x
The click stopped being the default outcome of a search.
Pew Research Center tracked 68,879 Google searches made by 900 US adults on its KnowledgePanel. Where an AI-generated summary appeared at the top of the page, people went on to click a traditional search result in 8% of visits. Where no summary appeared, 15%.
Two more numbers from the same study, both more revealing than the headline. Users clicked one of the source links inside the summary in 1% of visits — so the citation footer, added partly in response to publisher complaints, is in practice almost never used as a door. And sessions ended entirely after a summary page 26% of the time, against 16% without one.
Ahrefs measured the publisher side of the same shift across 300,000 keywords in Search Console data: pages ranking first now see a 58% lower click-through rate where an AI Overview is present, up from 34.5% eight months earlier. Position two lost about half its clicks. Even position ten dropped by nearly a fifth.
None of this was announced. There was no policy change to read, no email. It arrived as a slow change in other people's habits.
Less than the panic suggests, and differently than you'd think.
If your search traffic was never the thing bringing you work — and for most one-person businesses it wasn't — then a drop in clicks is not the story. Referrals, repeat clients and word of mouth are unaffected by any of this.
What does matter is the question underneath. When someone asks an assistant "who does bookkeeping for small creative agencies" or "what should I look for in a fractional finance person", something gets said back. A few names, or a description of what to look for, assembled from sources the asker never sees a list of.
You are either in that sentence or you are not, and there is no way to find out except to ask.
That's the real change, and it isn't about traffic at all. It's that a recommendation layer has appeared between you and people who don't know you yet, and unlike the ten blue links, it doesn't show its whole hand.
The old failure mode was ranking eleventh. The new one is being described as a category rather than a name — "look for someone who specialises in..." — which reads like advice and functions like being left out.
Three things, and one thing not to.
Take the direction, not the decimal. The 8%, the 58%, the 1% — those measure Google, US-weighted, at specific moments. The direction they all point in has been consistent for two years and is the only part safe to plan around. Any strategy that needs the number to stay at 58% is not a strategy.
Take the shift in unit. Ranking was a competition for a slot in a list. Citation is a competition to be quotable. Those reward different work, and the second one — plain claims, sourced numbers, a describable offer — is work a solo operator can do in an afternoon and a committee genuinely struggles with. The one-pager that doubles every AI answer was about briefing a model well. This is the same clarity discipline pointed at your own public pages.
Take it as a prompt to look. Most founders have never once asked an assistant what it says about their business. That's a fifteen-minute job and it's tomorrow's post.
Don't take it as a reason to buy anything. The market for "AI search optimisation" packages got very crowded very fast, and nobody can currently guarantee a position in a generated answer — the systems are non-deterministic and change without notice. If a pitch includes a guarantee, that's the tell.
Partly, and the overlap is the useful part.
The founding academic work on this — Aggarwal et al., GEO: Generative Engine Optimization, ACM SIGKDD 2024 — tested around 10,000 queries and found content changes could lift visibility in generated answers by up to roughly 40%. The strongest levers were adding statistics, citing sources, and including quotations.
Notice what's absent from that list. No keyword density, no link building, no technical trickery. The things that worked were the things that make a page more useful to a careful human reader, which is a rare and welcome case of the incentive pointing the right way.
Where it genuinely differs from SEO is the feedback loop. There is no rank tracker for this. Nothing tells you when a model stops mentioning you. That absence is why a rough monthly check matters more here than a precise one ever did in search — a crude instrument you actually run beats a good one that doesn't exist.
And it connects back to something this newsletter has argued since speed got cheap: when output becomes abundant, the scarce thing is being trusted and legible. Being the sentence rather than the eleventh link is that argument arriving in your marketing.
The week ahead takes it apart one move at a time. Today is only the news: the click was the unit, and it isn't any more.
For the audit that finds where your week actually leaks — free, 5-min read: The 10-Hour Audit.
Related Decodes
AI the Brief (@aithebrief) — AI advice you can act on and trust, for time-starved founders.
Not marketing advice. Figures are dated and sourced so you can re-check them; assistant behaviour changes without notice.
TL;DR: Pew tracked 68,879 Google searches and found that when an AI summary appears, people click a traditional result 8% of the time versus 15% without one — and they click a source link inside the summary in just 1% of visits, with 26% ending the session entirely. Ahrefs measured a 58% CTR drop for top-ranking pages where an AI Overview appears, across 300,000 keywords. For a one-person business the point is not lost traffic, which was probably never the main channel, but that a recommendation layer now sits between you and people who don't know you yet, and it doesn't show its whole hand. The new failure mode is being described as a category instead of named. Take the direction rather than the decimal, treat citation rather than ranking as the unit, and go and look at what assistants say about you — but don't buy a guarantee, because non-deterministic systems can't offer one.
