Active Note-Taking Is Thinking, Not Transcribing

The bot will keep the transcript. Your job is the synthesis the transcript will not do for you — the decision named while it is warm, the action picked up before the call ends, the open question held for next time. The writing is the thinking; the transcript is the record. Two notes, two jobs.

A 177-point Show HN in December 2025 was titled, in the author's own words, "I built a system for active note-taking in regular meetings like 1-1s."¹ The post is a builder describing what he kept doing by hand after the AI notetaker arrived: "not minutes, not transcriptions or AI summaries, but me using my brain to actively pull out the key points in short form bullet-like notes, as the meeting is going on."¹ That phrasing is the topic of this essay. The encoding-hypothesis literature — formal language for the same idea — calls it "the processing that occurs during the act of note taking improves learning and retention."² The shape underneath the AI-bot moment is not new; the moment is.

The setup: the transcript is not the note

Most people, when an AI notetaker joins the call, stop writing. The reasoning sounds clean: the bot captures every word, so typing alongside it is redundant. The trouble is that "every word" is not the same shape as "what mattered." A transcript records; a note resolves. They answer different questions for the reader who comes back to them next week.

The literature on note-taking is precise about what the writer is doing while writing. Mueller and Oppenheimer, in The Pen Is Mightier Than the Keyboard, frame the encoding hypothesis as the textbook reason note-taking helps in the first place: "The encoding hypothesis suggests that the processing that occurs during the act of note taking improves learning and retention."² What is doing the work is not the document; it is the choosing-what-to-write step that happens before each character is typed.

A transcript skips that step on the reader's behalf. That is its feature when the goal is to recover what was said. It is its limit when the goal is to leave the meeting knowing what to do. The two artifacts are not redundant — they are complementary. The honest framing is the one the Show HN author landed on: both can coexist; they do different jobs.¹

The pivot: the keyboard captures more and learns less

Mueller and Oppenheimer's 2014 finding in Psychological Science was that laptop note-takers wrote significantly more words than longhand note-takers — Study 1 reported a mean of 309.6 words typed versus 173.4 words by hand³ — and scored worse on conceptual-application questions about the lecture material.³ The keyboard was capturing more and synthesizing less.

The mechanism the authors proposed was transcription: "laptop note takers' tendency to transcribe lectures verbatim rather than processing information and reframing it in their own words is detrimental to learning."⁴ What the fingers do under load is type the room. What the eyes do, when the fingers are taking dictation, is glaze. The result is a longer document and a shallower understanding.

The most important Study 3 result is that telling the typists to stop transcribing did not work. "The instruction to not take verbatim notes was completely ineffective at reducing verbatim content (p = .97),"⁵ the paper reports. The medium drove the behavior; willpower did not. That is the finding that survives whatever happens to the test-score effect, because it describes a tendency, not a result.

The replication is where intellectual honesty matters. In 2021 a 81-co-author Tufts team led by Heather Urry ran "a direct replication of Mueller and Oppenheimer's (2014) Study 1" with 142 participants and reported, in the abstract: "laptop participants did not perform better than longhand participants on the quiz. Exploratory meta-analyses of eight similar studies echoed this pattern… Overall, results do not support the idea that longhand note taking improves immediate learning via better encoding of information."⁶ The word-count gap and the verbatim-overlap gap held. The immediate-test-score gap did not.

That is the careful claim this essay sits on. The mechanism the original paper named — transcription as shallow processing — is what replicated; the specific immediate-recall benefit of longhand did not. The actionable lesson is not handwrite everything. It is do not let your fingers capture what your brain did not decide to write.

The argument: the writing is the active ingredient

If transcription is the trap and synthesis is the work, the right question in 2026 is not handwriting vs typing; it is am I writing, or am I just typing the room. An AI notetaker removes typing entirely. What it cannot remove is the deciding. Two recent lines of evidence, flagged carefully, point the same direction.

The MIT Media Lab's Your Brain on ChatGPT (Kosmyna et al., arXiv preprint, 2025-06-10) reports an EEG study where "LLM users displayed the weakest connectivity. Cognitive activity scaled down in relation to external tool use… Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group. LLM users also struggled to accurately quote their own work."⁷ The headline is the direction: external-tool use scales cognitive engagement down, and authorship feel scales with engagement. The study is a preprint with 54 participants, peer-review pending, and has drawn a published critical comment; cite it as supporting, not load-bearing.⁷

The Gerlich survey AI Tools in Society (Societies, 2025) interviewed 666 participants and found "a significant negative correlation between frequent AI tool usage and critical thinking abilities, mediated by increased cognitive offloading."⁸ The study is correlational and the journal has issued a published correction; the direction matches the offloading mechanism but the magnitude is not the claim to lean on. Cite it as supporting, not load-bearing.⁸

Even the act-of-writing-itself literature outside the learning frame says something compatible. James Pennebaker — whose forty years of expressive-writing studies sit in a different domain entirely — frames the central question as why writing about a thing helps in the first place: "why do so many of us feel better after writing or talking about something that has been weighing on us?"⁹ Different field, same shape: the act of putting a thought into words is doing work the thought alone is not.

Steph Ango, writing on his personal site, lands the file-shape version of the same idea: "the files you create are more important than the tools you use to create them. Apps are ephemeral, but your files have a chance to last."¹⁰ The note you make during the meeting is one of those files. The transcript belongs to the bot vendor; the note belongs to you.

The practice: write the note while the meeting is happening

If the bot keeps the transcript, the meeting-time discipline is to write the three things the transcript will not surface for you next week: the decisions, the actions, and the open questions. Three bullets, written while the call is in motion. Then a thirty-second pass at the end to mark them clean.

1. Open one plain Markdown file before the call starts — 2026-06-30-1-1-alex.md is enough.¹¹ Three headings: Decisions, Actions, Open. No template framework, no app block types — just the headings the meeting needs. The full template version, with a five-minute and a thirty-minute variant, lives in how to take better meeting notes; this post is the while-it-happens discipline, that one is the file shape.
2. Write decisions in the speaker's words, as close to verbatim as a one-handed typist can manage. The decision is the thing the transcript would let you re-litigate later because the wording was loose. "Ship V2 to Customer A by Friday" beats "agreed to move forward on the V2 thing." Two minutes of attention now is half an hour of meeting-recap later.
3. Write actions with an owner and a date, even if both are guessed. A - [ ] checkbox with a name and a 2026-07-04 is a commitment the next 1-1 can audit; a transcript paragraph is not. The act of typing the date is the act of agreeing to the date. If the owner pushes back, the typing made the disagreement legible — that, also, is the work.
4. Write open questions verbatim, as questions. "What is the actual budget for the migration?" The format matters: a question phrased as a question is a question the next meeting can pick up; a question phrased as a vague worry disappears between meetings. Open questions are the inventory that compounds when you write them down and decays when you do not.
5. Let the bot do the bot's job. When the meeting ends, the transcript will land in your inbox or your drive. Read it once for the things your live note missed — usually a quote you want verbatim, occasionally a thread you were not in the room for. The bot is a backstop, not the primary record.

The artifact you actually own is the small Markdown file you wrote during the call. It opens in any editor, it lives on your device, it survives whichever AI notetaker your company is paying for this quarter. The three-bucket pattern is the same one the AI-coding diptych argues for in the IDE — see why one builder went back to writing by hand after vibe-coding — because the failure mode is the same: outsource the writing, outsource the thinking.¹² The cure in both domains is the same: keep doing the part the tool removed.

Frequently asked questions

Should I take notes if the AI is transcribing the meeting? Yes — the writing is the synthesis, and the transcript is not. Mueller and Oppenheimer reported that laptop typists captured more words (~309.6 vs ~173.4)³ but scored worse on conceptual application,³ and Urry's 2021 replication confirmed the word-count and verbatim-overlap gaps held even where the test-score gap did not.⁶ The transcript records; the note resolves.

What should I do during a meeting while the AI notetaker is running? Open a plain Markdown file with three headings — Decisions, Actions, Open — and write what the transcript will not surface for you next week. The Show HN that named it calls it "using my brain to actively pull out the key points in short form bullet-like notes."¹ Let the bot keep the receipt; you keep the thinking.

Active note-taking vs transcription — what is the difference? A transcript is everything that was said; an active note is what mattered. Mueller and Oppenheimer named the mechanism: typing tends toward verbatim capture, and "the instruction to not take verbatim notes was completely ineffective at reducing verbatim content."⁵ The point of writing is the choosing — the part a transcript does for you and a synthesis does not.

Is it better to take notes by hand or on a laptop? The honest answer is mixed: Mueller and Oppenheimer found a longhand advantage on conceptual application in 2014;³ Urry's 81-co-author 2021 replication did not find that immediate-learning gap.⁶ Both papers agree that laptops produce more words and more verbatim overlap. The actionable rule is active synthesis, not handwriting per se — write less, choose more.

Do I still need to take notes with an AI notetaker? Yes — keep both. The bot makes a complete record of what happened; you make a short record of what mattered. The two artifacts answer different questions for next week's reader. The Kosmyna preprint flags the cost of skipping the second one: "LLM users displayed the weakest connectivity… Self-reported ownership of essays was the lowest in the LLM group."⁷

What is the best way to take 1-1 notes? Use the three-bucket file structure — Decisions, Actions, Open — and write the names and dates while the call is in motion. The template companion, with a five-minute and a thirty-minute version, is how to take better meeting notes; this post is the why and the during-the-meeting discipline. The file shape is there; the writing habit is here.

Does taking notes actually improve memory? The encoding hypothesis, named in Mueller and Oppenheimer's introduction, is the standard textbook account: "the processing that occurs during the act of note taking improves learning and retention."² Urry's replication complicates the immediate-test version of the story; the processing-while-writing mechanism survives. Write to think, not to record — that is the part the transcript will not do for you.

The argument we sit on is David's, who built the practice into a tool and posted it on the day every meeting acquired a transcript bot,¹ and the cognitive-science scaffolding belongs to Mueller, Oppenheimer, Urry, and the encoding-hypothesis literature they all answer.²³⁶ MNMNOTE saves each meeting as a plain Markdown file on your device — three headings, your words, the same shape cat and every other editor on the machine already understands. The file is at mnmnote.com.