Does Your AI Notetaker Learn From Your Edits, and Can You See What It Learned?
You fix the same things in every summary and hope the tool notices. Some notetakers never learn from an edit, some build a profile of you from your meetings, and almost none show you which output that profile changed. Here is what Granola documents, how Driffle's opt in personalization works, and four questions to ask in a trial.
The edit you keep making
Most people who use an AI notetaker for a few weeks end up making the same edits. Decisions moved to the top. Action items cut to one line. The client's name for the project instead of the one the model guessed. The obvious hope is that the tool notices and stops making you do it.
Whether it does depends on which kind of learning the vendor means, and the two kinds carry very different privacy terms.
Two different things get called learning
The first is style: the tool watches how you edit and adjusts how it writes for you. The second is a profile: the tool reads your meetings and writes down who you are, what you care about and who you work with.
Granola is a useful example because its docs describe both. On style, the answer is no. Granola says it processes each meeting independently, so an edit applies only to that note and does not change how future notes are written. For a consistent style it points you to templates and to Settings, Profile, where you type your job title and some context about your work.
On profile, the answer is yes, on Business and Enterprise plans. After you have transcribed 10 meetings, Granola writes an agent brief from your meetings, with cards for current priorities, key relationships, projects, glossary, habits, communication style and identity and role. It refreshes overnight. You can edit or delete any card, and connected AI tools such as Claude or ChatGPT can read the brief over Granola's MCP connection. On Enterprise, an admin can turn that reading off.
Why the profile is the part to read closely
A style signal says something like keep action items short. A profile says something like key relationship: the client you are renegotiating with. The second is a new document about you that did not exist before you started recording, and it is written for other software to read.
That is not a reason to avoid it. It is a reason to know what it is built from, whether it is on by default for your plan, and which tools and people can read it. Being able to edit it after the fact is good. Seeing every item, and switching off one without deleting the rest, is better.
How Driffle handles it
Driffle learns style, not a profile of your relationships, and only if you ask it to. The switch is in Settings, Privacy and data, labelled Use my activity to improve replies, and it is off until you turn it on. The app's own description: when on, Driffle adjusts the wording, detail and order of your summaries and Ask answers using what you set and how you edit your summaries. Facts, owners and dates never change, and only you see it.
Under the switch, See what Driffle uses lists every signal. Each row says whether you set it or it was learned from how you edit, where it is used, or why it is not used yet, for example when Driffle has seen too little to be sure. Each row has Keep, Don't use and Use again, so you can turn off one signal without touching the others. Meeting summaries only follow your editing style; a preference about your role can shape an Ask answer but never reframes what a meeting was about.
When a summary or answer was actually shaped by these signals, a line under it reads Personalized for you, with a why link that opens the list. If nothing was applied, no line appears.
Two limits worth knowing. The rule that personalization never changes facts, owners or dates is an instruction to the model, with the facts still taken from the transcript, so check an important summary against the transcript as you would anyway. And Ask conversations are saved to your workspace so they stay searchable; that cannot be turned off yet. Processing runs in the cloud, and Driffle does not use your content to train any model.
Four questions to ask in a trial
What is it built from: your edits, your meetings, your email? Is it on by default on the plan you would buy? Can you see each thing it learned and turn off one at a time? And when an output was changed by it, does the tool tell you, so you can tell a personalised summary from a neutral one?
A quick test: make the same edit to three summaries in your first week, such as moving decisions to the top, then look at the next summary and at whatever settings page claims to show what the tool knows about you. If you cannot find that page, you have your answer.
If you want a notetaker that learns only what you let it and shows you where it used it, you can try Driffle on an Apple Silicon Mac at driffle.ai/download.
Sources
- AI enhanced notes. Read 16 November 2026. Granola processes each meeting independently, so edits to a note apply only to that note and do not influence future notes; templates or profile settings are the route to a consistent style; edits are taken into account when that note is regenerated.
- Profile and preferences. Read 16 November 2026. Settings, Profile takes your job and extra context to guide summaries; Granola uses anonymized data to train its models, with an opt out under Settings, Preferences, Data and sharing; third party providers may not train on your data.
- Agent brief. Read 16 November 2026. A brief written from your meetings on Business and Enterprise after 10 transcribed meetings, with cards for priorities, relationships, projects, glossary, habits, communication style and role; refreshed overnight; editable; readable by connected AIs over Granola MCP; Enterprise admins can block that reading.
FAQ
Does Granola learn from my edits to meeting notes?
Not across meetings. Granola's docs say each meeting is processed independently, so an edit applies only to that note. Your edits do count if you regenerate that same note. For a consistent style Granola recommends templates and the job and context fields in Settings, Profile.
Does Driffle learn from how I edit my summaries?
Only if you turn on Use my activity to improve replies in Settings, Privacy and data. It is off by default. When on, Driffle adjusts wording, detail and order in your summaries and Ask answers based on what you set and how you edit.
Can I see what Driffle has learned about me?
Yes. Under the switch, See what Driffle uses lists every signal, where it came from, where it is used or why not, and lets you Keep, stop using, or use again each one. An output it shaped shows Personalized for you with a why link to that list.
Can personalization change the facts in my notes?
In Driffle it is limited to presentation: emphasis, order, wording, detail and next steps. The model is instructed never to change a fact, owner, date or decision, and facts still come from the transcript. It is still sensible to check important summaries against the transcript.
Is learning my style the same as training on my meetings?
No. Style is a setting for your own outputs. Training means your content improves a model for other users. Driffle does not use your content to train any model. Granola's docs say it uses anonymized data to train its models unless you opt out in Settings, Preferences, Data and sharing.