Your customers didn't stop looking for you. They stopped clicking.

When an AI summary appears, clicks fall from 15% to 8%. Here's what replaces ranking for a one-person business, and the five things worth doing about it.

Pew Research Center tracked 68,879 real Google searches from 900 US adults. When an AI summary sat at the top of the page, people clicked through to a normal result 8% of the time. When no summary appeared, 15%. And the citation links inside the summary — the ones added partly to answer publisher complaints — got clicked in 1% of visits.

Ahrefs, working from Search Console data across 300,000 keywords, put the same shift another way: top-ranking pages now see a 58% lower click-through rate where an AI Overview appears, up from 34.5% eight months earlier.

Read those together and the conclusion is uncomfortable but simple. The summary is not the shop window. It is increasingly the shop.

So what actually replaces ranking?

Being in the answer rather than under it.

Ranking was a competition to appear in a list of ten, where the reader did the choosing. Citation is a competition to be the sentence, where there is no list and no choosing. That is a harder game in one specific way — there's less room — and an easier game in almost every other, because the thing it rewards is not budget.

It rewards being describable. A model assembling an answer needs a source it can lift from without mangling: a plain claim, a number with a source next to it, a definition in the words a customer would use, a stated limit. Those are cheap for a one-person business to produce and oddly expensive for a large one, because large organisations approve copy by committee and committees remove specifics.

This is the part worth sitting with. For fifteen years, being found rewarded scale — more pages, more links, more domain authority. The new mechanism rewards clarity, and clarity does not have economies of scale.

Is there evidence, or is this just a new thing to sell?

Fair question, because this space filled up with packages very quickly.

There is some. The founding academic work — Aggarwal et al., GEO: Generative Engine Optimization, presented at ACM SIGKDD 2024 — tested roughly 10,000 queries and found that content changes lifted visibility in generated answers by up to around 40%. The strongest levers were not clever ones. They were adding statistics, citing sources, and including quotations.

That is the whole finding, and it is almost boringly aligned with good writing.

Now the honest caveats, because this newsletter doesn't run a stat without them. The Pew and Ahrefs figures measure Google, US-weighted, and assistant behaviour differs by engine and changes without notice. The GEO study measured specific engines at a point in time; the principle is sturdier than the percentage. Anyone selling you a guaranteed position in an AI answer is selling something nobody can currently deliver.

What's actually worth doing this week?

Five things, in order, and none of them require a new tool.

1. Look. Ask two different assistants what they say about your business — by name, by need, and by "who are the best options for X". Most founders have never done this. Write down what it got wrong, who it named instead of you, and where it says it got that. Don't fix anything yet; the list is worth more than the first fix.

2. Invert your best page. Answer in the first two lines, then the why, then the how. Subheads as questions a buyer would type — including who is this not for, which is the line competitors never write.

3. Put something quotable on it. One number with its source beside it. One plain definition. One stated limit. That's the GEO finding, applied.

4. Fix the sentence. What you do, who for, what changes — no adjectives. If a stranger can't repeat it back, a model can't either.

5. Measure it roughly, monthly. Ten fixed buyer questions, two assistants, three columns: named, not named, named-but-wrong. Ten minutes. It's a rough instrument and the answers vary run to run — but three months of it beats a strategy nobody is measuring.

The bit that makes this safe to act on

Notice what happens if the prediction is wrong.

Suppose assistants plateau, or Google restores clicks, or the whole thing shakes out differently. You'd be left with a clearer page, a sourced claim, a describable offer and a monthly habit of checking what the market says about you. All of which were worth having in 2019.

That's the test for any change like this: if the advice only pays off when the forecast is right, be careful. If it pays off either way, get on with it. This one pays either way, which is the only reason it's this week's theme rather than a warning.

The rest of the week takes each of the five apart — the decode on Monday, the visibility check Tuesday, the page shape Wednesday, the positioning Thursday and Friday, the monthly check on Saturday, and the argument itself on Sunday.

Start with looking. You cannot manage what you have never once gone and read.

For the audit that finds where your week actually leaks — free, 5-min read: The 10-Hour Audit.

Related Decodes

• Speed Is Cheap Now. Trust Is the Premium. — what stays scarce when output gets abundant.
• Say It Out Loud: The Solo Founder's Guide to AI Disclosure — last fortnight's hub: saying what AI does in your business.
• The One-Pager That Doubles Every AI Answer — the same clarity discipline, pointed inward.
• Make It Show Its Working — how to check a claim before you publish it as your own.

AI the Brief (@aithebrief) — AI advice you can act on and trust, for time-starved founders.

Not marketing or legal advice. Search and assistant behaviour changes without notice; the figures here are dated and sourced so you can re-check them yourself.

TL;DR: When a Google AI summary appears, click-through to normal results falls from 15% to 8%, and the summary's own citations are clicked in 1% of visits (Pew, 68,879 searches); Ahrefs measured a 58% CTR drop for top-ranking pages where an Overview appears. So the unit of being found is shifting from ranking in a list to being cited inside an answer — a contest that rewards being describable rather than being big, which favours a one-person business. The founding GEO study (Aggarwal et al., SIGKDD 2024) found content changes lifted visibility in AI answers by up to ~40%, with statistics, source citations and quotations the strongest levers. Five moves: ask two assistants what they say about you and write down what's wrong; invert your best page to answer first; add one sourced number, one plain definition, one stated limit; fix your one-sentence description so a stranger could repeat it; then run ten fixed buyer questions monthly and log named / not named / wrong. The caveat matters — this data is Google and US-weighted and engines change — but every one of the five is worth doing even if the forecast is wrong.