AI Overviews in Google: what changes for website traffic and how to write for it

AI Overviews is the block with a generated answer above the regular Google results, and on informational queries it takes a noticeable share of clicks away from websites: people read the summary and go no further. Commercial and local queries suffer far less, because there the user still has to reach a product, a price or an address. Below we look at where exactly the Google AI Overview appears, how CTR changes by query type, which pages Google cites in the block, and how to write content so that it gets cited instead of simply losing traffic.

What AI Overviews are and where they show up

AI Overviews is an answer that the Gemini model assembles from several pages and shows above the organic results. To the right of the text or below it sit source cards linking to the sites the facts came from. The block launched in the US in May 2024, in the autumn of the same year Google rolled it out to more than a hundred countries, Ukraine included, and it keeps adding languages. As of publication, overviews are most visible on English queries, while in Ukrainian they appear on a growing share of informational questions.

Whether the block appears depends on the intent behind the query:

• "How", "what is", "why", "how much" questions, comparisons and instructions. Here an AI Overview is almost guaranteed.
• Queries with a clear purchase ("buy nike sneakers kyiv") or a brand name. The overview is rare or shows a product carousel instead.
• Local queries with a map ("dentist near me"). The block is either absent or sits below the map.
• Health, finance and legal topics. Google shows overviews more cautiously, often with extra disclaimers.

It is worth separating AI Overviews from AI Mode. The first is built into the regular results page, the second is a separate chat tab the user opens deliberately. For website traffic the first format matters more right now, because everyone sees it.

How CTR changes: an honest picture by query type

"Position zero" used to mean the featured snippet: one site got a block above the results and, with it, the largest share of clicks. An AI Overview works differently. It answers the question right on the results page, and even the site that gets cited receives fewer visits than first place used to bring without an overview. Independent industry studies from 2024 and 2025 give different numbers, but agree on the order of magnitude: on informational queries with an AI Overview, organic CTR for top positions drops by roughly a third, and in places by half. What we see on Query projects matches that: blogs with reference articles slide, while service and catalogue pages hold.

One important detail: impressions often do not fall but grow, because a link inside the overview counts as an impression too. That is why reports show "traffic down, impressions up", and this very gap is the signature of AI Overviews at work.

Which sources Google cites in AI Overviews

Google does not publish its source selection algorithm, but after analysing hundreds of overviews across niches the patterns are consistent. The cited page is not necessarily the top organic result. Pages from positions 5, 8 or 12 often make it into the block if they contain something the model can conveniently lift.

• A clear answer in one paragraph. The model looks for a passage that answers the question without a preamble. A 40-word paragraph with a direct answer gets cited more often than a sprawling section.
• Structure. An H2 phrased as a question, the answer beneath it, then details as a list or a table. Such a page is easy to break into fragments.
• Visible expertise. A named author with experience, links to primary sources, your own numbers instead of retold ones. In 2025 Google explicitly names experience and trustworthiness as criteria for AI Overviews.
• Freshness. For topics where information changes (prices, rules, service versions), the overview picks updated pages with a real modification date.
• Technical accessibility. The page is indexed, serves its content without JavaScript tricks, loads fast. Anything that does not make it into the index does not make it into the overview either.

In practice this means that working for AI Overviews overlaps with plain good SEO by about 80 percent. The difference is in emphasis: more attention to how the first sentences are phrased and to splitting the text into self-contained chunks. We covered how GEO and AEO differ from classic SEO in a separate article.

How to write content that gets cited

The working format we use for reference pages and articles has five elements. It does not guarantee a citation, but it raises the odds considerably, and, importantly, it does not hurt regular rankings.

1. A direct answer in the first two sentences. Right after the heading: the claim plus a specific (a number, a timeframe, a tool name). Preambles like "in this article we will look at" go entirely.
2. H2 headings phrased as questions. "How much does…", "Do you need…", "What to do if…". Every section opens with a short answer, and the details follow.
3. Tables for comparisons. Options, prices, conditions, before/after. The model reads tables well and often reproduces them in the block with a link to the source.
4. Lists for steps and criteria. Numbered for sequences, bulleted for enumerations. One thought per item, no paragraphs inside an item.
5. An FAQ at the end. Three to six short questions with two- or three-sentence answers, with FAQPage markup where appropriate. This is the easiest fragment for the model to use, and it is useful to the reader at the same time.

What not to do: do not inflate the text for length, do not bury the answer at the end "so they read to the bottom", do not repeat the same answer in several paragraphs with different wording. The overview takes the shortest clear fragment, and if your page does not have one, it will take it from a neighbour.

What a business should do when informational traffic drops

The worst reaction to AI Overviews is to write even more generic articles in the hope that volume wins. The better approach is to redistribute effort.

• Bet on commercial and local queries. These are the ones that convert, and the ones least affected. Service pages, categories, products, city pages: here SEO works as it always did, and at Query we are shifting the main focus exactly there.
• Your own expertise instead of a retelling. The overview paraphrases common knowledge easily, but it does not have your data: measurements, cases, internal statistics, photos from production. Such content gets cited and, more importantly, gets visited.
• Brand. Branded queries do not depend on overviews. The more often people search for you by name, the less you depend on changes in the results page. It is slow work, but it pays off on every algorithm shift.
• Informational content as part of the funnel, not as a traffic source. Articles should lead to a service or a product, collect subscriptions, show expertise. Judging them by search visits alone is no longer accurate.
• Presence in the sources that get cited. Overviews often lean on industry directories, marketplace reviews, forums. Mentions there work for brand citations even without a visit to your site. This is already GEO territory, which we wrote about separately.

How to track the impact in Search Console

There is no separate "AI Overviews" filter in Search Console: clicks and impressions from the block are counted together with the regular results under the "Web" search type. So the impact has to be inferred indirectly.

1. In the Performance report, compare two 3-month periods: before overviews appeared in your niche and after. Look not at total traffic but at "impressions up, CTR down" pairs for individual queries.
2. Split queries into groups with a regular expression: informational ("how", "what is", "why") separately from commercial ("buy", "price", "order"). The gap in CTR dynamics between the groups is the AI Overviews effect.
3. Check 20-30 key queries by hand in an incognito window. Record where an overview is present and whether you are cited. Repeat monthly: the set of sources changes often.
4. Add "shown in an overview" to your reporting as a separate metric. Rank tracking tools (Serpstat, SE Ranking, Ahrefs) can already show the presence of an AI Overview and your citation in it.

For clients on SEO services we now include this slice in the monthly report by default: the share of queries with an overview, the share of citations and CTR dynamics by group.

Where to start

1. Isolate informational queries in Search Console and compare CTR across the last two 3-month periods.
2. Rewrite the opening paragraphs of your 10-20 most important articles: direct answer first, details after.
3. Add tables and FAQs wherever there is something to compare or a question to answer briefly.
4. Move part of the budget to commercial and local pages that convert.
5. Once a month, check by hand whether you are cited in overviews for your key queries.

AI Overviews do not cancel SEO, they change what traffic pays for. Reference content now has to be written as a source for citation and as part of the funnel, while growth should be sought where the user still has to reach your site.