The AI build checklist
You built something with AI. It looks fine. Here is how to find out whether it is ready for customers.
Nothing here requires you to be a developer. Most of it you can check by asking the AI that built the thing. It has the whole codebase in front of it, and it will answer honestly if you ask directly. That is a better test than taking our word for it, because the tool ends up telling you itself.
Work through it and one of two things is true. Either your build clears it, in which case you have saved yourself real money and we would say so. Or it does not, and now you know precisely where and why.
Ask your AI
Paste these more or less as written.
"Audit every colour value in this project and tell me how many distinct ones there are, grouped by hue."
Do this one first, because the answer is usually startling. If it reports a hundred and fifty greens, there is no palette, just a fresh decision every time something needed to be green.
Why it matters: This is what makes a site feel subtly cheap, and it compounds every time anyone touches it.
"How many distinct font sizes, weights and spacing values are in use?"
A system has a handful of each, used on purpose. Drift has dozens, each chosen in the moment and never reconciled.
"Is there a design system here? Show me where the rules are defined."
If the answer describes what the code happens to do rather than pointing at defined tokens, the rules do not exist. They were improvised, and then inherited.
"What are the interface components built on?"
If buttons, menus, dialogs and form fields were hand-rolled rather than built on an established library, follow up with: "Was keyboard navigation and screen reader support ever implemented?" Usually it was not. A good library gives you focus states, keyboard handling and correct semantics for free. Bespoke components give you none of it while looking perfectly fine to someone testing with a mouse.
Why it matters: Accessibility failures are a live legal exposure, and short of that it is customers who cannot use the site quietly leaving.
"Is anything being measured?"
Analytics at all? Is a booking, purchase or enquiry tracked as a conversion? Can you see where people leave? Most generated builds have none of this, because the tool was never told what success looks like.
Why it matters: Without it you cannot tell whether the site works. You can only tell whether it loads.
"What happens when someone submits this form?"
Where does the message go, who gets notified, is it stored. Prototype forms very often go nowhere at all, silently. The person who discovers it is a customer who believes they contacted you.
"Check the basics search engines read: heading structure, alt text, page titles and descriptions, link text."
Related to accessibility but not the same thing. Commonly skipped, because nothing looks broken without it.
Check with your own eyes
Open two pages side by side. Do the headings match in size and weight? Are the buttons identical? If not, you have pages rather than a site.
Add a new section yourself. Did you have to guess at a spacing value or pick a colour by eye? Guessed values are how drift starts, and it does not stop on its own.
Look at your smallest text on a phone, outside, in daylight. Can you read it?
Find the one thing you would delete if you had to. If nothing is deletable, nothing was decided.
Use it, don't look at it
Walk the whole path as a stranger. On a phone. Arrive cold, work out what this is, decide, and try to buy or book. Go all the way to the confirmation. This is where uneven journeys surface, and they surface no other way: some paths feel smooth and others feel convoluted, because a feature was prompted and the model filled the gap plausibly with nobody holding the whole flow in mind.
Check how fast it loads on mobile data, not on your laptop on office wifi. Generated pages often carry uncompressed images and more script than they need.
Ask what happens when someone who is not you edits this next year. If the answer depends on them having taste, the system lives in a person rather than in the work.
If it didn't clear
That's normal, and it is not an argument against building with AI. It is the difference between output and a finished thing. Most of what is on this list is invisible while you are building and obvious afterwards.
If you would rather not work through it alone, that is most of what we do.
Tell us about your business