AI Overviews and SEO: How to Keep Clicks When Google Answers
Google now answers a large share of searches on the results page itself. For some queries that has quietly removed most of the reason to click anything. For others it has changed almost nothing. Knowing which of your keywords sit in which group is the whole job — and most reporting hides the difference.
- What an AI Overview actually is
- What the studies say, and why they disagree
- Which queries lose clicks, and which do not
- How to measure it in your own data
- What still earns the click
- Getting cited inside the AI Overview
- What to stop doing
- What this changes for paid search
- Frequently asked questions
What an AI Overview actually is
An AI Overview is a generated summary that sits above the traditional results, assembled from several pages Google considers relevant, with a handful of source links attached. It is not a ranking position you win. It is a panel Google builds, and your page either gets used as a source or it does not.
Three things about it matter more than the technology behind it:
- It does not appear on every query. Triggering is uneven and it changes constantly. The same keyword can show an Overview one week and not the next.
- It pushes everything down. Even when your page still ranks first organically, first is now further down the screen than it used to be. On a phone that difference is severe.
- Being a cited source is not the same as ranking. Pages that are not in the top few organic positions get cited, and pages ranking first sometimes are not.
What the studies say, and why they disagree
Several independent studies have measured the click impact, and the honest summary is that they do not agree on the size of it.
An industry study reported roughly a 58% reduction in clicks on queries showing an AI Overview. Search Engine Journal covered a study putting the figure closer to 38%. Pew Research approached it from the user side, studying whether people click the links inside AI summaries at all. Different methodologies, different query sets, different date ranges — and a range wide enough that quoting any single number as “the” impact is misleading.
What every study does agree on is direction: where an AI Overview appears, clicks to organic results fall. The disagreement is about magnitude, and magnitude depends almost entirely on query type.
The useful takeaway: do not plan around somebody else’s percentage. The number that matters is what happened to your queries, and you can measure that yourself in Search Console.
Which queries lose clicks, and which do not
Queries that lose heavily
Anything with a short, factual, complete answer. Definitions. Conversions and calculations. “What is”, “how many”, “when does”. Opening hours. Simple how-to steps. If a competent answer fits in three sentences, the Overview gives those three sentences and the click has nowhere to go. Informational blog traffic built on this kind of query is the traffic that has already gone.
Queries that barely move
Anything where the searcher needs to do something on a specific site, compare real options, or check something only the destination knows. Buying a product. Booking an appointment. Getting a quote. Checking price or availability. Reading reviews of a named business. Logging in. An AI Overview can describe how a service works; it cannot quote your price or take the booking.
Queries in between
Commercial research — “best X for Y”, “X vs Y”, “how much does X cost”. Here the Overview often summarises the landscape and the searcher still clicks, because they do not trust a summary with a decision that costs money. These queries lose some clicks, not most.
| Query type | Example | Click impact |
|---|---|---|
| Definitional | what is a landing page | Severe |
| Simple factual | google ads character limit | Severe |
| Basic how-to | how to add a negative keyword | High |
| Commercial research | ppc management cost | Moderate |
| Comparison | agency vs freelancer | Moderate |
| Local intent | seo agency near me | Low |
| Transactional | book a website audit | Low |
| Branded | your company name | Minimal |
Sort your top pages into those rows before changing anything. Most sites discover that the traffic they lost was the traffic that never converted, and that the pages driving enquiries are largely untouched. That is a very different problem from “our SEO is broken”, and it needs a very different response.
How to measure it in your own data
You do not need a third-party study. Search Console holds the answer, and the method takes about twenty minutes.
Search Console shows the click side. For the answer side, the AI visibility scan checks whether the business is named or cited in those AI answers at all.
- Open the Performance report and compare a recent period against the same period last year. Use twelve-month comparisons rather than month over month, so seasonality does not do the talking.
- Add Impressions, Clicks, CTR and Position as columns, then look at Queries.
- Find the pattern that matters: queries where impressions held or rose while clicks fell, and average position stayed roughly the same. That combination is the signature of something appearing above you. You are still being shown; people are stopping earlier.
- Ignore queries where position dropped. That is a normal ranking problem with a normal ranking fix, and blaming AI for it will send you after the wrong solution.
- Group the affected queries using the table above. Then count how many of them ever produced an enquiry.
What still earns the click
When Google can produce a competent summary, a summary is no longer a product. What is still worth clicking for:
- Something only you have. Your own numbers, your own tests, a dataset nobody else published. A model cannot summarise a study that does not exist yet.
- Specificity Google will not commit to. Real prices, real timelines, real limits. An Overview will say “costs vary”. Your page can say what it costs and why.
- A decision, not a description. Comparison tables, honest trade-offs, and a clear recommendation with the conditions attached.
- Depth beyond three sentences. Edge cases, worked examples, what to do when the standard advice fails.
Getting cited inside the AI Overview
Citation is not guaranteed and cannot be bought, but the patterns are consistent enough to work with. The full service for this, across Google and the AI assistants, is described on the AI SEO agency page.
- Answer the question in the first two sentences under the heading. Then expand. Burying the answer under 400 words of preamble makes your page harder to lift from.
- Use question-shaped headings that match how people actually phrase things, with the answer immediately beneath.
- Keep facts atomic. Short, self-contained, checkable statements get used. Long conditional paragraphs do not.
- Structure honestly. Clean headings, real lists, real tables, and FAQ schema where a FAQ genuinely exists. Schema does not force citation, it just removes ambiguity.
What to stop doing
- Stop publishing definition posts. “What is X” content aimed at a general audience is the single worst investment in search right now.
- Stop measuring content by sessions alone. Sessions were always a proxy. Now they are a broken one. Measure enquiries, calls and revenue per page.
- Stop chasing keyword volume without checking the SERP. A 5,000-a-month keyword whose result page is an Overview plus four ads is not a 5,000-a-month opportunity.
What this changes for paid search
Shopping campaigns behave differently again, and the numbers there are easy to misread — that is covered separately in ads in AI Overviews and Shopping ad CTR.
Paid placement gets more valuable on commercial queries. When the organic result you used to own is pushed below a generated panel, the ad slots above it are the only reliable way to hold the top of the screen for a query someone is ready to act on. For high-intent commercial terms, the case for paid has strengthened rather than weakened.
Paid gets less attractive on informational queries. If the searcher’s question is fully answered before they scroll, paying for a click on that query buys somebody who has already been served. Those are the terms to trim.
The practical move is to look at organic and paid together for the same query set rather than in separate reports. Where organic clicks fell on a query that still produces enquiries, that is a candidate for paid coverage. Where organic clicks fell on a query that never produced anything, it is a candidate for a negative keyword, not a budget increase. Our approach to that is on the Google Ads management page, and the broader organic side sits under SEO services.
If you are auditing your own site after reading this, the common SEO mistakes list is the fastest way to find what is holding pages back, and the UK SEO cost guide covers what fixing it properly tends to cost.
Frequently asked questions
Do AI Overviews mean SEO is dead?
No, but the mix has changed. Informational traffic aimed at simple questions has fallen sharply. Commercial, local and transactional search is broadly intact. Content whose only value was explaining something is worth much less than it was; content that helps somebody decide is worth more.
How much traffic will I actually lose?
Published studies range from roughly 38% to 58% on affected queries, and they disagree because query mix differs. Your own figure depends on how much of your traffic sits on definitional queries. Measure it in Search Console rather than adopting a headline number.
Does being cited in an AI Overview send meaningful traffic?
Some, but far less than a normal organic click in the same position. Treat citation as visibility and credibility rather than as a traffic channel.
Want this checked on your own site?
If your impressions are steady but clicks are falling, that pattern is measurable and it is worth knowing whether the lost traffic was ever going to become customers. Send me your site and the pages you care about, and I will tell you which of your queries are affected and which are noise.
Send the details through the contact form, or message me on WhatsApp at +92 309 6405463.