How verdicts are decided

This site answers one question: is this product reasonable for ordinary business data, based on what the vendor says in public? It is not a penetration test, a legal opinion, or a promise that a tool is safe for health, finance, or secrets.

Safe means the vendor's own pages say this plan does not use customer content to train models by default, and they describe an organization control: an admin setting, a contract term, or a default that is already off.

Careful means training is on unless someone opts out, the default is not stated, the pages disagree, or a business plan exists but the control is easy to miss. Use it only after you change the setting and know which plan you are on.

Avoid means the vendor says customer content can be used to train models, and we did not find a business plan where that is off by default. A privacy-rights opt-out is not the same thing as an admin control.

Every factual line is taken from a vendor page we opened. The source list on each tool is those pages. If a page did not answer a question, the field says Unknown. We do not fill gaps from memory, reviews, or other people's summaries.

A verdict can be Safe for ordinary business documents and still be the wrong place for health information. Read the BAA line. "Enterprise only" means a business associate agreement exists for a specific plan or switch, not for the consumer product.

Suggestions sit in a review queue. They are not published until an entry is approved through the admin API. The date on each page is the day we last read the sources, not the day the vendor last changed them.

Checked November 10, 2026 for the first published set.