AI Search and Click-Through Rates
Published September 2026 · RankNibbler analysis
Google can now answer a search using your content without sending anyone to your website. When it does, you keep the ranking and lose the click. In an analysis we ran of organic search performance, click-through rates fell by as much as 58%. Impressions held. Positions held. The traffic did not. This page explains the mechanism behind that, how to tell whether it is happening to your site, and what still earns a visit.
What changed was the machinery, not the mind
Picture Google the old way. You type one query, it returns ten blue links, and the best page wins. That machine is being replaced. What changed at Google was not its mind, it was its machinery.
The old machine was a router. It read your query, decided which pages were most likely to satisfy it, and handed you a list. Google made its money from the journey. You made yours from the destination. The arrangement only worked because you were the destination.
The new machine is a manufacturer. It reads your query, breaks it into related sub-queries, gathers what it needs from the pages that answer them, and assembles a reply on its own page. Google documents this as a query fan-out technique, issuing multiple related searches across subtopics to find supporting pages (Google Search Central: AI features and your website).
Your content still does the work. It just does it inside Google's answer rather than on your page. The ranking is no longer the product. The answer is, and Google builds the answer and keeps the customer.
Two things follow from that, and both are uncomfortable. This is not a penalty, so there is no quality problem to fix and no recovery to wait for. And it is not something you can opt into either: the same Google documentation says there is no special schema.org structured data you need to add to appear in these features. Being eligible is not the lever. Being worth leaving the results page for is.
What we measured
We ran an analysis of organic search performance and found click-through rates falling by as much as 58%.
That is the whole of the claim and it is deliberately narrow. "As much as" is a ceiling, not an average, and a ceiling tells you what is possible rather than what is likely. Before anyone acts on the number, they should know what it was measured against.
[Methodology to be added: sample, period and comparison.]
Google says total clicks are stable, and that can be true at the same time
In August 2025, Liz Reid, who runs Google Search, wrote that "total organic click volume from Google Search to websites has been relatively stable year-over-year", and that "average click quality has increased and we're actually sending slightly more quality clicks to websites than a year ago" (Google: AI in Search is driving more queries and higher quality clicks). Google published no figures alongside those sentences.
Assume it is accurate. It still does not tell you anything about your site. "Total" is the sum across the entire web. A total can sit perfectly flat while the distribution underneath it moves a very long way, because clicks lost by one kind of page are clicks gained by another. Nobody publishes an aggregate you can bank, and you do not run an aggregate. You run a handful of pages.
So the argument between Google and the various third-party studies is not the useful one. The useful question is narrower: on the queries that actually matter to your business, is the results page now doing the job your page used to do?
Why "rankings are fine" is true and useless
The most common version of this problem looks like nothing is wrong. Your reporting shows stable positions. Your impressions are flat or up. Someone checks the rank tracker, sees position three, and concludes SEO is working. Meanwhile the enquiries have stopped.
Impressions and position describe where you appear. Clicks describe whether appearing was worth anything. When the answer is assembled above you, those two things come apart, and every report built on position alone stops being informative on the day it matters most.
Click-through rate is the metric that catches it, because it is the ratio between the two. Here is what to look at in Google Search Console, and what each pattern would mean.
| What to check | The pattern | What it would mean |
|---|---|---|
| Clicks vs impressions, site-wide | Impressions flat or rising, clicks falling | You are being shown as often as before and chosen less often |
| Average position vs clicks | Position steady or improved, clicks falling | The ranking still exists, it is simply worth less than it was |
| CTR by query type | Question and definition queries falling faster than others | The queries that can be answered in a paragraph are being answered in a paragraph |
| CTR by page | A few pages carrying the whole decline | The problem is page-level, not site-level, and it is fixable page by page |
| Branded vs non-branded | Branded queries holding, non-branded falling | People who already know you still arrive. Discovery is what has been absorbed |
Do this over a long enough window to rule out seasonality, and compare like with like. A three-week view will tell you a story whether or not one exists.
What it means if you depend on organic traffic
If you run a small business, organic search was probably your cheapest channel, and cheap channels are the ones nobody audits. The risk is not that traffic falls overnight. It is that it drains slowly while every dashboard you own keeps reporting success.
The exposure is uneven, and it is worth being honest about which side of it your pages sit on.
- Most exposed: pages whose entire value was answering a question that can be answered in a paragraph. Definitions, "what is", simple how-to, opening hours, basic comparisons. If a summary of your page is as useful as your page, the summary will win.
- Least exposed: anything that cannot be summarised into a satisfying answer. A price for your specific job. A tool that does the calculation. A decision that needs your judgement. A booking. A result you produced that exists nowhere else.
The strategic move is to shift weight from the first group to the second, not to abandon the first. Content that answers simple questions still builds the topical coverage that makes you a candidate in the first place. It just cannot be the whole business any more. If you are starting from scratch on this, the small business SEO guide covers the basics, and what organic traffic is sets out where the rest of it comes from.
What actually works now
Five things. They are not new inventions. They are the parts of SEO that survive the machine change, because each of them produces something a generated answer cannot substitute for.
1. Information gain
Publish something only you have. Your own numbers, your own results, what happened when you tried it, what your customers actually asked. The test is simple: if the same article could have been written by someone who has never done the work, it adds nothing to the pile, and a model trained on the pile has no reason to cite it.
This is the opposite of the advice that dominated the last decade. Writing a better version of an article that already exists used to be a viable strategy, because better versions outranked worse ones and ranking produced clicks. Now the answer layer flattens ten versions of the same article into one paragraph. Being the eleventh version of a known answer is a losing position. Being the only source of a fact is not.
In practice this means running things and writing up what happened. A small piece of first-hand data beats a long article assembled from other people's articles, and it is cheaper to produce than most people assume.
2. Answer the question in the first 100 words
Put the direct answer at the top, in plain language, in a self-contained paragraph. No preamble, no throat-clearing about how important the topic is, no promise that you will explain it shortly.
Two reasons. The first is that an answer built from your page has to be liftable: a clean, complete paragraph near the top is far easier to quote than an answer scattered across a page. The second is that if someone does click, you have a few seconds to prove the click was worth it, and burying the answer is the quickest way to send them back.
This applies to headings too. Write headings that are the question a person would type, then answer that question directly underneath. The structure that makes a page easy to skim is the same structure that makes it easy to quote. More on this in the guide to writing SEO content.
3. Topical authority
Cover one subject completely rather than scattering across a dozen. A site with thirty pages that thoroughly covers one subject is a more credible source on that subject than a site with three hundred pages covering everything and nothing.
Completeness means the obvious pages and the awkward ones: the definitions, the how-to, the comparisons, the objections, the edge cases people run into at three in the morning. When a system is selecting sources on a topic, depth of coverage is one of the few signals that is genuinely hard to fake at scale.
Practically: pick the subject you have a real claim to, map everything a person needs to know about it, and work through the list. Do not start a second subject until the first is finished.
4. Internal linking
Coverage on its own is a pile of pages. Internal links are what turn it into a subject. They tell search engines which of your pages is the authoritative one on a topic, they push authority towards the pages you care about, and they give a visitor who did arrive somewhere else to go, which is the entire point when individual clicks are scarcer.
Link with descriptive anchor text, link from your strongest pages, and link to the new page from the old ones rather than only the other way round. The full method is in the internal linking guide, and you can see which of your pages are under-linked with the link analysis tool.
5. Third-party mentions
Everything on your own site says you are good. Of course it does, you wrote it. Someone else saying it is the half that counts, and it is the half most sites never work on.
Mentions, citations and links from other sites are the independent corroboration that you are a real entity with a real reputation. That mattered before and it matters more now, because when a system has to choose which of several sources to lean on, what the rest of the web says about you is the part you did not control. Being named on other people's pages also means you can be cited in answers you were never going to rank for.
This is slower and less comfortable than writing another article, which is exactly why it is worth doing. The link building guide covers the tactics that are still worth the time.
How to check your own site
An hour of work, in order.
- Pull twelve months of Search Console data. Compare clicks, impressions, CTR and average position year on year, not month on month. Sort by clicks lost. The pages at the top of that list are your investigation.
- Split queries by type. Branded against non-branded, question queries against everything else. If the decline is concentrated in question queries, you are looking at the answer layer rather than a ranking problem.
- Search your own top queries. Actually type them into Google and look at the results page. If the answer is complete before your listing appears, no amount of on-page work will bring that click back. That page needs a different job.
- Check your rankings really are holding. If position has slipped as well, this is an ordinary SEO problem and the usual fixes apply. Use the keyword rank tracker to separate the two.
- Audit the pages carrying the decline. Run them through the RankNibbler SEO checker or a full site audit. Check the answer is in the first 100 words, check the headings match how people ask, and check the page is linked to from your strongest related content.
- Decide what each page is for. For every page that has lost its click, either add something only you have, or accept that its job is now credibility and coverage rather than traffic, and stop measuring it on traffic.
All the tools above are free and need no account. The full set is on the free SEO tools page.
Related reading
If you want the longer view on how search is changing, optimising for Google AI Overviews covers the mechanics of the answer layer, the state of AI search in 2026 looks at why ranking and citation have come apart, and SEO in the age of AI covers what the discipline looks like from here.
Frequently asked questions
Is a falling click-through rate a Google penalty?
No. A penalty suppresses your rankings. This does the opposite: your rankings stay where they are and the clicks stop arriving. There is nothing to appeal and no recovery process, because nothing has been done to your site. The results page has simply taken over part of the job your page used to do.
Can I stop Google using my content in AI answers?
Google's documentation lists the preview controls nosnippet, data-nosnippet, max-snippet and noindex as ways to limit what is shown from your pages, while noting that content can still appear in AI features despite them (Google Search Central). Restricting previews also restricts your ordinary search listings, so for most sites it costs more than it saves.
My impressions are up and my clicks are down. What does that mean?
It means you are being shown as often as before and chosen less often. Check whether average position has held. If it has, the ranking is intact and its value has fallen, which is the pattern described on this page. If position has slipped too, you have a conventional ranking problem as well.
Should I stop publishing content that answers simple questions?
No, but stop expecting it to bring traffic. Those pages still do useful work: they build the topical coverage that makes you a credible source and they give your important pages something to be linked from. Judge them on coverage, and put your effort for traffic into pages that hold something only you have.
Last updated: September 2026