After this lesson you can spot a made-up answer, and explain why chatbots make them.
About 9 minutes, or 15 with the Deep layers
Reading depth for every section
In short
A chatbot writes likely text, one piece at a time. Nothing in those steps checks the facts.
So an answer can hold a made-up detail that looks just like a real one. Sounding sure is not a sign of being right.
Check any detail that matters against a source you can open. Asking the same chatbot is not a check.
Reading depth forListen to
Three studies, with authors
A chatbot names three studies, with authors, years and journals. Would you put them in an essay?
A journal is where researchers publish their studies. The answer looks ready to use. But a detail in it could be made up. This lesson shows why, and what to do.
This lesson looks at these beliefs: “Confident and detailed means correct”, “Hallucination is basically solved now” and “Its citations are real evidence”. Each is set against dated evidence below. Every study, test and case carries its date.
Why it happens
A chatbot writes its answer one likely piece at a time (lesson 1.4). Each piece comes from chances its model learned in training.
Nothing in those steps checks the facts. So likely is not the same as true.
In September 2025, four researchers (three at OpenAI, one at Georgia Tech) argued that training itself leads to some errors. Their paper says: “we show that even if the training data were error-free, the objectives optimized during language model training would lead to errors being generated.” It is their argument, with a mathematical analysis. It is not a measured rate for any chatbot.
One part of their first reason is about facts seen only once in training. A base model is the model after its first training on text, before it is tuned into a chatbot. They argue that facts that appear only once in its training text are the kind it is expected to get wrong. A famous birthday, like Einstein’s, appears more than once. Others may appear only once, for example in an obituary. If a fifth (20%) of birthday facts appear exactly once, they expect a base model to get at least a fifth of birthday facts wrong. That is a minimum error rate for such facts, not a share of all errors. They also note that language models seldom get wrong the facts that are mentioned often, such as Einstein’s birthday.
Their second reason is about tests. They argue that most tests of language models give a point for a right answer and nothing for “I don’t know”. That is like an exam with no penalty for guessing: a model that guesses scores higher than one that says it is unsure. Of ten widely used tests they reviewed, only one gave any credit for showing uncertainty. They argue this is why made-up answers survive the later training that tunes a model into a chatbot. In their words, “the training and evaluation procedures reward guessing over acknowledging uncertainty”.
A dated example from the same paper: on 9 May 2025, the authors asked three popular language models for the title of one author’s PhD thesis. They were GPT-4o in ChatGPT, DeepSeek R1 in the DeepSeek app, and Llama 4 Scout on Hugging Face. None gave the right title or year, and none of them searched the web.
A search tool does not remove the problem. The same paper argues that “the binary grading system itself still rewards guessing whenever search fails to yield a confident answer.” Lesson 1.5 covers search, and how a searched answer can still get its source wrong. Lesson 1.7 covers training in full.
A made-up detail
So sometimes the text is false. An answer can hold a detail that sounds real and is not. It could be a study that does not exist, a wrong year, or a name that does not belong.
People call this a hallucination. The word is borrowed from people. Researchers note it is very different from what a person goes through. Google’s glossary says “confabulation” is probably a more accurate word.
A made-up detail can look just like a real one. So the look of an answer does not show which parts are true.
Anthropic makes Claude. It says even the most advanced language models, like Claude, can sometimes write text that is factually wrong. Its web page was checked in September 2026.
The International AI Safety Report 2026 came out in February 2026. It says even systems that do complex tasks well may give citations, biographies or facts that do not exist. A citation names a source, such as a study, so a reader can find it.
Google’s glossary defines a hallucination as “The production of plausible-seeming but factually incorrect output by a generative AI model that purports to be making an assertion about the real world.” Its entry for confabulation says: “Confabulation is probably a more technically accurate term than hallucination. However, hallucination became popular first.” The paper above notes that the error “differs fundamentally from the human perceptual experience”.
The OECD and European Union’s framework for AI literacy, the skills and knowledge to use AI in an informed way, says generative AI “can make fact and fabrication hard to distinguish”. The AI literacy training of Ireland’s Higher Education Authority (HEA) speaks of outputs that are “sometimes fluent yet wrong”.
Anthropic’s page says: “Even the most advanced language models, like Claude, can sometimes generate text that is factually incorrect or inconsistent with the given context.” That is Anthropic’s statement, checked in September 2026. The International AI Safety Report 2026 says: “Even systems that excel at complex tasks may generate non-existent citations, biographies, or facts”.
Sounding sure is not being right
How sure an answer sounds is not a sign that it is right.
In a study published in January 2025, people in the US read answers from three language models. They were GPT-3.5, PaLM2 and GPT-4o. The people tried to tell right answers from wrong ones. With the models’ usual explanations, they did only slightly better than guessing.
In that study, 301 people living in the US, all native English speakers recruited through Prolific, read answers with explanations written by GPT-3.5, PaLM2 and GPT-4o. They judged how likely each answer was to be right. With the models’ usual explanations, the authors write, they “were only slightly better than random guessing”. Longer explanations made people more confident without making them any better at it: “longer explanations increased user confidence, even when the extra length did not improve answer accuracy.”
The words mattered. Explanations that began “I am not sure” made people less confident than ones that said “I am somewhat sure”, whichever model wrote them. The authors then ran a simulation, not a new experiment. From experiment 2’s answers, they kept only the estimates people made on explanations whose wording fitted how likely the model’s own numbers said the answer was to be right. In that simulation, people’s estimates told right answers from wrong ones better than with the usual explanations of experiment 1.
In the multiple-choice experiments (answers from GPT-3.5 and PaLM2), people chose the answer the model’s explanation backed in most responses (82%). The models’ own internal chances told right answers from wrong ones better than the people did.
The paper does not say when its experiments ran, so it is dated here by when it was published. Its first posting, in January 2024, tested only GPT-3.5 and PaLM2.
A number a chatbot gives for its own confidence is not a check either. In a test posted by OpenAI researchers in November 2024, four OpenAI models (GPT-4o, GPT-4o-mini, o1-preview and o1-mini) were asked for an answer and how confident they were, as a percentage. They were right less often than they said. The authors write that the models “consistently overstate their confidence”.
Try it: Spot the made-up answer
The same question went to two AI models, ten times each. They were Claude Sonnet 5 by Anthropic, and Ministral 3 14B by Mistral AI. Neither had a search tool. Recorded on 28 September 2026.
Name three studies published since 2020 on whether taking notes by hand or on a laptop affects how much students remember from a lecture. For each study, give the title, the authors, the year, the journal and the main finding. Use plain text, with each study on its own line.
Here is one answer. It is the first answer in recording order with one-line studies and a made-up detail. Every answer is below.
The answer, from Ministral 3 14B
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:43 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
"The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Learning in Higher Education" – Mueller, P. A., & Oppenheimer, D. M. (2014) – Psychological Science (2014, but widely cited; a 2020 follow-up study by the same authors, "Revisiting the Pen Is Mightier Than the Keyboard: A Critical Review and Meta-Analysis of Note-Taking Research" – Mueller, P. A., & Oppenheimer, D. M. (2020) – Educational Psychology Review – Main finding: Meta-analysis confirmed that handwritten notes lead to better conceptual understanding and retention compared to laptop notes, though laptop notes may be superior for verbatim recall.
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a separate 2020 follow-up by the same authors. We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
"Laptop Multitasking Hinders Classroom Learning: An Experimental Study" – Feng, Y., Spilman, L., & Woger, B. C. (2021) – Computers & Education – Main finding: Students who took notes on laptops while multitasking (e.g., browsing the web) performed worse on conceptual questions than those who took handwritten notes or used laptops without distractions.
Checked: a study with a close title exists, but its title, its authors and its year do not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Its authors are Faria Sana, Tina Weston and Nicholas J. Cepeda. Its year is 2013. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Study 3:
"The Impact of Note-Taking Mode on Learning: A Systematic Review and Meta-Analysis" – Karpicke, J. D., & Blunt, J. R. (2021) – Educational Psychology Review – Main finding: While handwritten notes generally improved long-term retention, the advantage varied by task type—laptop notes were better for transcription but worse for deeper learning unless students engaged in active processing (e.g., summarizing or self-testing).
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
The checks are below, so decide first if you want to guess.
What all the runs showed
Ten answers were recorded from each AI, twenty in all.
From Claude Sonnet 5: no answer had at least one made-up detail.
From Ministral 3 14B: every answer had at least one made-up detail.
Some answers named a study we could not find.
Show all runs
Claude Sonnet 5, Run 1
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:40 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I want to flag a limitation before answering: I don't have reliable, verified knowledge of three specific post-2020 studies matching all those details (exact title, full author list, journal, year, and finding). If I generated specific-sounding citations from memory, there's a real risk I'd be fabricating or misattributing details—mixing up authors, inventing journal names, or misremembering findings. That's not useful to you and could actively mislead you if you cite it.
This line names no study.
What I can tell you with more confidence:
This line names no study.
Study 1:
The most influential paper in this space is Mueller, P. A., & Oppenheimer, D. M. (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," published in Psychological Science. It found that longhand note-takers performed better on conceptual questions, likely because handwriting forces more selective, generative processing rather than verbatim transcription. This predates 2020 but is the study most later work responds to or tries to replicate.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Since 2020, there has been continued research and some failed replications/extensions of the original Mueller & Oppenheimer effect, along with neuroscience-oriented work (e.g., using EEG) comparing handwriting and typing, but I can't vouch for specific titles/journals/years without a real risk of getting it wrong.
If this is for something like a paper, lit review, or presentation, I'd suggest:
This line names no study.
Searching Google Scholar with terms like "handwriting laptop note-taking memory 2020..2024"
This line names no study.
Checking who has cited Mueller & Oppenheimer (2014) since 2020 via Google Scholar's "Cited by" feature
This line names no study.
Looking at journals like Psychological Science, Educational Psychology Review, Computers & Education, or Memory & Cognition
This line names no study.
Would you like me to help you construct a search strategy, or would a summary of what I do know about the research area (without fabricated citations) be useful?
This line names no study.
Claude Sonnet 5, Run 2
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:40 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I don't have reliable, verified citation details (exact authors, journal, and year) for three specific studies published since 2020 on this topic that I can confidently vouch for. Rather than risk giving you fabricated or inaccurate citations, I want to flag this clearly.
This line names no study.
What I can tell you with more confidence:
This line names no study.
Study 1:
The foundational study everyone builds on is Mueller & Oppenheimer (2014), "The Pen Is Mightier Than the Keyboard," Psychological Science — finding that longhand note-takers performed better on conceptual questions than laptop note-takers, likely due to shallower "verbatim" transcription when typing.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Study 2:
A notable replication/extension is Morehead, Dunlosky & Rawson (2019), "How Much Mightier Is the Pen Than the Keyboard for Note-Taking?," Educational Psychology Review — which found much weaker and less consistent effects than the original study, suggesting the "laptops are worse" conclusion may not generalize as strongly as commonly believed. This is 2019, just before your cutoff, so it doesn't meet your "since 2020" requirement.
Search for and verify actual post-2020 papers on this topic right now if you have search/browsing enabled for me, or
This line names no study.
Give you well-known researchers/labs (e.g., those studying note-taking, encoding, and generative vs. transcriptive processing) whose recent work you could look up directly, or
This line names no study.
Summarize the general state of research since 2020 without pretending to cite specific papers I can't verify.
This line names no study.
Would you like me to proceed with any of these instead?
This line names no study.
Claude Sonnet 5, Run 3
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:40 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I don't have reliable, verified information about three specific studies published since 2020 on this exact topic that I can cite with confidence. I want to avoid fabricating titles, author names, journals, or findings, since that would risk giving you false citations.
This line names no study.
What I can tell you with more confidence:
This line names no study.
Study 1:
The most well-known and frequently cited study on this topic is Mueller & Oppenheimer (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," published in Psychological Science. It found that students who took notes by hand performed better on conceptual questions than those who typed notes, likely because handwriting encourages more selective, generative processing rather than verbatim transcription. This predates 2020, but it's the foundational study that most later research responds to or tries to replicate.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Since 2020, there has been a body of replication and follow-up work (including some studies that failed to replicate the original effect, or found it was more nuanced/context-dependent), but I don't have confident, specific recall of exact titles, author lists, years, and journals for three such papers.
Describe the general trends and debates in this research area since 2020 without specific citations, or
This line names no study.
Suggest search strategies/databases (like Google Scholar, PsycINFO, or ERIC) where you could find peer-reviewed studies matching your criteria, or
This line names no study.
You could share a study you've come across, and I can help you understand or summarize it.
This line names no study.
Would any of these be helpful?
This line names no study.
Claude Sonnet 5, Run 4
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:40 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I can't confidently provide specific citations for three post-2020 studies on this topic — I don't have verified details (exact titles, author lists, journals) for papers meeting that description, and I don't want to risk giving you fabricated or inaccurate references.
This line names no study.
What I can tell you honestly:
This line names no study.
Study 1:
The most famous and widely-cited study on this topic is Mueller, P. A., & Oppenheimer, D. M. (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," Psychological Science. It found that longhand note-takers performed better on conceptual questions, likely because handwriting encourages more selective, synthesized processing rather than verbatim transcription. This predates 2020 but is the foundation almost all later work responds to.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Since 2020, there has been continued academic discussion and some attempts to replicate or challenge these findings (including debates about effect size and reproducibility), but I don't have reliable, verified bibliographic details for specific 2020+ papers to share with you as fact.
Describe the general trends and debates in this research area since 2020 without citing specific papers as fact, or
This line names no study.
Suggest search strategies (e.g., Google Scholar queries, specific journals like Psychological Science, Journal of Educational Psychology, or Computers & Education) so you can find and verify the actual studies yourself, or
This line names no study.
You could share any studies you've already come across, and I can help you summarize or analyze them accurately.
This line names no study.
Would any of these be helpful?
This line names no study.
Claude Sonnet 5, Run 5
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:41 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I want to flag a limitation before answering: I don't have reliable, verified knowledge of specific post-2020 studies on this topic with accurate titles, author lists, and journal names. If I generated three specific citations right now, there's a meaningful risk I'd be fabricating details—mixing up real researchers with invented paper titles, or misattributing findings—since this is exactly the kind of specific bibliographic request where language models tend to produce plausible-sounding but incorrect citations.
This line names no study.
What I can tell you reliably:
This line names no study.
Study 1:
The foundational study most people reference is Mueller & Oppenheimer (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," published in Psychological Science. It found that longhand notetakers performed better on conceptual questions, likely because handwriting forces more selective, generative processing rather than verbatim transcription. This predates your 2020 cutoff, though.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
I'm aware there has been follow-up and partially contradicting research since then (some studies have failed to replicate the original effect, or found it depends on factors like note-taking strategy, test type, or laptop use restrictions), but I can't confidently give you exact titles/authors/journals/years for post-2020 papers without risking invented details.
Search Google Scholar for "handwriting vs laptop note-taking" with a 2020+ date filter—this will surface actual recent papers.
This line names no study.
Check who has cited Mueller & Oppenheimer (2014) in recent years, since that will surface direct follow-ups and replications.
This line names no study.
If you want, I can help you evaluate studies once you've found them, or discuss the general debates/mechanisms in this research area without pretending to cite specifics I can't verify.
This line names no study.
Would any of those alternatives be helpful?
This line names no study.
Claude Sonnet 5, Run 6
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:41 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I don't have reliable, verified citation details for three specific studies published since 2020 on this exact topic, and I don't want to give you fabricated authors, journal names, or dates that sound plausible but may be wrong.
This line names no study.
Study 1:
What I can tell you with more confidence is background: the most famous and heavily cited study here is Mueller & Oppenheimer (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," Psychological Science — which found that laptop note-takers, who tend to transcribe verbatim, performed worse on conceptual questions than longhand note-takers. Since then there has been a body of follow-up and replication work (some supporting the original effect, some finding much weaker or no effect when factors like note-taking instructions, verbatim overlap, or test type are controlled for), but I can't vouch for exact 2020+ titles, author lists, or journal names from memory without risking inaccuracy.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
If this is for something like a paper, lit review, or presentation, I'd suggest one of these approaches instead:
This line names no study.
I can help you search strategies/keywords (e.g., "longhand vs laptop note-taking," "transcription vs generative note-taking," "verbatim note-taking laptop") to find recent studies via Google Scholar, PsycINFO, or ERIC.
This line names no study.
If you have access to a database, I can help you evaluate or summarize specific papers you find and paste in.
This line names no study.
I can discuss the general state of the research debate (including critiques of the original 2014 study) without pinning it to specific unverified 2020+ citations.
This line names no study.
Would you like me to go one of those routes?
This line names no study.
Claude Sonnet 5, Run 7
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:41 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I want to flag a limitation here: I can't reliably provide three specific studies with accurate titles, author names, journals, and years for this topic. Academic citations are exactly the kind of detail I can get wrong in a way that sounds convincing but isn't accurate—I might blend real researchers' names with incorrect paper titles, or cite a journal that didn't actually publish that work. Given that, I'd rather not guess and risk giving you fabricated references.
This line names no study.
Study 1:
What I can tell you with more confidence is the general landscape: this research area builds heavily on the influential 2014 study by Pam Mueller and Daniel Oppenheimer ("The Pen Is Mightier Than the Keyboard"), which found longhand note-takers often outperformed laptop note-takers on conceptual questions. Since then, there has been a series of follow-up and replication studies, some of which failed to replicate the original effect or found it was more nuanced (e.g., depending on note-taking strategy, verbatim transcription vs. summarizing, or the specific memory measures used).
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
If you need actual citations from 2020 onward, I'd suggest:
This line names no study.
Searching Google Scholar with terms like "handwriting laptop note-taking memory" and filtering by year (2020-present).
This line names no study.
Checking journals like "Journal of Educational Psychology," "Memory & Cognition," "Educational Psychology Review," or "Computers & Education," which frequently publish in this area.
This line names no study.
Looking at citation trackers for the original Mueller & Oppenheimer paper to find studies that cite and build on it post-2020.
This line names no study.
Would you like me to help you construct effective search queries or search strategies instead?
This line names no study.
Claude Sonnet 5, Run 8
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:41 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I want to flag a limitation before answering: I don't have reliable, verified recall of three specific post-2020 studies with accurate titles, author lists, journal names, and findings on this topic. If I listed them from memory, there's a real risk I'd fabricate or misattribute details—getting a journal name wrong, inventing an author, or misremembering a finding—which could mislead you if you cited them.
This line names no study.
What I can tell you with more confidence:
This line names no study.
Study 1:
The foundational study most people reference is Mueller, P. A., & Oppenheimer, D. M. (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," published in Psychological Science. It found that longhand note-takers performed better on conceptual questions, likely because handwriting encourages summarizing rather than verbatim transcription. This is pre-2020, but it's the study most later work responds to or tries to replicate.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Since 2014, there's been debate and some failed replications questioning how robust this effect is, and I'm aware there's ongoing research into note-taking modality, but I can't confidently reconstruct three specific 2020+ citations without a strong chance of error.
Search for and verify current studies right now if I have that capability in this context, or
This line names no study.
Suggest search terms (e.g., "laptop vs longhand note-taking recall 2021 2022 2023" on Google Scholar) so you can pull accurate, citable sources yourself, or
This line names no study.
Discuss the general findings and debates in this research area without pretending to cite specific papers I can't verify.
This line names no study.
Would any of these help?
This line names no study.
Claude Sonnet 5, Run 9
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:41 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I want to flag a limitation here rather than risk giving you inaccurate information: I can't reliably recall three specific post-2020 studies on handwriting vs. laptop note-taking with confidently accurate titles, full author lists, exact journals, and findings. Given how specific this request is (exact citations), guessing could mean giving you fabricated or misattributed sources, which wouldn't be useful for actual research or citation purposes.
This line names no study.
What I can tell you with more confidence:
This line names no study.
Study 1:
The foundational study most people reference is Mueller & Oppenheimer (2014), "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking," published in Psychological Science — but this predates 2020, so it doesn't fit your request.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
I'm aware that this study's findings have been contested and that replication attempts and follow-up research have produced mixed results since then, but I don't trust my memory to give you precise, accurate citations for that follow-up literature.
To get accurate, verifiable studies from 2020 onward, I'd suggest:
This line names no study.
Searching Google Scholar with terms like "laptop note-taking longhand memory 2020..2024"
This line names no study.
Checking PubMed or PsycINFO if you have institutional access
This line names no study.
Looking at papers that cite Mueller & Oppenheimer (2014) — Google Scholar's "Cited by" feature will show you newer studies that engaged with or tested that original finding
This line names no study.
Would you like me to help you construct a more targeted search strategy, or help you evaluate studies once you've pulled some candidates?
This line names no study.
Claude Sonnet 5, Run 10
claude-sonnet-5 on Anthropic Messages API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Effort: high. Thinking: adaptive. Up to 16,000 tokens. Default sampling settings. No instructions from us besides the question. Anthropic says it adds its own instruction when thinking is on.
I want to flag a limitation here: I don't have verified, reliable access to specific citation details (exact titles, author lists, journal names) for studies published since 2020 on this topic. I could try to recall specifics, but there's a real risk of misremembering or blending details from different papers, which would give you inaccurate citations—not helpful for something you might want to cite or rely on.
This line names no study.
What I can tell you with more confidence:
This line names no study.
Study 1:
The foundational study most people reference is Mueller & Oppenheimer (2014), "The Pen Is Mightier Than the Keyboard," published in Psychological Science, which found longhand note-takers performed better on conceptual questions than laptop note-takers. This predates your 2020 cutoff but is the study most later work responds to or tries to replicate.
Checked: this study exists, but it came out before 2020, which the question did not ask for. This paper is from 2014. Its influence ranking is not checked. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Since 2014, there have been replication attempts and critiques, and I recall the literature has become more mixed—some studies failing to replicate the original effect, others pointing to factors like verbatim transcription versus generative note-taking as the real driver rather than the device itself.
Rather than risk giving you fabricated citation details, I'd suggest:
This line names no study.
Searching Google Scholar with terms like "laptop notes longhand memory lecture 2020 2021 2022 2023" and filtering by date.
This line names no study.
Checking who has cited Mueller & Oppenheimer (2014) since 2020—Google Scholar's "Cited by" feature is useful for this.
This line names no study.
Looking at journals likely to publish this kind of work: Psychological Science, Journal of Educational Psychology, Computers & Education, or Memory.
This line names no study.
If you'd like, I can help you think through what a good search strategy would look like, or discuss the general debate/mechanisms in this research area without pretending to cite specific papers I'm not certain about.
This line names no study.
Ministral 3 14B, Run 1
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
Year: 2014 (but widely cited; a 2021 follow-up study by the same authors, "The pen is mightier than the keyboard: Effects of note-taking mode on learning" in Psychonomic Bulletin & Review, reinforced these findings with updated meta-analytic support)
Checked: this study exists, but it came out before 2020, which the question did not ask for. This year matches a study from 2014. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a separate 2021 follow-up by the same authors. We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Main Finding: Handwritten notes lead to better conceptual understanding and long-term retention compared to laptop notes, likely because handwriting encourages deeper processing and reduces verbatim transcription.
Title: Laptop multitasking hinders classroom learning: An experimental study
Checked: this study exists, but its title does not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Main Finding: Students who took notes on laptops were more likely to multitask (e.g., browsing the web), which impaired comprehension and memory compared to those who wrote by hand or typed without distractions.
Title: The impact of note-taking mode on learning from lectures: A meta-analysis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Authors: Korbach, J., & Papassotiriou, A.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Year: 2020
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Journal: Educational Psychology Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Main Finding: Meta-analysis confirmed that handwritten notes generally enhance retention and conceptual learning, while laptop notes (especially when used for verbatim transcription) reduce deeper processing and memory performance. The effect was stronger for complex material.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Ministral 3 14B, Run 2
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its impact on memory:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Learning
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Main Finding: While the original 2014 study found handwritten notes led to better conceptual understanding, later research (e.g., meta-analyses) suggests the effect depends on note-taking quality and lecture structure—laptops may help with verbatim recall, but handwriting can improve deeper learning when students engage actively.
Title: The Effect of Note-Taking Mode on Learning from Lectures: A Meta-Analysis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Authors: Daniel M. Oppenheimer, Daniel J. Levinson, and Pamela A. Mueller
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Year: 2021
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Journal: Educational Psychology Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Main Finding: The meta-analysis confirmed that handwritten notes generally enhance conceptual learning and retention compared to laptop notes, but only when students take meaningful notes (not verbatim transcription). Laptop users often engage in passive transcription, reducing comprehension.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Title: Note-Taking Technology and Student Learning: A Critical Review and Synthesis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Authors: Sarah E. McDaniel, Jeffrey D. Halpern, and Katherine A. Graesser
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Year: 2022
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Journal: Educational Psychology Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Main Finding: The review highlighted that handwritten notes improve memory and understanding, especially for complex material, while laptop notes can hinder learning if students rely on autocompletion or multitasking. However, the advantage of handwriting diminishes if students use laptops strategically (e.g., organizing notes later).
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Note: The 2014 Mueller & Oppenheimer study is foundational, but the 2021 and 2022 papers provide updated syntheses. For primary sources, the meta-analyses are key.
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Learning
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Main Finding (Updated Context): While this 2014 study found handwritten notes improved conceptual understanding, a 2021 meta-analysis (Educational Psychology Review) by Karpicke & Roediger (2021) reinforced that handwritten notes often lead to better retention, especially for complex material, due to deeper processing. Laptop notes may encourage verbatim transcription, reducing encoding.
This line makes a claim that the linked source does not settle. The linked record does not settle this finding. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a 2021 meta-analysis by Karpicke and Roediger, with no title. This citation gave no title. We could not find a study with its authors, year and journal. On 28 September 2026, we searched Crossref's top 20 results and the web for its details. The same searches found a real study. Search: Crossref (opens in a new tab)
Main Finding: While not exclusively about note-taking, this study found that laptop use (even for non-note-taking tasks) impaired learning for both the user and nearby students. Handwritten note-takers showed better comprehension, suggesting laptops may distract from deep processing.
Title: The Effects of Note-Taking Mode (Longhand vs. Laptop) on the Learning and Retention of Lecture Material
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Authors: Hennessy, S. A., & Korb, A. E.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Year: 2022
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Journal: Educational Psychology
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Main Finding: This study found that students who took notes by hand recalled more factual details and demonstrated better conceptual understanding than laptop note-takers, particularly for lectures with complex ideas. Laptop users tended to take more verbatim notes, which correlated with poorer retention.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
(For a fourth, highly relevant 2023 study:)
This line names no study.
Study 4:
Title: Handwritten Notes Enhance Learning and Memory Compared to Laptop Notes
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Authors: Boon, C. J., et al.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Year: 2023
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Journal: Journal of Educational Psychology
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Main Finding: Meta-analysis showing handwritten notes consistently improved long-term retention and conceptual learning, while laptop notes led to shallower processing. The effect was strongest for students who transcribed verbatim rather than paraphrasing.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Ministral 3 14B, Run 4
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Conceptual Learning
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Main Finding: While the original 2014 study found handwritten notes led to better conceptual understanding, later work (e.g., 2021 critiques) notes that the effect may depend on note-taking quality (e.g., verbatim vs. paraphrasing) rather than the tool itself.
Title: Laptop Multitasking Hinders Classroom Learning: Effects on Performance and Perceptions
Checked: this study exists, but its title does not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Study 2:
Authors: Fiddick, M. A., Kay, A. C., & Christ, J. A.
Main Finding: Laptop use (even for note-taking) was linked to lower performance and more distractions, while handwritten notes correlated with better focus and retention—though the study emphasizes multitasking (e.g., web browsing) as a key factor.
Title: The Effects of Note-Taking Mode (Longhand vs. Laptop) on Learning and Retention
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Authors: Boon, A. J., et al.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Year: 2022
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Journal: Educational Psychology
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Main Finding: Handwritten notes improved retention of lecture content compared to laptop notes, but only when students engaged in elaborative (summary-based) note-taking. Verbatim laptop notes led to poorer recall unless students actively processed the material.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Title: Digital Note-Taking and Learning: A Meta-Analysis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Authors: Wang, Q., et al.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Year: 2023
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Journal: Educational Research Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Main Finding: Meta-analysis of 20+ studies found that while handwritten notes generally enhanced memory for factual details, laptop notes were superior for speed and volume of notes. The advantage of handwriting diminished when students used laptops for active processing (e.g., summarizing) rather than passive transcription.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Ministral 3 14B, Run 5
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its impact on memory:
"The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Learning" – Mueller, P. A., & Oppenheimer, D. M. (2014) – Psychological Science (2014, but widely cited; a 2021 follow-up study by the same authors: "Revisiting the Pen Is Mightier Than the Keyboard Debate: A Meta-Analysis" – Mueller, P. A., & Oppenheimer, D. M. (2021) – Educational Psychology Review)
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a separate 2021 follow-up by the same authors. We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 1:
Main finding (2021): A meta-analysis confirmed that handwritten notes lead to better conceptual understanding and retention compared to laptop notes, though the effect varies by task type (e.g., note-taking vs. transcription).
"Laptop Multitasking Hinders Classroom Learning for Both Users and Nearby Peers" – Feng, Y., Spilka, A., & Ward, A. F. (2021) – Computers & Education
Checked: a study with this title exists, but its authors and its year do not match the record. The record's authors are Faria Sana, Tina Weston and Nicholas J. Cepeda. Its year is 2013. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Study 2:
Main finding: While not exclusively about note-taking, this study found that laptop use (including note-taking) in lectures—especially when multitasking—reduced comprehension for both the user and nearby students, whereas handwritten note-takers performed better.
"The Effects of Note-Taking Mode (Laptop vs. Longhand) on Learning from Lectures" – Hunt, E., & Einstein, G. O. (2012) – Psychonomic Bulletin & Review (2012, but a 2023 replication/extension: "Re-examining the Pen vs. Keyboard Debate: A Preregistered Study on Note-Taking and Learning" – Hunt, E., & Einstein, G. O. (2023) – Journal of Experimental Psychology: Applied)
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab) Related study: The line names a separate 2023 replication. We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Main finding (2023): Replicated earlier findings that handwritten notes improved retention, particularly for conceptual questions, while laptop notes were better for verbatim recall—though the advantage of handwriting was stronger in conditions requiring deeper processing.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
(Note: Many foundational studies predate 2020, but these are key post-2020 extensions or meta-analyses. If you need strictly post-2020 primary research, let me know—I can refine the search!)
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Learning
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Year: 2014 (updated meta-analysis referenced in later work, but key findings were widely cited in 2020+ studies; see below for a more recent replication)
Main Finding: While the 2014 study found handwritten notes improved conceptual understanding, a 2021 follow-up (Journal of Experimental Psychology: General) by Mueller et al. confirmed that handwriting still enhanced retention for complex material, but laptop notes were better for verbatim recall when students used effective strategies (e.g., summarizing).
This line makes a claim that the linked source does not settle. The linked record does not settle this finding. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a 2021 follow-up by Mueller and others, with no title. This citation gave no title. We could not find a study with its authors, year and journal. On 28 September 2026, we searched Crossref's top 20 results and the web for its details. The same searches found a real study. Search: Crossref (opens in a new tab)
(Note: For a more recent direct study, see below.)
This line names no study.
Study 2:
Title: The Effect of Note-Taking Mode on Learning from Lectures: A Meta-Analysis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Authors: Sarah E. McGill, Sarah E. Butler, and Daniel M. Oppenheimer
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Year: 2022
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Journal: Educational Psychology Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Main Finding: Handwritten notes led to better conceptual learning (e.g., understanding ideas) compared to laptop notes, but laptop notes were superior for transcribing information verbatim. The effect was stronger for low-stakes learning (e.g., lectures) than high-stakes tests.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Title: Laptop Multitasking Hinders Classroom Learning: Effects on Performance and Study Strategies
Checked: this study exists, but its title does not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Main Finding: While not exclusively about note-taking, this study showed that laptop use (even for note-taking) reduced comprehension when students multitasked (e.g., checking email). A 2021 study (Journal of Educational Psychology) by Karpicke et al. found that handwritten notes improved long-term retention even when students later typed them up, suggesting the process of writing by hand (not just the final product) boosts memory.
This line makes a claim that the linked source does not settle. The linked record does not settle this finding. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab). Related study: The line names a 2021 study by Karpicke and others, with no title. This citation gave no title. We could not find a study with its authors, year and journal. On 28 September 2026, we searched Crossref's top 20 results and the web for its details. The same searches found a real study. Search: Crossref (opens in a new tab)
This line names no study.
More recent (2020–2023) study:
This line names no study.
Study 4:
Title: The Pen Is Mightier Than the Keyboard: Differences in Learning and Retention Between Handwritten and Typed Notes
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Authors: Sarah E. Butler, Sarah E. McGill, and Daniel M. Oppenheimer
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Year: 2021
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Journal: Journal of Experimental Psychology: Applied
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Main Finding: Handwritten notes led to better performance on conceptual questions (e.g., applying ideas) compared to laptop notes, even when students later reviewed their notes. The advantage was most pronounced for low-stakes learning (e.g., lectures) and when students didn’t engage in deep processing (e.g., re-studying notes).
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
This line names no study.
Key Takeaway: Most recent research (2020–2023) supports that handwritten notes improve conceptual understanding, while laptops may be better for verbatim recall—but only if students avoid distractions (e.g., multitasking). The act of writing by hand (not just the medium) seems to enhance memory.
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:42 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Study 1:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking
Main Finding: While the original 2014 study found handwritten notes improved conceptual understanding, a 2021 meta-analysis (Mangen et al.) confirmed that handwriting still enhances retention, especially for complex material, though laptop note-taking can be effective if students engage in active processing (e.g., summarizing, self-testing).
(Note: Mueller & Oppenheimer’s work remains foundational, but here’s a more recent study:)
This line names no study.
Study 2:
Title: The Effect of Note-Taking Mode on Learning from Lectures: A Meta-Analysis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Authors: Audun Mangen, John-Peter Sund, and Torbjørn M. Brønnick
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Year: 2021
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Journal: Educational Psychology Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Main Finding: Handwritten notes led to better retention of conceptual information (effect size d = 0.27) and higher performance on application-based tests, while laptop notes were superior for verbatim recall. The advantage of handwriting was stronger for lower-ability students and when lectures were complex.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Title: Laptop Multitasking Hinders Classroom Learning: Effects on Performance and Study Strategies
Checked: this study exists, but its title does not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Study 3:
Authors: Daniel M. Oppenheimer, Adam J. Albarracín, and Yael M. Mark
Main Finding: Students using laptops for note-taking remembered less than those writing by hand, but the gap widened when students multitasked (e.g., browsing the web). Handwritten notes improved deep processing (e.g., elaboration, organization), while laptop users relied more on transcription, reducing long-term retention.
(Bonus: For a 2023 study on note-taking and digital distractions, see Hwang et al. (2023) in Computers & Education, which found that even offline laptop use—withoutmultitasking—reduced comprehension compared to pen-and-paper.)
This line makes a wider claim that the linked source does not settle. The linked source does not settle this wider claim. Record: doi.org/10.1007/s10648-019-09468-2 (opens in a new tab). Related study: The line names a 2023 study by Hwang and others, with no title. This citation gave no title. We could not find a study with its authors, year and journal. On 28 September 2026, we searched Crossref's top 20 results and the web for its details. The same searches found a real study. Search: Crossref (opens in a new tab)
Ministral 3 14B, Run 8
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:43 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Conceptual Learning, but Not for Transcription
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab).
Year: 2014 (but widely cited; a 2021 follow-up study: Mueller, H.A., & Pesta, J. – "The Pen Is Mightier Than the Keyboard: Advantages of Longhand Note Taking for Learning" – Journal of Applied Research in Memory and Cognition, 2021)
Checked: this study exists, but it came out before 2020, which the question did not ask for. This year matches a study from 2014. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a separate 2021 follow-up. We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 1:
Journal: Psychological Science (2014) / Journal of Applied Research in Memory and Cognition (2021)
Main Finding: Handwritten notes lead to better conceptual understanding and retention compared to laptop notes, likely because typing encourages verbatim transcription rather than processing information.
Title: Laptop Multitasking Hinders Classroom Learning: An Experimental Study
Checked: this study exists, but its title does not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Study 2:
Authors: Lewandowsky, S., Oberauer, K., & Brown, A.
Main Finding: Students who took notes on laptops (especially while multitasking) performed worse on conceptual questions compared to those who wrote by hand, reinforcing that digital note-taking can reduce deeper learning.
Title: The Effects of Note-Taking Mode (Longhand vs. Laptop) on Learning from Lectures
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Authors: Pesta, J., & Mueller, H.A.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Year: 2022
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Journal: Journal of Educational Psychology
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 3:
Main Finding: Handwritten notes improved long-term retention and comprehension, while laptop notes (even when discouraged from typing verbatim) still led to lower performance, suggesting that the medium itself influences cognitive processing.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Ministral 3 14B, Run 9
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:43 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Here are three studies published since 2020 on handwritten vs. laptop note-taking and its effects on memory:
"The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking for Learning in Higher Education" – Mueller, P. A., & Oppenheimer, D. M. (2014) – Psychological Science (2014, but widely cited; a 2020 follow-up study by the same authors, "Revisiting the Pen Is Mightier Than the Keyboard: A Critical Review and Meta-Analysis of Note-Taking Research" – Mueller, P. A., & Oppenheimer, D. M. (2020) – Educational Psychology Review – Main finding: Meta-analysis confirmed that handwritten notes lead to better conceptual understanding and retention compared to laptop notes, though laptop notes may be superior for verbatim recall.
Checked: this study exists, but its title does not match the record. The record's title is “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking”. Record: doi.org/10.1177/0956797614524581 (opens in a new tab). Related study: The line names a separate 2020 follow-up by the same authors. We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
"Laptop Multitasking Hinders Classroom Learning: An Experimental Study" – Feng, Y., Spilman, L., & Woger, B. C. (2021) – Computers & Education – Main finding: Students who took notes on laptops while multitasking (e.g., browsing the web) performed worse on conceptual questions than those who took handwritten notes or used laptops without distractions.
Checked: a study with a close title exists, but its title, its authors and its year do not match the record. The record's title is “Laptop multitasking hinders classroom learning for both users and nearby peers”. Its authors are Faria Sana, Tina Weston and Nicholas J. Cepeda. Its year is 2013. Record: doi.org/10.1016/j.compedu.2012.10.003 (opens in a new tab).
Study 3:
"The Impact of Note-Taking Mode on Learning: A Systematic Review and Meta-Analysis" – Karpicke, J. D., & Blunt, J. R. (2021) – Educational Psychology Review – Main finding: While handwritten notes generally improved long-term retention, the advantage varied by task type—laptop notes were better for transcription but worse for deeper learning unless students engaged in active processing (e.g., summarizing or self-testing).
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Ministral 3 14B, Run 10
ministral-14b-2512 on Mistral AI API. Recorded 28 September 2026 at 16:43 UTC.
No search tool. Temperature: the model's default (0.3, read from Mistral on 28 September 2026). Mistral's safety prompt: off. Up to 4,096 tokens. No instructions from us besides the question.
Study 1:
Title: The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking
Year: 2014 (Note: While this is a seminal study, a follow-up or related work was published in 2020—see below for a more recent study. Here’s a corrected entry for a 2020+ study instead.)
Title: The Effects of Note-Taking Mode (Laptop vs. Longhand) on Learning from Lectures
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Authors: Jeffrey D. Karpicke, Henry L. Roediger III, and Katherine A. McDermott
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Year: 2021 (published in Psychological Science)
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Journal: Psychological Science
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 2:
Main Finding: Handwritten notes led to better conceptual understanding and retention compared to laptop notes, particularly when students were later tested on their ability to apply or explain the material rather than recall verbatim details.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
This line names no study.
Study 3:
Title: Laptop Multitasking Hinders Classroom Learning for Both Users and Nearby Peers
Main Finding: While not exclusively about handwriting vs. laptops, this study reinforced that laptop use (especially for multitasking) impairs learning, and handwritten notes were implicitly shown to mitigate some of these distractions by encouraging deeper engagement.
Title: The Impact of Note-Taking Method on Learning and Retention: A Meta-Analysis
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Authors: Michael D. Scaife and Richard J. D. Astle
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Year: 2022
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Journal: Educational Psychology Review
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
Study 4:
Main Finding: The meta-analysis found that handwritten notes generally improved long-term retention and conceptual understanding compared to laptop notes, though the effect varied by context (e.g., lecture type, testing method). The advantage of handwriting was strongest for complex or abstract material.
We could not find this study. On 28 September 2026, we searched Crossref's top 20 results and the web for its exact title. The same searches found a real study. Search: Crossref (opens in a new tab)
This line names no study.
Note: Mueller and Oppenheimer’s 2014 study is the most cited in this field, but the above are more recent (2020–2022) works that build on or replicate their findings. If you’d like a strictly 2020+ study, the Karpicke et al. (2021) and Scaife & Astle (2022) papers are the most relevant.
A line can sound sure and still be made up, so check what matters against a source.
What the recorded answers showed
We put the same question to Claude, from Anthropic, and to Ministral, from Mistral. Each try is a run. A run counts if any line has a made-up detail, or names a study we could not find.
In this recording, Claude Sonnet 5’s answer had a made-up detail in 0 of its 10 runs. Ministral 3 14B’s answer had one in 10 of its 10 runs. So every Ministral run counted, and no Claude run did.
That is 20 answers to one question, not a test of either model.
A run counts here when at least one of its lines had a made-up detail, or named a study we could not find. Each check of a named study links its record or the search.
Made-up citations, and what they cost
Citations can be made up too.
Two researchers had ChatGPT write short reviews in the first week of April 2023. Then they checked every citation. More than half (55%) of GPT-3.5’s were to works that do not exist. So were about 1 in 5 (18%) of GPT-4’s.
Made-up citations can cost people. In June 2023, a US court fined two lawyers and their law firm $5,000.
They had filed papers citing court cases that ChatGPT made up. They kept standing by them after the court questioned them. The judge also found that the two lawyers acted in bad faith.
The citation study: in the first week of April 2023, two researchers had ChatGPT (GPT-3.5 and GPT-4) write 84 short literature reviews on 42 topics. They checked all 636 works cited. More than half (55%) of GPT-3.5’s citations and about 1 in 5 (18%) of GPT-4’s were to works that do not exist. Many citations to real works had errors too: about 4 in 10 (43%) for GPT-3.5 and about 1 in 4 (24%) for GPT-4. It was published on 7 September 2023. These are 2023 models, and newer ones were not tested.
The court case is Mata v. Avianca, in a New York federal court. The lawyers filed papers on 1 March 2023 citing court decisions that ChatGPT had made up, with fake quotes. They kept standing by them after the court questioned them. The judge found “bad faith on the part of the individual Respondents based upon acts of conscious avoidance and false and misleading statements to the Court.” On 22 June 2023 the judge fined the two lawyers and their law firm $5,000. The judge also ordered them to write to each judge falsely named as the author of a fake decision.
The judge also wrote: “there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.” The duty to check what is filed stays with the lawyers.
In Ireland, the High Court’s practice direction HC 142, on generative AI in court documents, is dated 29 July 2026 by the President of the High Court. It is in operation from 1 September 2026, for civil proceedings. Among the known risks of AI tools, it lists “fictitious cases” and “academic references”. It says people using AI tools to prepare submissions “must independently verify all legal authorities, citations, quotations and references relied upon in Court Documents using authoritative sources.” That is a rule for court documents in Ireland’s High Court.
The HEA’s AI literacy training, written for staff, asks for “A curated repository of real cases, including hallucinated facts, fabricated citations, and confidently incorrect analyses”. The cases in this lesson follow that idea.
Is it solved?
Do newer systems still make things up? An international report from February 2026 says many kinds of failures, such as hallucinations, have generally become less likely.
But it says systems still often make mistakes on more complex tasks. Less often is not never.
The International AI Safety Report 2026 says: “Since the publication of the previous Report (January 2025), AI systems have generally become more reliable”. It goes on: “Many kinds of failures, such as hallucinations, have generally become less likely, but systems still commonly make mistakes when performing more complex tasks.”
Newer is not always fewer. In its system card for o3 and o4-mini, dated 16 April 2025, OpenAI reported its own test, PersonQA: questions about public facts about people, built to draw out made-up answers. OpenAI’s hallucination rate on it was 0.33 (about 1 in 3) for o3 and 0.48 (almost half) for o4-mini, against 0.16 (about 1 in 6) for the older o1. On its other test, SimpleQA, the rates were 0.51 (about half) for o3, 0.79 (about 4 in 5) for o4-mini and 0.44 (more than 4 in 10) for o1.
OpenAI says o3 makes more claims overall, right and wrong. It calls this effect minor on SimpleQA and more pronounced on PersonQA, and says: “More research is needed to understand the cause of these results.” These are one company’s tests of its own models, on one date. They are not a rate for AI in general, and they do not show that newer models make more mistakes.
What students say. HEPI surveyed UK full-time undergraduates through Savanta: 1,041 in December 2024 (published February 2025) and 1,054 in December 2025 (published March 2026). It asked: “Which of the below, if any, are reasons which make you less likely to use AI tools for your studies?” About half (51%) picked “Getting false results / hallucinations” in the first survey, and about a third (35%) in the second. HEPI marks the drop as significant. The responses were weighted on demographics such as gender, institution type and year of study.
Over the same two surveys, students using AI in at least one way went from 92% to 95% (almost all). This counts a reason students give for using AI less. It does not measure how often AI is wrong, and it does not show that students think the problem is solved.
What helps
Check any detail that matters against a source you can open yourself, such as the study itself.
Asking the same chatbot is not a check. In a US court case in 2023, a lawyer asked ChatGPT whether its cases were real. It said they were.
Lesson 4.1 will show how to check an answer’s claims and citations against real sources.
In the Mata case, the lawyer asked ChatGPT “Is Varghese a real case” and “Are the other cases you provided fake”. ChatGPT replied that they were real and could be found through Westlaw, LexisNexis and the Federal Reporter.
Ireland’s High Court, for court documents, warns that AI outputs “may incorrectly assert, when asked, that information is accurate when it is not” (paragraph 10). Paragraph 13 says: “It is not sufficient to use the same or other GenAI Tools to confirm that information is accurate.” GenAI means AI that makes new text or pictures. Its next sentence allows one route: “However, persons responsible for a Court Document may verify GenAI content by referring to original source material housed within a GenAI Tool.” That is a rule for court documents in Ireland’s High Court.
The OECD and European Union framework’s basic level says: “Learners understand why AI-generated outputs should be verified and recognise how they can verify the content themselves.”
Anthropic’s guide for developers suggests one way to cut made-up answers from its own model: “Explicitly give Claude permission to admit uncertainty.” That is Anthropic’s advice about Claude, checked in September 2026, not a guarantee. Lesson 2.6 will cover prompting, and why asking again is not checking.
The limits. The people study is from the US, and the student surveys are from the UK. The citation study and the court case are from 2023 and involve 2023 models. OpenAI’s figures are its own tests of its own models. The paper on why it happens is an argument. These studies do not report results for neurodivergent learners. The Irish High Court direction applies to court documents, not to every chatbot answer.
Check-in
A chatbot's answer names a study, with authors and a year, and it sounds sure. What does that tell you?
It wrote a likely-sounding answer. The study still needs checking.
Yes. A chatbot writes likely text, and nothing in that checks the facts. How sure it sounds is not a sign of being right.
The study is real.
Not this one. One study checked citations in answers written in the first week of April 2023. More than half (55%) of GPT-3.5's were to works that do not exist.
The study is made up.
Not this one. In that same study, many citations were to real works. How sure it sounds tells you neither way.
The chatbot looked it up.
Not this one. A chatbot can write it one likely piece at a time, with no search at all. And a search does not remove the problem (lesson 1.5).
You ask the same chatbot, "Is that study real?", and it says yes. Is it checked now?
No. Find the study itself, for example on the journal's own website.
Yes. Asking the same chatbot is not a check. The study itself is where you check.
Yes, it would know its own sources.
Not this one. In a US court case in 2023, a lawyer asked ChatGPT whether cases it had given him were real. It replied that they were. They were made up.
Yes, if a second chatbot agrees.
Not this one. Ireland's High Court has a rule for court documents. It says: "It is not sufficient to use the same or other GenAI Tools to confirm that information is accurate." GenAI means AI that makes new text or pictures. Its next sentence allows one route. "However, persons responsible for a Court Document may verify GenAI content by referring to original source material housed within a GenAI Tool."
Do newer chatbots still make things up?
Less often for many kinds of mistakes, but yes, sometimes.
Yes. The International AI Safety Report 2026 says many kinds of failures, such as hallucinations, have generally become less likely. It also says systems still often make mistakes on more complex tasks.
No, that problem is solved.
Not this one. The same report says systems still often make mistakes on more complex tasks.
Yes, more than ever.
Not this one. The same report says many kinds of failures have generally become less likely. OpenAI's own tests, reported in April 2025, found its newer o3 made more of them than its older o1. Both are OpenAI models. Those are one company's tests of its own models.
Opinion. The rest of this lesson is sourced fact; this part is one person's view.
A chatbot can sound certain and be wrong. You've just seen why. So should we stop trusting it? I don't think so.
I think of it like a map. A map is useful, and a map can be out of date. So you look up and check the road. That's all I'm asking for here. Informed trust. Blind faith hands your judgement away. Total rejection throws a useful tool away. Both are easier than thinking, and both cost you.
Understanding sits between them. Know how the answer was made. Know where it tends to slip. Then check what matters. Trust that has been checked is worth having.
A made-up detail in an AI's answer that sounds real. The word is borrowed from people. Google's glossary says "confabulation" is probably a more accurate word.
made-up detail
A detail in an answer that sounds real and is not. It could be a study that does not exist, a wrong year, or a name that does not belong.
citation
A reference to a source, such as a study, so a reader can find it. It gives details like the authors, year and journal.
journal
A publication, printed or online, where researchers publish their studies.
base model
A language model after its first training on text, before it is tuned into a chatbot.
Where next
An article in Scientific Reports, 2023. It shows citations checked at scale. Free to read, with no login. It is on a different website.
