How we use AI

We are a publication about AI, and we use AI in producing it. This page sets out where AI helps in our work, what we never use it for, and who is accountable for every published claim. The line does not move for convenience.

Why we disclose this

We investigate AI products for a living. Every review on this site weighs what a vendor claims about its own AI against what the evidence actually supports. A publication doing that work owes its readers the same scrutiny of itself, in the same detail, on a page anyone can read before they trust a verdict.

So we publish where AI sits in our production and where it does not. Transparency over marketing is one of our editorial principles, and AI use is one of the things we disclose openly, alongside affiliate ties, paid placements, and method. If our use of AI changes in a way that affects what you read, this page changes with it.

What AI helps with

AI helps us with four things:

• — Research assistance and source discovery
• — Drafting and structure
• — Data cleaning and cross-checking
• — Language editing

That is the full list. In each case a person directs the work, and a person is answerable for what comes out of it.

What AI is never used for

• — Inventing sources or quotes. Every source we cite is a real document or a real user report that a person opened.
• — Fabricating tests or first-hand experience. We do not run hands-on product trials, and no AI output may be written up as though we did.
• — Generating fake user reports. User evidence comes from real, attributed posts and reviews on public platforms. We never synthesize it.
• — Publishing without human editorial review. Nothing reaches the site on a model's say-so.

These are not preferences. If any of them ever appears on our site, it is an error to be corrected, not a practice to be defended.

Human accountability

Every page ships under named human editorial responsibility. AI Tools Police is founded and edited by Mucahit Kaya, and the final call on any published claim rests with a person.

Verdicts, scores, and corrections are human decisions, not model outputs. A draft is a starting point; it becomes a published claim only after a person checks it. Sources are opened and read by the editor rather than trusted blind from any AI output, and a source we cannot open does not become a citation.

Our evidence labels apply to AI-assisted work exactly as they apply to everything else. If a claim is labeled verified, it is because a person read the document or the live pricing page it comes from. If it is vendor-reported or unknown, we say so instead of letting a confident sentence stand in for a checked one.