Google AI Overviews: what they are and what to do about them
How AI Overviews pick their sources, what they have done to click through rates, and the practical work that gets you included.
An AI Overview is the generated summary Google puts above the ordinary results. It answers the question directly and cites a handful of sources.
For anyone whose traffic depends on informational search, it is the most consequential change to results in a decade.
What it does to traffic
The honest summary, category by category.
Informational queries. Significant click loss. A query like "how long does X take" gets answered in the overview and a large share of searchers stop there. This is the traffic that is genuinely gone, and no optimisation recovers it.
Commercial queries. Smaller effect. Somebody comparing options still wants to reach a page, read reviews and see prices. The overview frames the choice rather than completing it.
Transactional queries. Least affected. You cannot buy inside a summary.
The planning consequence: content written purely to capture informational search volume has lost much of its value, and content that serves someone who is deciding or buying has held it. If your content strategy was built on top of funnel volume, that is the thing to revisit.
How sources get picked
Three observations from watching this closely.
Ranking is necessary but not sufficient. Cited pages are generally pages that rank for the query or a closely related one, but the top result is frequently not cited while a page from position seven is.
Passage level matters more than page level. The overview pulls a specific passage that answers a specific sub question. A page that answers the question cleanly in one place beats a better page that buries it.
Multiple sub questions, multiple sources. One overview often cites four pages, each answering a different part. You do not have to be the best page on the topic. You have to be the clearest page on one part of it.
The practical work
Answer the question in the first hundred words
Directly, plainly, in a self contained paragraph. Not after a preamble about why the topic matters.
If somebody could copy one paragraph from your page and it would fully answer the query, you have given the overview something to use. If your answer is distributed across six paragraphs, you have not.
Use question shaped headings
An H2 that reads like the question a person asks, followed immediately by the answer. This makes the passage boundaries obvious.
Keep the facts specific and current
Numbers, dates, prices, timeframes. Vague content does not get summarised because there is nothing to summarise. And put a visible last updated date on the page, because currency is weighted.
Cover the sub questions
Look at the People Also Ask box for your target query. Each of those is a passage opportunity on your page.
Mark it up
FAQ and article schema help machines parse structure. Modest effect, low cost, worth doing after the writing is right.
What to stop doing
Long preambles. The 300 words before the answer are now actively harmful, not merely tedious.
Writing for word count. A 3,000 word page that answers the question at word 1,200 loses to a 700 word page that answers it at word 40.
Chasing pure informational volume. That traffic is being absorbed. Write informational content to build authority and to be cited, not as a traffic plan.
Measuring it
Search Console will not separate AI Overview impressions cleanly. What you can do:
Watch click through rate by query. A query where impressions hold and clicks fall is likely being answered above you.
Check your main queries by hand, monthly. Search them, note whether an overview appears, and note which sources it cites. Fifteen minutes and it tells you more than the data will.
Track the position of queries where you are cited. Being cited from position eight is a strong signal that your passage structure works, and it is worth copying that page's shape elsewhere.
Where this fits with the assistants
AI Overviews, ChatGPT, Claude, Gemini and Perplexity are the same problem in four costumes. All of them read a small number of pages and write a summary naming a few options.
The work overlaps almost entirely: be in the sources that get read, make your facts liftable, answer questions in self contained passages, stay consistent everywhere.
Measure it the same way too. Twenty buyer questions, asked across the assistants, recording who is named and what is cited, repeated weekly with competitors in the same run. Our own Figo does that from $49 a month, but the method works with a browser and a spreadsheet, and doing it by hand once is the best way to understand what you are actually looking at.
Questions people ask
Do AI Overviews reduce clicks?
For informational queries, substantially. For commercial and transactional queries the effect is smaller, because people still want to reach a site to buy. Plan your content mix accordingly.
Can I opt out of AI Overviews?
There are markup options that limit how your content is used in snippets, and they tend to cost you visibility elsewhere too. For most sites opting out trades a small principle for real traffic.
How do I get cited in an AI Overview?
Rank well for the query, answer the specific question early and plainly on the page, and structure the content so a single passage fully answers it.
Are AI Overview citations the same as the top ranking pages?
Overlapping but not identical. Pages ranking outside the top three get cited regularly when they answer the specific sub question better.
See it on your own competitors
Figo checks their ads, pages, rankings, reviews and AI answers every week, then tells you what to do in plain words. Set up in two minutes.
Keep reading