What an AI Notetaker Actually Lets You Export

The button is not the answer

Ask any meeting notes vendor whether you can export your data and the answer is yes. It is yes at Otter, yes at Fireflies, yes at Granola, yes at Fathom. The word export appears in every help centre and on most pricing pages, which is precisely why it tells you nothing.

The useful questions are narrower. Which fields come out. On which plan. In which file format. Scoped to you, or to your company. Available while you are paying, or still available on the day you stop. Four of those five have answers published by the vendors themselves, and once you line them up the picture is less reassuring than the marketing word suggests.

The transcript leaves and the action items stay

Otter's help centre is unusually direct about this. On exporting a conversation it says that Action Items "can't be exported directly, but you can copy them manually", and that the Outline "can't be exported as a file", though you can copy the summary and pull the items out by hand. The transcript exports fine. The two artefacts a busy person actually opens after the call do not.

Fathom inverts the same asymmetry in a different way. Its documentation says that "while transcripts can't be downloaded directly, you can copy and paste them anywhere you need", with a Copy Transcript button above the text. Meanwhile the recording has a real download: pick the meeting, click the three dots, choose Download Video, and you get an MP4 named after the meeting.

Sit with that for a second. The heaviest, least useful artefact in the whole system, a video file of people talking, has a one click download. The text you would actually paste into a doc, a ticket or a handover has a clipboard button and nothing else. You can walk away with two hundred MP4 files and not one text file.

None of this is a conspiracy. Export is engineering work, and it gets built for whichever object is easiest to serialise and most often requested by support tickets. But the effect on a buyer is real: the parts of the product you were sold on, the summary, the action items, the structure, are often the parts with the weakest exit.

Export is frequently a paid feature, so a trial cannot test it

Fireflies states it plainly in the article describing its free plan: "Downloading transcripts requires a paid plan." Everything else on that plan is generous, unlimited transcription through auto-join, more than a hundred languages, all integrations available. You can put a year of meetings in and take none of them out.

Otter gates it by format and by volume. Its export page notes that "Basic plan users can export conversations only as TXT files", with other formats behind an upgrade, and its bulk export article opens with the line "The bulk export feature is only available on the Otter Business and Enterprise plans." Below that upgrade line, export means one conversation at a time.

The consequence is the part nobody mentions in a buying guide. If export sits behind the upgrade, then the trial you run to evaluate the product cannot evaluate the exit. You test capture, quality, search and share on the free tier, sign a contract on the strength of it, and only meet the export path a year later when you are trying to leave. By then the data you want out is the whole year.

The file is a subset, not a copy

Granola documents its export carefully enough to show what this looks like in practice. You go to Settings, then Profile, then Generate CSV, and a CSV arrives by email within a few hours carrying the title, note summary, transcript and other basic details for each note. It includes your full history, which is more than the app itself shows on the free tier.

Then come the conditions. The docs state that the export "only includes notes that have a summary", that deleted notes and notes without a summary are left out, and, in their own words, "this means the number of notes in your export may be lower than the total count shown in the app". So a meeting you scribbled into during a call and never let the AI process is not in the file. The download link is valid for 24 hours, and you can only generate one export per account per 24 hours, which makes a botched attempt cost you a day.

Otter's bulk path has a mechanical version of the same problem. Its own FAQ explains that Select all "will only select the conversations loaded on the current page", so to bulk export everything you first have to scroll until the whole account has loaded into the browser. A year of daily meetings is a lot of scrolling, and the help article adds that an adblocker can stop the export working at all.

Your export is yours alone, which is the company's problem

Here is the finding that should change how a team evaluates this. Granola's export documentation asks whether workspace admins can export notes for the whole team and answers, in one word, no. There is no company-wide or admin-level export, and each person's export contains only their own notes.

Read that as an operator. Your company's meeting record is not one archive with an export button. It is a stack of individual archives, and a complete copy requires every person to run their own export, which means every person has to still be there, still have an account, and still be willing. Granola's docs are candid about the edge case too: a departed colleague's private notes are not reachable by admins, and the suggested remedy is to temporarily re-add them to the workspace so they can export their own.

This is not unique to one vendor, and it is not obviously wrong either. Per-person scoping is a reasonable privacy position, and a company-wide export of everyone's one to ones would be its own scandal. But it does mean the honest answer to "can we get our data out" is usually "each of us can get some of ours out", and that is a different sentence to put in front of a board or a buyer.

Four questions for day one of the trial

Run these in the first week, while you are still curious rather than annoyed. First, export one meeting and open the file: are the summary and the action items inside it, or only the transcript? Second, check whether bulk export and the good formats exist on the tier you are actually going to buy, not on the tier in the sales deck. Third, count what came out against what is in the app, because a documented subset is easy to miss when the file is large and you are in a hurry.

Fourth, and this is the one people skip, find the vendor's written answer to what happens after you cancel. How long the data stays, whether export still works during that window, whether a downgrade removes the export you would need. If the answer is not published, that is the answer. Getting these four right during a trial costs an hour. Getting them wrong costs whatever a year of your team's meetings is worth.

Where Driffle sits, including the limits

Driffle's own position is on the Trust page, in specifics rather than adjectives. Settings then Export generates a ZIP containing every note as Markdown, every transcript as plain text, and a JSON manifest of the metadata. It is produced asynchronously and emailed when ready, usually within minutes. Markdown and plain text because those open in anything, and the manifest because metadata is what makes the pile navigable once it is out.

Export is not a tier. The pricing page commits to full data export being identical on every plan, alongside the sensitive floor, the capture inspector and per-item delete, on the stated grounds that paywalling a privacy control is a dark pattern. Cancellation is written down as well: memory is frozen rather than deleted, access runs to the end of the billing period, and data is preserved for 30 days in case you come back. The point of publishing that is so you can test the exit on day one instead of trusting a promise on the day you leave.

The limits, so you can price them in. Driffle runs on a Mac with Apple Silicon on macOS 14.4 or later, as a direct download, and there is no Windows, Linux or Intel build to hand you today. Capture happens on your machine but transcription and the AI work run in the cloud, so this is not an offline or on-device tool. All customer data sits in AWS regions in the United States, with EU residency on the roadmap rather than available. If those work for you, the honest test of everything above takes about ten minutes: download it at https://driffle.ai/download, capture one real meeting, then export it and read the file.