How We Test
Our full method, including the parts of it that are weaker than we would like.
The six criteria
Every platform on the comparison page is rated against the same six criteria. Each is scored out of ten, then combined into the overall figure shown on the card.
- Conversation quality. Does the writing read naturally, or does it fall into obvious loops? Judged from the provider's own demos, public transcripts and user reports.
- Memory between sessions. Whether details carry across separate conversations, based on what the platform documents and what users consistently report.
- Feature range. Text, voice, images, calls, character customisation — how much is included versus sold separately.
- Platform availability. Whether there are real iOS and Android apps or only a mobile web wrapper.
- Pricing. Entry price, what the free tier allows before it stops being useful, and how much annual billing saves.
- Cancellation. Whether you can cancel from account settings or have to email support. This is weighted more heavily than its share of the product, because it is where this category most often goes wrong.
How the criteria are weighted
Conversation quality and memory carry the most weight, because they determine whether a subscription survives past the first week. Pricing and cancellation come next. Feature range and platform availability carry the least, because a long feature list does not rescue a product that is dull to talk to.
What this method cannot tell you
This is the part most comparison sites leave out.
Our ratings are built from published information and reported experience, not from controlled side-by-side trials of every platform. That means we can tell you what a service offers, what it charges and how hard it is to leave — all verifiable facts. We are on softer ground judging how a conversation feels, which is subjective even when you have used the product, and varies enormously between users.
Two further limits worth stating. These platforms update frequently, so a rating reflects a moment in time rather than a permanent verdict. And we cover four products out of dozens; a service being absent from this page means we have not covered it, not that it is bad.
When ratings change
We review the page monthly and after any major pricing or feature change we become aware of. When a score moves, it is because a criterion underneath it moved — not because commercial terms changed.
Tell us we are wrong
If you have used one of these platforms and our rating does not match your experience, we want to hear it. Reader reports are the main way this page gets more accurate over time: hello@aicompa.org