AI Search
Rankings Steady, Traffic Falling:
The AI Overview Effect
This is one of the most disorienting patterns in search right now. Your positions have not moved. Your rank tracker shows green. Your traffic is down twenty or thirty percent and nobody can explain why. The explanation is usually sitting directly above your listing.
When Google shows an AI Overview, pages not cited inside it lose a substantial share of clicks even while holding their ranking position. The response is not to rank harder but to become citable: fix rendering so AI crawlers can read your content, restructure pages so individual passages answer questions completely, and build the entity clarity that makes a system confident enough to name you.
What Is Actually Happening
An AI Overview appears above the traditional results and answers the query directly, citing a handful of sources. If the user gets what they needed from that summary, no click occurs at all. If they do click, it is disproportionately to a cited source.
The result is a genuinely new failure mode. Your page can hold position three, be perfectly optimised, load quickly, and still lose most of its traffic because it is not among the sources the summary drew from.
This is why the situation feels so unexplainable in conventional reporting. Every metric a rank tracker measures is unchanged. The thing that changed is not measured by rank trackers at all.
Confirming This Is Your Cause
Before assuming, verify. Several other causes produce similar patterns.
- Check Search Console for impressions versus clicks. The signature of this problem is stable or rising impressions with falling clicks. That combination means you are still appearing and being chosen less.
- Identify which queries lost the most clicks. Filter to your highest volume queries and compare periods.
- Search those queries manually. In an incognito window, run the affected queries. Does an AI Overview appear? Are you cited in it? Who is?
- Check the query type. If the affected queries are informational and question shaped, this cause is very likely. If they are transactional, look elsewhere.
If impressions fell alongside clicks, this is not your cause. That pattern indicates a genuine ranking or indexation problem instead.
Why Ranking Harder Does Not Solve It
The intuitive response is to push for a higher position, on the assumption that position one would restore the clicks. This works less well than expected, because citation and ranking are evaluated differently.
Research published in March 2026 analysing four million AI Overview citations found only around 38 percent of cited pages ranked in Google’s top ten for the triggering query. That figure had fallen from roughly 76 percent the previous July. Around a third of citations came from pages ranking beyond position 100, or not ranking for that query at all.
What this means practically is that citation selection is increasingly independent of ranking. Google appears to be running related sub-searches around a query and citing whatever answers each component best, regardless of whether that page ranks for the original phrasing.
The opportunity is that a page answering one specific sub-question extremely well can be cited even when your domain has no chance of outranking a national publisher for the main term.
What Actually Works in Response
1. Make sure crawlers can read you at all
Most AI crawlers do not execute JavaScript. If your content only renders client side, you are not a candidate for citation regardless of quality. Disable JavaScript in your browser and reload your key pages to see what a non-rendering crawler receives.
Also check your robots.txt for blocks on GPTBot, ClaudeBot, PerplexityBot, or Google-Extended, which are frequently inherited rather than chosen.
2. Restructure for passage extraction
Systems extract passages, not pages. Content built as a flowing argument that reaches its point in the conclusion provides very little extractable material.
- Frame headings as the exact question a user would ask
- Open each section with a direct answer in under fifty words
- Make each section comprehensible without the surrounding context
- Use tables and lists where the content is genuinely comparative or sequential
3. Answer the sub-questions, not just the head term
Since Overviews assemble answers from multiple sub-searches, coverage depth matters more than targeting a single high volume phrase. A cluster of pages covering selection, pricing, comparison, and implementation for your topic gives a system many more opportunities to cite you than one page targeting the main keyword.
4. Build entity clarity
Systems cite sources they can confidently attribute. Clear organisational schema, consistent representation across the web, and a factual About page all raise the confidence with which a system can name you as the source.
5. Shift what you measure
Rank tracking alone now measures an incomplete picture. Add a monthly manual check: run your twenty most important queries, record whether an Overview appears, and record who is cited. It is unglamorous and it gives you the baseline that rank tracking cannot.
What Not to Do
| Common response | Why it disappoints |
|---|---|
| Publishing more of the same content | Volume without extractability does not increase citation likelihood. |
| Chasing higher rankings on affected terms | Citation selection is increasingly independent of position. |
| Blocking Google-Extended defensively | A blunt instrument that usually reduces visibility rather than protecting value. |
| Abandoning informational content | Informational content still builds the topical depth that makes citation possible. |
A realistic expectation. Not every lost click is recoverable. When a query can be fully answered in two sentences, some of that traffic is genuinely gone regardless of what you do, because the user never needed your page.
The recoverable portion is the queries where the summary raises further questions, where a decision needs comparing, or where the user needs a provider rather than an explanation. Concentrate there rather than trying to win back everything.
Frequently Asked Questions
How much traffic do AI Overviews actually take?
Estimates vary by query type, but the pattern is consistent: informational queries are affected most, and pages not cited inside the Overview see materially reduced click through even at unchanged ranking positions. Commercial and transactional queries are less affected so far.
Should I stop investing in SEO because of this?
No, but the definition of success has to change. Strong organic rankings still matter and still correlate with citation. What has changed is that ranking alone no longer guarantees the click, so measurement needs to include AI citation presence alongside position.
Can I opt out of AI Overviews?
Not selectively in any practical way. Blocking Google-Extended affects AI training and some AI surfaces but is a blunt instrument that can reduce visibility rather than protect it. For most businesses the better strategy is becoming the cited source rather than trying to disappear.
Which queries are most affected?
Informational and question shaped queries, where an AI generated summary can genuinely resolve the intent. Queries requiring a transaction, a location decision, or a comparison of specific providers still drive clicks because the answer cannot be fully delivered in a summary.
How do I know if I am being cited?
Search your priority queries manually in an incognito window and read the Overview sources. Doing this monthly across ten to twenty core queries gives you a workable baseline without specialist tooling.
Losing Traffic Without Losing Rankings?
Get in touch and we will look at whether AI Overview citation is the cause, and what it would take to change it.
Talk to Our Team View Our Services