Guides · ← All guides

Which AI Tools Train on Your Data?

Published: November 7, 2026 · Reviewed November 2026 · 8 min read

This guide is informational, not legal advice.

By the Ingenious AI team — building tools that watch AI vendors' terms so you don't have to.

The short answer

It depends on the vendor — and on your tier. TermsWatch tracks 13 AI vendors daily, checking their terms, privacy, license, and usage-policy pages. Its Vendor Data Scorecard reads each vendor's current wording and grades the training clause into one of four verdicts, from "does not train (claimed)" to "trains on your data." Vendor list and tracked documents as seeded in TermsWatch, November 2026.

Which vendors does TermsWatch track, and which documents?

The honest answer to "which tools train on your data" starts with knowing where each vendor's answer actually lives. A training clause can sit in the terms of service, the privacy policy, or a usage policy — and the license page answers a different question (what you may ship, covered in AI Model Licenses Explained). This is the exact document set TermsWatch checks for every tracked vendor. Document URLs were verified against the vendors' own legal pages on 2026-10-08 (terms, privacy, license) and 2026-10-12 (usage policies); a dash means no verified page of that type is tracked.

Scope note, stated plainly: TermsWatch does not publish a static per-vendor verdict table on this page, because a verdict printed here would go stale the day a vendor quietly edits its clause — which is the whole problem this platform exists to catch. The verdicts below are computed live from each vendor's latest snapshot in the Vendor Data Scorecard. What this page gives you is the tracked document set, the verdict scale, and the method — so any answer you read anywhere can be checked against the current wording.

What are the four possible verdicts on a training clause?

The scorecard derives a per-dimension verdict from the vendor's own latest snapshot text — a heuristic scoring of the wording, never a legal judgment. For training-data use, the four verdicts are:

The same snapshot also scores retention, deletion rights, human review of data, third-party sharing, data residency, and — where a license document exists — commercial use. Our longer guide, Is My Data Training Their Model?, walks the manual version of this check: find the documents, run the ten-phrase search ("train," "improve," "opt-out," and the rest), and classify the answer into the same buckets by hand.

Why does the tier you pay for change the answer?

Because the free tier and the paid tier are often different products wearing the same logo. Free tiers frequently reserve the right to train on inputs; paid and business tiers frequently don't. The pattern isn't universal — some vendors train on paid-tier data too, and some free tiers are clean — which is why the check has to run on the tier you're actually on, not the tier the vendor advertises. It's also why three teammates on individual free accounts can hold three separate data deals your workspace opt-out never touches, the blind spot we mapped in Who Owns What You Type Into an AI Tool?

How do you check a vendor TermsWatch doesn't track yet?

Use the free check on the TermsWatch home page: name a vendor or paste its terms URL and it resolves against the tracked library, returning its real recorded changes. A vendor that isn't tracked gets an honest "not tracked" answer — changes are never invented for untracked vendors. You can also add any AI vendor to a free watchlist by pasting its terms, privacy, or license page URL, and the public terms-change feed shows the format every detected change is filed in.

Frequently asked questions

Which AI tools train on your data?
It varies by vendor and by tier, and it changes when the terms change. TermsWatch tracks 13 AI vendors daily and grades each vendor's current training wording in its Vendor Data Scorecard as: does not train (claimed), training opt-out available, trains on your data, or not stated clearly.

How many AI vendors does TermsWatch track?
Thirteen seeded vendors are monitored daily: OpenAI, Anthropic, Google, Meta, Mistral AI, Stability AI, Midjourney, Runway, Kling, xAI, Cohere, Perplexity, and Replicate. Users can add any other AI vendor by pasting its terms, privacy, or license page URL.

Where does a vendor's training policy actually live?
Usually scattered: the terms of service hold the license you grant, the privacy policy holds the "how we use information" wording, and a data processing addendum or help-center article may describe the opt-out. Only the binding documents count, so the check reads all of them.

Does the free tier train on my data more often than paid?
Often, yes. The common industry pattern is that free tiers reserve training rights while paid and business tiers do not, but it is not universal. Read the clause for the tier you are actually on.

Can a vendor start training on my data after I sign up?
Yes. Most terms allow the vendor to update its policies, sometimes with only a "last updated" date change, and continued use counts as acceptance. That is why the training clause needs watching, not a one-time read.

What if a vendor's training policy is ambiguous?
Treat "we may use data to improve our services," with no mention of training or an opt-out, as a yes until proven otherwise, and ask the vendor the question in writing, including whether non-generative ML models are covered.

See your vendors' training verdicts in the free scorecard →

More from Ingenious AI: the fleet behind TermsWatch builds one focused tool per failure point — the full map is on the hub.