Skip to content
nokofy

How to rank in AI Overviews — and keep your clicks

The AI overview at the top of the results is taking clicks. Why it happens, which pages lose most, and how to rank inside the overview yourself.

· Nokofy

You type a question into Google and get an answer straight away, at the top, assembled from several sites. Below it are the ordinary results — but only after you scroll. If the answer is good, many people won’t scroll at all.

That’s the biggest change in search since it began, and there’s no point pretending otherwise. But the panic around it mixes up two different things.

What actually changes

Not everyone loses equally. The loss depends on the kind of question:

Type of queryExampleWhat happens
Fact“how long to roast a turkey”AI answers, the click doesn’t happen
Definition“what is a canonical tag”AI answers, the click doesn’t happen
Comparison“laminate or engineered wood”AI summarises, some people still click
Buying“laminate flooring price London”the click still goes to the site
Local“plumber near me”the map, not AI
Specific task“VAT return deadline 2026”AI answers, but the source gets checked

The pattern is clear: the closer a question is to buying, the less AI takes. Nobody buys a kitchen from a summary.

That’s why a company selling something concrete loses less than it seems — it loses informational visits, which weren’t producing orders directly anyway. It stings, but it isn’t the same as losing sales.

Where AI pulls its answers from and who it mentions

Why this is happening and why it won’t pass

Because it’s faster for people. There’s no conspiracy and no penalty here — if somebody gets an accurate answer in two seconds instead of two clicks, they’ll take the answer. Search goes where people’s behaviour goes.

The same applies outside Google: more and more people put the question straight to ChatGPT, Perplexity or Gemini and never open a list of results. That’s why it’s no longer enough to ask “what position am I in” — you also have to ask “do they mention me at all when somebody asks”.

What can be done

Three things, in order of usefulness.

1. Write so that you’re easy to quote

An AI answer is assembled from sources that are easy to read. That means:

  • the subheading is a question, and the answer sits immediately below it, in the first two sentences
  • concrete numbers, dates and prices instead of “it depends on the situation”
  • a table or a list wherever you’re enumerating
  • one topic per page — a text covering five things doesn’t get quoted for any of them

This isn’t a new skill. It’s the same thing that makes a text good for a person in a hurry.

2. Give a reason to be trusted

The systems that assemble answers choose sources by how reliable they seem: who the author is, whether there’s a company behind the site, whether the text is dated and updated, whether you cite a source when you claim something. A site with no name and no contact details gets quoted less often.

3. Move towards what AI can’t do

AI summarises what already exists. It can’t summarise:

  • your price and your lead time
  • your photograph of the job you finished yesterday
  • experience from the field — what breaks most often, what people most often order wrongly
  • your calculator, configurator or tool on the site

The more of that there is, the more reason to click despite the summary.

What not to do

  • Don’t remove content that lost clicks. If AI quotes you, it builds recognition even when the click doesn’t happen — and having your name mentioned in an answer has its own value.
  • Don’t block AI bots in the hope that clicks will come back. You’ll return to invisibility, and the clicks won’t come.
  • Don’t chase “AI SEO” tricks. For now, what works is the same thing that works for ordinary search: clear text, accurate data and a reason to be trusted.

How to know where you stand

For ordinary results you have Search Console. For AI answers you have nothing — no official report says how many times ChatGPT mentioned you.

The only way is to ask directly: put the questions your customers actually type and check who the models name as the answer, and which site they pull from. We do that across several models at once — that’s what the AI visibility check is.

In short

  • The AI overview takes most from factual and definitional queries, and least from buying and local ones.
  • Write so that you’re easy to quote: the question as a subheading, the answer in the first two sentences, concrete numbers.
  • Give a reason to be trusted — an author, a company, a date, sources.
  • Invest in what AI doesn’t have: your prices, your experience, your tools.
  • Measure mentions in AI answers too, not just your position in the results.

Check this on your own site

Everything we write about is something you can measure in the tool — the audit, positions and the real questions people type in your industry. The free plan asks for no card.

No card · No contract · Cancel anytime