Which checks use AI
Of the eight checks you can run on their own, six need no AI: Page health, AI compatibility, Data saving, Regression check, Staging vs live and Report. Site mapping, which IT’s checks run, needs none either. They are free and unlimited on every plan. When a run has AI, Report adds a short written review, though every number on it is counted without.
Two of the eight use it. Site review reads every page the way a first-time visitor would and ranks the fixes. User simulation sends a tester through your product. So do Strategy judgment, which Finance’s checks add on a paid plan, and a validation: the panel of simulated buyers and the walks through your product. The verdict on Home cannot say Ready to launch without one, because until somebody has tried the product, simulated or real, the best it can say is Not yet.
Whose AI it is
On the hosted console, every AI call runs on our account with a third-party AI provider. You cannot paste a key and you cannot pick a model. Where a console on your own machine has API keys in its settings, the hosted one has Variables, which is only for values your runs need to reach your product, such as a preview bypass secret.
Page text and screenshots go to the third-party AI providers we use to read them. Nothing is sold, shared, or used for training. The privacy page names them.
Which model
We choose the model for each job based on measurements. The simulated people run on a cheaper model that measured as well or better for that work. Site review runs on the model that made up fewer claims in our tests, because it is the one that files findings. It is not a setting on your account.
How it counts against your plan
Everything that uses AI draws on your plan’s monthly allowance, which is counted in validations. Pro covers about 10 a month, Team about 25 and Agency about 100, under fair use. The pool is shared by everyone in the workspace and starts again on your renewal day.
Free runs every check that needs no AI, and one full check with AI each month, with no card. Nothing is charged until you choose a plan.
When it runs out
Usage, under More, shows how many validations are left this month and the runs and model calls that used them, and Home shows the number left too. Settings → Plan says in words what you have used, tells you when only a few validations are left, and says plainly when you have reached the month’s fair use, with the day it starts again. Until then the checks that use AI wait. The free checks are not affected.
To get more before then, buy a top-up ($20 for about 5 validations, or $80 for about 25; they never expire and are used after the month’s pool) or move up a plan. Plans and billing has the details.
If you run it yourself
On your own machine, it works differently. The checks that use AI run on a model key you set there, billed to you by that provider, and nothing is metered or capped by us. You choose the model too, and your console shows what each run cost, at the model’s list price. Run it yourself shows where the key goes.