How we rank — and what our verdicts mean

Independent, evidence-based, honest about its limits.

By Ivar Berg · Updated May 2026 · How we rank

Some links are affiliate links: if you sign up through them we may earn a commission, at no extra cost to you. It never affects our ranking. Full disclosure.

What we do

Bot or Person compares AI customer-support bots on what actually matters to a support team: whether the bot escalates to a real human cleanly, whether it keeps context at the handoff, and what it truly costs once usage and seats are counted. Every page is built from primary sources we link and date — each vendor's own pricing and docs — and from real, attributed user reports on Reddit, vendor community forums, Hacker News and Product Hunt.

What we do not do (yet)

We do not run a hands-on lab benchmark of every tool. Verdicts are based on documented pricing/capabilities and real-user evidence, not a private test we ran. If we publish hands-on tests, each result will carry dated evidence you can check.

How we make money

Some links are affiliate links: sign up through them and we may earn a commission at no extra cost to you. It never changes a ranking. Tools with no affiliate program (Intercom, Ada, Decagon, Forethought) are covered on the same terms. See our full disclosure.