To know whether a meeting tool trains AI on your conversations, you have to check two things, not one: whether the tool itself trains on your data, and whether the third-party AI vendors it sends your audio to — the speech-to-text engine and the language model — are contractually barred from doing the same. Most "we don't train on your data" promises only cover the first. Here's how to actually check, and the questions that surface the gap.
Why "we don't train on your data" isn't the whole answer
Almost no meeting tool runs its own speech recognition and its own large language model. It sends your meeting audio to specialist AI vendors to transcribe and summarize it. That means your conversation passes through at least one or two companies other than the tool you signed up with — and each of them has its own data policy.
So when a tool says "we don't train AI on your data," the honest follow-up is: and your vendors? If the tool doesn't train on your data but hands it to a speech vendor whose default is to use customer audio for model improvement, your conversation is still training someone's model. The promise is only as strong as the contract behind the subprocessors.
Where to look — not the marketing page
The marketing page will always say the reassuring thing. Check the documents that are actually binding:
- The privacy policy and DPA. Search for language like "to improve our services" or "to improve our models" — that's training, phrased gently. A clear policy says plainly that your content is not used to train models.
- The subprocessors list. A serious tool publishes the third parties it sends your data to. If there's no list, you can't evaluate the real question. If there is one, you can look up each vendor's own training stance.
- The default, not the toggle. "We don't train on your data" should be the default, not a setting you have to find and switch off. Opt-out buried in account settings is a different promise than opt-out by contract.
The questions to ask any tool
- Do you train any model on our meeting content? (Looking for: no.)
- Which subprocessors receive our audio or transcripts, and are they contractually barred from training on it? (Looking for: a published list + a yes.)
- What's actually retained — the audio, or only the text? For how long? (Looking for: audio discarded, text kept only as long as you want it.)
- Where is it stored, and can we choose the region? (Looking for: a real answer — US / EU / APAC.)
- For the strictest cases, can we self-host so nothing leaves our infrastructure? (Looking for: yes, at least on an enterprise tier.)
A tool that can answer all five concretely is being straight with you. One that can only answer the first is doing the quiet work in the gap.
How Sageio answers them
Plainly, because these are the questions we get asked:
- We do not train AI models on your meeting content — and our AI vendors are contractually restricted from doing the same. That second clause is the one most tools can't make.
- Audio is processed in memory and discarded. Only the text transcript and summary are kept, encrypted in transit and at rest, in the region you choose (US, EU, or APAC).
- We publish our subprocessors so you can verify the chain yourself, and a DPA with Standard Contractual Clauses is available for review.
- Enterprise customers can self-host the entire stack, so your meeting data never leaves your own servers.
You can read the actual documents rather than take our word for it: the subprocessors list, the DPA, and the security overview.
(Today the bot runs on Google Meet; Zoom and Microsoft Teams support is coming soon.)
Frequently asked questions
Does AI meeting transcription train models on your conversations? It can — it depends on the tool and its vendors. Most meeting tools send your audio to third-party speech and language-model providers, and some of those providers use customer data to improve their models by default. The tool's own "we don't train" promise doesn't cover a vendor whose contract allows it.
How do I check whether a meeting tool trains AI on my data? Read the privacy policy and DPA (not the marketing page) for phrases like "to improve our models," check whether the tool publishes its subprocessors, and ask specifically whether those subprocessors are contractually barred from training on your content. No published subprocessor list means you can't verify the real answer.
What's the difference between "we don't train" and "our vendors don't train"? A tool can avoid training its own models while still passing your audio to a vendor that trains on it. The conversation is only protected end-to-end if both the tool and every subprocessor in the chain are barred from using it for training.
Does Sageio train AI on my meetings? No. Sageio does not train models on your meeting content, and its AI vendors are contractually restricted from doing so. Audio is discarded after processing; only the encrypted text transcript and summary are retained, in the region you choose, and Enterprise customers can self-host so data never leaves their servers.
What does it cost to try? Every plan starts with a free 60-minute trial, no credit card required. After that, Professional is $49/month and Teams is $99 per seat/month (annual billing includes 2 months free); Enterprise is custom-priced.
If data handling is the thing standing between you and a decision, the fastest path is to read the three documents linked above and ask any tool the five questions. Then add the bot to one real meeting and see the rest for yourself.