What an AI Overview actually is
An AI Overview is a synthesised answer generated by Google's Gemini model and placed at the top of the search results page. It reads like a short paragraph, sometimes with bullets, sometimes with a step list. It cites between three and eight sources, shown as a stack of small clickable icons beside or beneath the answer.
The rest of the search page still exists. The map pack, the blue links, the related searches, all of it slides down below the Overview. On mobile, the user often has to scroll past the full Overview before a single organic result is visible. The page is not gone. It is just further away.
Why this is different from featured snippets
A featured snippet pulled one source and quoted it. If you ranked in position zero, you earned the whole box. AI Overviews pull from several pages at once and rewrite the content into a new paragraph. No single page is quoted. Several are credited.
The site that ranks number one in the classic sense is not always among the cited sources. We see it often. A small site with clean schema and a direct answer paragraph gets cited above a much larger competitor that outranks it in the blue links. Position still matters, but clarity, structure and direct answers matter more for who the Overview quotes.
The three fixes that move citations
Run these in order. Each one can ship in an afternoon.
1. Direct-answer paragraph in the first 100 words
Write a forty to sixty word paragraph near the top of the page that answers the primary question the page is about. Declarative sentences. No throat clearing. No "in today's fast-paced world". The paragraph should stand alone and make sense out of context, because that is exactly how the Overview will use it.
2. JSON-LD schema that matches the visible page
Pick the schema type that fits. FAQPage for a questions page. HowTo for a step guide. ProfessionalService or LocalBusiness for a service business. Put the JSON-LD in the head of the page. Every field must reflect current, accurate data that is also visible on the page. Wrong prices, old opening hours, or phantom FAQ entries will hurt you.
3. Clean page structure
One H1. A clear H2 for each sub-question. Short paragraphs underneath. The Overview model lifts sentences out of structured pages because they are easier to extract and rewrite. A page that is one long wall of text gives it nothing to grip. A page with H2s that mirror how a customer phrases the question gives it a map.
What about the click-through
The honest data. AI Overviews reduce blue-link clicks on simple informational queries, where the answer is short and the user is done. On commercial and local queries, the picture is mixed. Cited sources inside the Overview still receive meaningful click-through, sometimes higher than their old organic position delivered.
The sites that take the real hit are the uncited ones. If the Overview answers the question and your site is not among the three to eight citations, you effectively drop off the page. The strategy is not to avoid AI Overviews. It is to be the cited source inside them.
What to skip
Three traps worth naming.
- Do not write content aimed specifically at pleasing AI Overviews. Google has said the system is dynamic and the signals shift. Write for a human reader who wants a clear answer, add structure, and the Overview will follow.
- Do not stuff the page with FAQ schema that does not match the visible content. Google has been explicit about penalising this, and recent guidance removed FAQ rich-result eligibility for most sites as a result.
- Do not chase every "people also ask" that appears under your query. Pick the two or three that match what your customers actually want, answer them inline, and move on.
How to monitor
Weekly cadence, not daily. Open an incognito tab, run eight to twelve of your target queries by hand, and screenshot the AI Overview for each. Log whether your site appears in the citation stack, and if not, which three to five sites did. That log is the brief for your next round of fixes.
For a faster version, the free checker gives you the exact prompts for your business across ChatGPT, Perplexity and Google. Apply the three-prompt consistency rule. If the same business is cited across three different phrasings of the same intent, it is a real pattern. One hit is noise. Three hits is a signal you can trust.
Start here
If you want a ranked backlog with the exact copy and schema rewritten for your pages, the $197 audit does that work for you with a 48 hour turnaround. If you want to ship the schema changes yourself, the free schema pack gives you paste-ready JSON-LD for the four types that matter.
Happy to answer anything, Bob.