When you search on Google today, you often get the answer before you click anything: an AI summary, or a box with the facts. That's great for searchers and hard for anyone whose website depends on those clicks.
I saw it on a children's learning site I run. Four pages had shown up on Google's first or second results page for weeks and got almost no visitors. People saw them, and nobody came.
I chased the wrong problem first, which is the part worth telling. Search Console flagged around ninety pages as excluded from Google, and that looked like the emergency. It wasn't. Those are game pages we keep out of Google on purpose, so that the topic pages rank instead. The loud warning needed no action. The quiet number, good rankings and no clicks, was the one costing us.
The alarm is rarely the invoice
This keeps happening to me. Four hundred and sixty-six disapproved products turned out to be a feed we'd decided not to use. A four-day dip turned out to be noise before anyone went bug-hunting. Platform consoles are built to flag states, not to rank problems by cost, and the two orderings have almost nothing to do with each other. Triage before diagnosis, every time.
Once I'd set the noindex count aside, the actual defect was sitting in plain sight in the search results. Every one of those four pages had the same shape of snippet — a list of what the article covers, ending in an invitation to read the guide. And directly above them, for most of the queries involved, Google was already answering the question.
What a search result has to do now
The numbers on this have stopped being arguable. Pew Research tracked the browsing of 900 US adults and found that when an AI summary appears, people click a traditional result link on 8 per cent of visits — against 15 per cent when no summary is shown. SparkToro, working from Similarweb clickstream data, puts the share of US Google searches that end with no click at 68 per cent in the first four months of 2026, up from 45 per cent a decade earlier.
Those aren't ranking problems and no amount of technical SEO addresses them. They describe a change in the job a result does. A listing used to be a door: its purpose was to make you curious enough to walk through. Now it is frequently the last thing the reader will look at, sitting underneath a block of text that already told them what they came to find out.
A teaser has become a cost
The old snippet pattern assumed scarcity of information. You had the answer, the reader didn't, and the gap between those two facts was the click. Withholding slightly was the whole technique.
Run that pattern in an answer-first results page and you're asking the reader to pay attention for something they have already been given for free. Worse, you're asking it in the one position where you look least authoritative — below the thing that just answered them.
Matthew Dicks, writing about why audiences disengage in Storyworthy: "More often than not, you will find yourself two or three minutes into a story, unsure of where the story is going and why you should continue to listen."
Two or three minutes is generous. In a results page you have roughly two lines.
What we actually changed
New titles and descriptions for four posts across five languages, rewritten so that the snippet states the answer rather than advertising it. Where a query asks a factual question, the description now contains the fact. Every claim was already verified in the page's own fact register, so nothing new had to be researched and nothing could drift from the body copy — a build check enforces that.
This is the opposite of clickbait, and it's worth being precise about why. Google's own snippet guidance asks for descriptions that are "truly descriptive", unique per page, that summarise the whole page rather than one anecdote, and that include the key details a user needs to decide — dates, prices, specifics. Answer-first writing isn't a trick played on the algorithm; it's the documented request. The same goes for titles, which Google generates rather than accepts: it will rewrite a title that doesn't reflect the page, and it draws on your heading, your main visual title and the anchor text pointing at you to do so.
So the snippet gives the answer, and the page earns the visit by being where you go for everything that follows the answer. Which is a harder product requirement than a good title tag, and the correct one.
The uncomfortable part, and how to measure it honestly
If your snippet answers the question, some readers will take the answer and leave. They were never going to convert anyway. What you're competing for is the reader whose next question is the one your page is genuinely good at, and that reader now has a reason to believe you can answer it, because you just did.
Measuring it requires some discipline. The comparison that means anything is the click-through rate of those same pages before and after the change date, not traffic in aggregate — and in this case impressions on the site had already fallen sharply in the same fortnight for reasons that have nothing to do with snippets, most likely ordinary volatility on a young site. I deliberately kept the two questions apart rather than letting the rewrite take credit for a recovery it didn't cause. The re-check is diarised for two weeks out, on one metric, on those pages only.
If you publish at any scale
Then this is a library problem, not a copywriting problem. I've replaced ninety-seven posts per market with thirteen pages after they produced ten non-brand clicks between them, and spent a day where the real work was deciding what not to index. Thin pages whose entire value proposition was "we also have a page about this" were already a liability. In a results page that answers questions itself, they're invisible.
The lane that still pays is being the place a task gets finished — and being readable to the machines doing the summarising, which is its own emerging discipline now that shopping agents are being blocked by sites that don't realise they're turning customers away.
Rank is only permission to speak. The snippet is the sentence, and most of us are still writing trailers.
Sources & further reading
External
Pew Research Center — Google users are less likely to click on links when an AI summary appears
SparkToro — When Google stops sending clicks, what still works?
Google Search Central — Control your snippets in search results
Google Search Central — Control your title links in search results
Related posts
How to test whether a dip is a signal before you go hunting for a bug
97 blog posts per market, ten non-brand clicks
I shipped a magazine section in a day. The work was deciding what not to index.
AI shopping agents are getting blocked