What Are AI Overviews? How They Work

What Are Google's AI Overviews, and Why Do They Matter?

AI Overviews are generated summaries that Google can display above or among its conventional search results. They combine information retrieved from the web with Google's generative AI systems, then provide links that allow searchers to examine supporting sources. Understanding how to get in Google's AI Overviews starts with understanding what they actually are and how they differ from familiar search features.

They are not simply longer featured snippets. A featured snippet normally extracts a short passage from one page. An AI Overview can investigate several related questions, synthesise information from multiple sources and change its response according to the wording, location, device and context of the search.

For businesses, this creates a second layer of organic visibility. A page may still rank as a conventional result, but it may also be cited, linked or mentioned within the generated response. That opportunity comes with a challenge: if the summary satisfies the searcher's immediate need, fewer people may click through to any website.

Key takeaways

AI Overviews are generated summaries combining web information with Google's generative AI, distinct from featured snippets.
Outcomes should be measured separately: citation, mention and referral are not the same thing.
Google interprets the query, may fan out to related searches, retrieves from the Search index, then assembles a generated answer.
AI Overviews are most likely for how/why questions, comparisons, multi-step tasks and conversational searches — not navigational queries.
AI Mode and AI Overviews share foundations but are not identical reporting surfaces.
Generated answers are not infallible; risk rises for fast-changing, personal, disputed or YMYL topics.
There is no special submission process, AI Overview schema type or llms.txt requirement — good SEO fundamentals apply.
Businesses should track presence, visibility, engagement and commercial outcome as four separate measurement layers.

What does an AI Overview look like?

The presentation varies. A typical AI Overview contains a generated answer, links or citation cards, and options to explore the subject further. Some answers include lists, steps, comparisons, images or product information. Searchers may also be able to ask follow-up questions.

Google continues to test how its AI features appear, so there is no permanent layout that publishers can optimise around. The useful distinction is between three outcomes:

Citation: a page is shown as a source supporting part of the answer.
Mention: a brand, organisation or product is named without a prominent link.
Referral: a searcher follows a link from the AI experience to the website.

These outcomes should be measured separately. A citation is evidence of visibility, not proof that the page received traffic or generated business.

Three AI Overview outcomes
Outcome	What it means
Citation	A page is shown as a source supporting part of the answer.
Mention	A brand, organisation or product is named without a prominent link.
Referral	A searcher follows a link from the AI experience to the website.

How do AI Overviews work?

Google does not publish every component or weighting used to create an AI Overview. Its official guidance does explain the broad process. This is also central to how Google chooses AI Overview sources.

The search is interpreted

Google first determines the likely meaning and needs behind the query. A conversational search such as "what happens to my workplace pension if I change jobs?" contains several implied questions: whether the pension remains invested, whether contributions stop, whether it can be transferred and whether any guarantees could be lost.

Information is retrieved from the Search index

Pages must be accessible to Google and eligible to appear in Search. Crawling, indexing, canonicalisation and snippet controls therefore remain important. There is no separate AI Overview submission process.

A generated answer is assembled

The system selects information it considers useful, creates a response and associates links with relevant parts of that response. Different claims can be supported by different sources.

The generated answer is not a permanent copy of one page. It can change as the query changes, new information is indexed or Google's systems select different evidence.

Which searches trigger AI Overviews?

Google does not show an AI Overview for every search. They are most useful where a question benefits from explanation or synthesis, including:

how and why questions;
comparisons involving several factors;
multi-step tasks;
decisions with conditions or exceptions;
longer conversational searches; and
topics where supporting sources help the searcher verify the answer.

A navigational query such as "HMRC login" already has a clear destination. An exploratory query such as "should a sole trader register for VAT before reaching the threshold?" requires context, trade-offs and current rules. The latter is more suited to a generated explanation.

Triggering is not fixed. The same query may produce an AI Overview on one date or device and not another. Any research dataset should therefore record country, language, device and observation date.

Our guide to finding keywords that trigger AI Overviews explains how to build a controlled query set.

How are AI Overviews different from AI Mode?

AI Overviews appear within the familiar Google results experience and offer an initial generated summary. AI Mode is a more conversational search experience designed for continued exploration and follow-up questions.

The two experiences share important foundations, including Google's Search index and ranking and quality systems. However, a business should not treat them as identical reporting surfaces or assume that a citation in one guarantees visibility in the other.

The practical lesson is to build content that answers the first question clearly while supporting the deeper questions a searcher may ask next.

Are AI Overviews accurate?

They can be useful, but generated answers are not infallible. A system may misunderstand a source, combine claims that require different qualifications or present information that has changed.

Risk rises when:

the subject changes quickly;
the correct answer depends on personal circumstances;
sources disagree;
an important exception is buried;
the page does not state its date or jurisdiction; or
the topic concerns health, finance, safety or law.

Publishers should make claims easy to verify. Date volatile information, cite primary sources, explain limitations and distinguish general information from personal advice.

For example, a page explaining inheritance tax should name the country, tax year and relevant thresholds. "Inheritance tax is charged at 40%" is incomplete without explaining allowances, exemptions and the portion of an estate to which the rate may apply.

How do AI Overviews affect website traffic?

The effect is not uniform. AI Overviews can reduce clicks when the generated answer resolves a simple informational need. Research based on observed browsing has associated AI summaries with fewer clicks to conventional results, while clicks to sources inside the summaries are relatively rare.

That does not make visibility worthless. A useful citation can:

introduce a brand during early research;
reinforce perceived expertise;
produce a highly informed visit;
influence a later branded search; or
place a smaller specialist beside a well-known publisher.

Commercial value depends on the query. A citation for "what is compound interest?" may bring little immediate revenue. A citation for "questions to ask before transferring a final salary pension" could influence a consequential decision, even if the eventual enquiry occurs later.

Read how AI Overviews affect SEO for the full traffic, measurement and strategy analysis.

Can websites optimise for AI Overviews?

Yes, but not through a special submission file or guaranteed markup. Google says its generative features are rooted in core Search ranking and quality systems. Good AI-search optimisation therefore starts with good SEO:

make important pages crawlable and indexable;
answer a defined search intent;
provide accurate, original and useful information;
demonstrate relevant experience and expertise;
structure the answer clearly;
support material claims with evidence;
maintain consistent entity information; and
keep time-sensitive content current.

Google also states that publishers do not need to create arbitrary "AI chunks", add an AI Overview schema type or publish an llms.txt file to qualify.

Our practical guide explains how to rank in AI Overviews without relying on unproven shortcuts. It's also worth understanding our approach to AI Overview optimisation, and why some businesses choose to work with a dedicated AI Overview optimisation agency rather than tackle it alone.

What should a business measure?

Do not replace one vanity metric with another. Track four layers:

Presence: which priority searches produce an AI Overview?
Visibility: is the brand mentioned or a page cited?
Engagement: does the visibility produce visits or branded demand?
Commercial outcome: do those journeys contribute to enquiries, sales or qualified actions?

Google launched a dedicated generative-AI performance view in Search Console in 2026. Use it alongside controlled manual observations, analytics and conversion data. Reporting should state what a metric proves and what it does not.

Sources

Frequently asked questions

Are AI Overviews the same as featured snippets?+

No. Featured snippets usually extract content from a particular result. AI Overviews can generate a broader response using several searches and sources.

Does every search have an AI Overview?+

No. Google decides when a generated answer may be helpful. Availability and triggering can vary by query, market, device and time.

Can you pay to appear in an AI Overview?+

Organic citations are not purchased placements. Advertising can appear around AI search experiences, but that is different from being selected as an organic supporting source.

Is there special schema for AI Overviews?+

No dedicated AI Overview schema type exists. Use supported structured data that accurately describes visible page content and is relevant to an eligible search feature.

Do AI Overviews replace SEO?+

No. They change how information is presented and how visibility should be measured, but crawling, indexing, relevance, quality and trust remain fundamental.

What's the difference between a citation and a referral?+

A citation means a page is shown as a source supporting part of the AI Overview's answer. A referral means a searcher actually followed a link from the AI experience to the website. A citation does not guarantee a referral.

Can a page rank well but never be cited in an AI Overview?+

Yes. Query fan-out means an AI Overview may draw on pages answering supporting subtopics rather than only the page ranking first for the exact query typed.

Why might an AI Overview get something wrong?+

The system may misunderstand a source, combine claims that need different qualifications, or surface information that has since changed. Risk is higher for fast-moving, personal or disputed topics.

AI Overviews add a new layer to search

AI Overviews are best understood as generated search answers grounded in Google's wider Search systems. They can help users explore complex questions quickly, but they also redistribute attention away from the conventional results page.

Businesses should respond by becoming more useful and more measurable, not by producing content shaped around speculative AI tricks. Build pages that resolve real questions, make evidence visible and give the right reader a reason to continue beyond the summary.

Undeniable Growth helps organisations connect SEO, content strategy and AI-search visibility to meaningful commercial outcomes. If you'd like help turning that into a plan for your website, get in touch.