How we test AI software

We buy or trial our own access

We sign up the way you would: a standard trial, or a paid plan bought at the advertised price. We do not score a product based on a vendor-configured demo account, because a demo account is a showroom, not a test.

Where a vendor provides extended access for testing, we say so on the page. It changes nothing about the score.

Every product in a category runs the same tasks

Each category has a fixed task list. For AI receptionists that includes an inbound call with background noise, an off-script question, an interruption mid-sentence, and a request the agent cannot fulfil. For automation platforms it includes an eight-step workflow with branching and a deliberate failure.

Running the same tasks is what makes scores comparable. It is also why we cannot score a product quickly.

We verify pricing separately

Advertised price, renewal price, and the cost at three times typical usage. The date we checked appears on the page. Where you see no date, we have not verified it ourselves yet and you should confirm before buying.

Nothing gets a score until it has been tested

Products enter our directory as soon as we can verify their pricing and features. They carry no rating until we have used them. A page marked "awaiting testing" means exactly that.

This is deliberate. A site that scores forty products in a week has not tested forty products.

We re-test on a schedule

Pricing every 60 days. Scores every 90 days, or immediately when a product ships a significant change. The date on the page is the date we last looked, not the date we first published.

Tell us when we are wrong

Prices change without notice and features ship weekly. If something here is out of date, tell us and we will check it.