Plenty of people have tried this already. They ask an assistant how to improve their SEO, get a competent summary of generic best practice, follow some of it, and conclude that AI is overrated for this.

The conclusion is understandable and wrong. The problem is not the reasoning. It is that the assistant is working blind.

Why the answers are generic

Ask it what to do about your SEO and consider what it actually knows: nothing about where you rank, nothing about what you nearly rank for, nothing about your competitors, nothing about the sixteen months of Search Console data quietly telling you which questions Google already associates with you.

Given nothing, it can only return what is true for everyone. Write good content. Get some links. Improve your speed. It is not being lazy; it is being asked to advise on a business it cannot see.

The fix is data, not a better prompt

A great deal of effort goes into prompt engineering for this, and it barely helps. A more elaborately worded request for advice about an invisible website produces more elaborately worded generic advice.

What changes the outcome is connecting the assistant to your actual measurements. Once it can read your rankings, your competitors, your Search Console history, your audit findings and your AI visibility, the questions it can answer change category entirely:

  • Which of my keywords is close enough to the first page to be worth a week of work?
  • Did my traffic drop because of a Google update, or because of something we changed?
  • What should I write next quarter, and by when, given my seasonality?
  • What is wrong with my site, in priority order, as a brief my web company can act on?

Those are answerable with data and unanswerable without it.

How the connection works

There is now a standard for this called MCP, the Model Context Protocol. A product publishes a set of tools, your assistant calls the ones it needs, and you just ask questions in ordinary language. No copying screenshots into a chat window.

Reasonable things to check before connecting any business system to an assistant: can it change anything, can it spend money, and what happens to the data. For ours: it has no access to your website at all, no tool can trigger paid data collection, and of twenty tools the only two that write add a dated note to your own charts and update your business context. The full explanation is here.

What it still cannot do

Be straight about this. It cannot earn links from real people, which is relational work. It cannot decide your commercial priorities. It cannot make a page good on its own, though it can draft one and tell you exactly what the page has to answer.

What it replaces is the analysis: the expensive, slow, judgement-heavy step of looking at everything and working out what matters this month. That part it does well, and it does it in seconds, as often as you like.