The short version
- An AI summary above the results changes what a ranking is worth, not how ranking works. The ten blue links are still there, just further down.
- Impact is highly uneven by query type. Simple informational queries are hit hardest; transactional, navigational and genuinely complex research queries much less.
- Being a source *inside* the summary is the new position zero. It is drawn from content that already ranks and is credible.
- Anyone quoting you a precise industry-wide click-through decline is extrapolating. The effect is real, well-attested by practitioners, and not uniformly quantified in public data.
The arrival of AI-generated summaries at the top of search results is the most visible change to the results page in a decade, and it has generated a proportionate amount of panic. The useful framing is narrower than the panic suggests: the mechanics of ranking have not changed, but the payoff curve has, and it has changed very differently depending on what people are searching for.
Before adjusting your strategy, it is worth working out how much of your traffic sits in the query types that are actually affected. For a lot of businesses the honest answer is "less than I feared".
Which queries are actually affected
Exposure ratings are our editorial judgement based on query intent, not measured click-through data. Segment your own Search Console queries by these categories to get a figure that is actually about you.
The practical exercise is straightforward and most teams have not done it. Export your top queries by impressions, tag each one against the rows above, and calculate what share of your impressions sit in the high-exposure bands. A site whose traffic is mostly navigational and transactional has a very different problem from a site living on definitional informational content — and they should not be following the same advice.
What still earns the click
- Detail a summary cannot carry. A full pricing table with limits and caveats does not compress into three sentences.
- First-party evidence. Your own data, screenshots, test results. Nothing else can produce them.
- Tools and interactivity. Calculators, checkers and configurators are not summarisable.
- Opinion and judgement. "Which of these should *I* buy given my situation" resists flattening.
- Currency. Genuinely current information on a fast-moving subject, with dates attached.
The common thread is that summarisable content gets summarised. If a page’s entire value is restating commonly available facts, the summary has replaced it and no amount of optimisation will restore the click. That page was always fragile; the summary just made it obvious.
How to get into the summary
There is no submission process and no markup that guarantees inclusion. What is observable is that sources drawn into summaries tend to be pages that already rank well, come from sites with recognised standing on the subject, and state the relevant fact clearly enough to be lifted. That points back at ordinary SEO plus the extractability habits covered in SEO vs AEO.
Two things are worth checking that people often miss. First, that you are not blocking the crawlers involved — review your robots.txt against what you actually intend. Second, that your structured data is valid rather than merely present; broken markup is functionally invisible, and the Rich Results Test will tell you in seconds.
On the click-through statistics circulating
You will see confident figures for how much AI Overviews reduce click-through rate. They vary enormously between studies because they measure different query mixes on different sites at different times. The effect is real and widely reported by practitioners; a single industry-wide percentage is not a thing that exists. Measure your own Search Console data before and after, segmented by query type. That number is about you and is the only one worth planning against.
What to change
- Segment your queries by the exposure table above and size your actual risk.
- For high-exposure pages, add something unsummarisable — data, a tool, a table, a judgement.
- For low-exposure pages, change nothing. A lot of wasted effort is currently going into fixing pages that were never at risk.
- Front-load answers so you are a candidate source rather than a page the summary talks over.
- Track clicks, not average position. Position became a much weaker proxy for traffic the day summaries appeared.