How the AI data-policy index is made
Where each answer comes from, what Stated, Inferred and Not stated mean, how quotes are checked, and how the index is kept current. Policies last captured 2026-11-08.
What this is
A side-by-side reading of the data policies behind popular AI products: chat apps, APIs, enterprise plans and personal agents. For each product it answers the same 16 questions, from how long your prompts are kept to the license you grant over them. It is a reference, not legal advice. Your contract, enterprise agreement or regional terms can override what is here.
Where answers come from
Only the providers' own pages: privacy policies, terms, help-center articles, API data documentation and the privacy FAQs for agent products. We save the full text of each page with the time it was captured and a SHA-256 hash of the text, so a later edit to a policy shows up as a changed hash. Each provider page lists the documents read.
Every answer quotes the policy it rests on. A quote has to appear word for word in the captured text, and the site will not build if one doesn't, so a paraphrase can't be shown as the provider's words.
What the labels mean
- Stated The policy says it outright.
- Inferred It follows from the policy's wording; the answer says how.
- Not stated None of the captured pages addresses the question. That is not the same as the practice being absent. Each provider page lists its open questions and the page most likely to answer them.
How it is checked
Answers are kept to 30 words. After drafting, a separate review reads each answer against its quotes, looking for dropped qualifiers ("up to", "may", "unless"), the wrong tier or product, and inferences labelled as statements.
How it is updated
A weekly job reads every policy page again. When a page's text has changed, it saves the new text, re-checks every answer that quotes that page, and proposes the updates. A person reviews each change before it appears here. The capture date shows when the text we quote was last saved, so it moves only when a policy changes. A few providers block automated readers; their pages are saved by hand and show the date they were saved.
Use the data
data-policies.json has every answer, quote and source. llms.txt lists the pages. To report an error or suggest a provider, email support@gjalla.io.