Skip to content
FANKAARstudio

Practice · 5 min read

What AI cannot do for your website

An honest list from a studio that uses it daily. The problems it does not solve, the ones it makes worse, and where the money is better spent.

We use these tools every day and they have genuinely changed how quickly work gets done. This is the other half of that, because a studio that only lists benefits is selling rather than advising.

It cannot decide what to build

The expensive mistakes on a project are almost never implementation. They are building the wrong thing well: a feature nobody asked for, a platform that does not fit, an assumption about customers that was never checked. No model has access to what your customers actually do, and it will confidently help you build the wrong thing faster.

It cannot make thin content rank

Search engines specifically target content produced at scale to manipulate rankings. Thirty generated articles is not thirty chances to rank, it is one signal that your site produces scaled content, and the penalty applies site-wide.

The thing that ranks is a page answering a real question with information the writer actually had. That is a research problem, and research is the part that cannot be skipped.

It cannot own the outcome

When a build ships with a security hole, or the migration loses rankings, or the checkout miscalculates tax in one region, nobody accepts "the model wrote it". Accountability sits with whoever shipped it, which is the reason review is not optional.

It does not know your codebase

It knows the general shape of code. It does not know that this project already has a currency formatter, that a table is intentionally denormalised, or why a decision was made two years ago. Ask it enough times and you get five near-identical helpers, none aware of the others, and a codebase that is quietly harder to change.

Where the money goes further

  • Fixing the four support questions people actually ask, on the pages where they ask them.
  • Making the site fast on a mid-range phone, which changes revenue on every page.
  • Writing three genuinely useful pages instead of thirty generated ones.
  • Talking to ten customers about why they nearly did not buy.

None of that is exciting and all of it outperforms adding an assistant to a site whose fundamentals are not right.

Working on something like this?

We build websites, stores and custom applications, and we will tell you honestly if the thing you are describing does not need one.