How to Get Cited by Google AI Overviews in 2026

Google AI Overviews are the single largest GEO opportunity by query volume. They sit above the blue links, summarize an answer from 4-8 cited sources, and intercept clicks before the user ever scrolls. The good news: getting cited is more accessible than ranking. The bad news: it requires structural changes to your top pages, not just on-page tweaks.

This post is the operator manual for the AI Overview citation game in 2026. For the broader context, see the full GEO guide. For Perplexity-specific tactics, see the Perplexity playbook.

When do AI Overviews actually appear?

AI Overviews appear on 20.5% of keywords, per Ahrefs' analysis of 146 million SERPs (September 2025 desktop data). The trigger rate skews hard by intent: 21.4% of informational queries return one, versus 4.3% of commercial and 2.1% of transactional queries. Local searches trigger at just 7.9%. Question-formatted queries are the outlier — 57.9% of them return an AI Overview.

If your top 50 commercial queries are mostly head terms ("emergency plumber Austin"), AI Overview exposure is currently more modest than on informational terms. If you target conversational long-tail ("why does my water heater make a popping sound"), AI Overview presence is already substantial. Map your trigger rate per query before you prioritize the work.

How does Google pick which sources to cite?

Google's AI Overview retrieval runs on Vertex AI with an LLM Re-Ranker. The pipeline starts with the classic organic ranking, expands to ~50-100 candidate URLs via vector similarity, then re-ranks based on passage extractability and synthesis fit. The Re-Ranker scores each passage on relevance, declarativeness, and entity coverage. The top-scoring passages get pulled into the synthesis.

The implication: ranking matters as a candidate-set qualifier, but passage quality decides citation. Ahrefs analyzed 863,000 SERPs and 4 million AI Overview URLs and found only 38% of pages cited in AI Overviews also rank in Google's top 10 for that query — down from roughly 76% a year earlier (March 2026). Ahrefs attributes the drop partly to query fan-out, where Google issues several related sub-searches and pulls citations from all of them. Authority gets you into the pool. Extractability wins the citation.

What's the passage pattern that actually gets cited?

The inverted pyramid. Under every H2, the first paragraph is a 40-80 word answer capsule that fully answers the H2's question, names the relevant entity in the first sentence, and includes at least one specific number. Subsequent paragraphs add nuance, examples, and supporting facts. This mirrors how journalists write news ledes and how reference manuals open chapters.

• Lead sentence: complete answer in 15-25 words, with the named entity front-loaded.
• Supporting sentences: 2-3 sentences adding specific numbers, ranges, or named comparisons.
• Optional 3rd sentence: scope qualifier ("for service businesses," "on transactional queries," "as of 2026").
• Word count target: 40-80 words. Below 40 = too thin to extract. Above 80 = the AI truncates and you lose the citation.
• Avoid pronouns at the start. AI Overviews prefer self-contained passages that don't require external context.

Compare these two openers under an H2 reading "How much does dental SEO cost?". The bad one: "That's a great question. The answer depends on a lot of factors. Let's dive in." The good one (illustrative ranges, not a quote): "Dental SEO is typically billed as a monthly retainer in the low-thousands per month for single-location practices, rising for multi-location DSOs. Pricing scales with the number of city pages, the competitive density of the metro, and whether the program includes content production." The second one gets cited because it answers concretely; the first one doesn't.

Which schemas actually matter for AI Overviews?

Article + FAQPage compound on Q&A-style pages. Service or LocalBusiness on commercial pages. Organization with sameAs at the site level. BreadcrumbList for site-structure understanding. Everything else is noise. Specifically: Speakable is dead, HowTo rich results were removed in 2023 (but the schema still helps AI extraction), and Review schema is only useful when paired with Service or LocalBusiness.

Set expectations honestly, though. Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 against 4,000 controls: AI Overview citations moved −4.6%, AI Mode +2.4%, ChatGPT +2.2%. Adding schema boosted citations on no platform. The caveat Ahrefs names — those pages already had 100+ AI Overview citations — means this proves schema won't lift the already-visible, not that it's worthless everywhere. Treat the Article + FAQPage compound as hygiene: it keeps entity and editorial signals clean and supports rich results, but it won't rescue weak passages. See the schema deep-dive for full JSON-LD templates.

Can original data make you the cited source?

Often, yes. AI Overviews disproportionately pull passages that carry a specific, attributable fact — because when the synthesis needs that fact, you are the only place to attribute it to. Proprietary data you genuinely own (a real customer survey, an internal benchmark you can substantiate, a first-party case study with consented figures) becomes citation bait precisely because no competitor can replicate it. The non-negotiable condition is that the data is real: inventing a statistic to look quotable is both dishonest and, once detected, corrosive to the trust that makes a citation worth anything.

The practical move is to publish the data you actually have and cite everything you borrow — report true sample sizes, link every third-party statistic to its source, and let genuine first-party findings carry their own weight. Even the layout of the page helps: research on where AI Overviews source their citations describes a strong top-of-page bias, so put your most quotable, best-evidenced passages high on the page rather than burying them below the fold.

What's the right H2 strategy for AI Overviews?

Question-formatted H2s. Roughly 80% of your H2s should end in a question mark. AI Overviews trigger most heavily on question-form queries, and matching the question structure on-page boosts both retrieval and synthesis. The remaining 20% can be declarative (e.g., "The 4-stage citation pipeline") for the structural backbone of the post.

• Use "How," "What," "Why," "When," "Which," "Can," "Should," "Is" as H2 openers.
• Mirror the actual queries your buyers type. Pull them from GSC, Ahrefs Questions, and AlsoAsked.
• Avoid clever or vague H2s ("The truth about pricing"). The Re-Ranker can't match these to user intent.
• Keep H2s under 70 characters. Longer H2s get truncated in Search Console and look weak in featured snippets.

How do you track AI Overview presence?

Three tools and one manual check. SE Ranking and Ahrefs both have AI Overview presence tracking baked into their rank-tracking tools — you'll see a per-keyword flag when an AIO is triggered and which domains were cited. Otterly is the dedicated player; it tracks AIO citations across a custom prompt set with daily granularity. Plus a manual incognito search for your top 10 commercial queries once a month.

Build a monthly dashboard showing AIO trigger rate per tracked keyword, your citation share when triggered, and your competitors' citation share. Set the baseline in month 1 and measure delta against it. The point of the baseline is that citation share is the metric to move — track it honestly against your own starting point rather than against a borrowed benchmark, because trigger rates and competition differ sharply by vertical.

How do you tell if you're actually being cited?

Three checks, from crude to precise. First, run your target queries manually in an incognito window and read the AI Overview's source links — if your URL is in the card, you're cited. Second, watch Google Search Console for the pages you rewrote: a citation often shows up as an impression bump paired with a softer click-through rate, because the overview answers part of the query on the results page. Third, watch your referral data for small, qualified spikes from AI surfaces. None of these is decisive alone; together they tell you whether a rewrite actually landed.

Expect the impressions-up, clicks-flat pattern to be structural rather than a measurement error. Pew Research Center tracked the real browsing behavior of 900 US adults across 68,879 Google searches and found that when an AI summary appeared, users clicked a search result on just 8% of visits — versus 15% when no AI summary was present. They clicked a link inside the summary itself on only 1% of visits (March 2025 data). So a citation buys you visibility and brand exposure well before it buys you a click, and judging a GEO program purely on organic clicks will under-report it.

For scale, an AI Overview citation tracker (SE Ranking, Ahrefs, Otterly) flags per keyword whether an overview triggered and which domains it cited, so you can watch your citation share move over time instead of spot-checking by hand. Set your baseline the month you start and measure the delta against it — see how AI Overviews changed SEO for why the impressions-up, clicks-flat pattern is the signature of this shift.

What common mistakes keep you out of AI Overviews?

Most exclusions trace back to content, not schema. Thin, generic pages that restate what everyone else already says give the synthesis layer nothing distinctive to pull. Stale pages that haven't been refreshed in years lose to fresher, credible competitors — especially on anything time-sensitive. And pages with no supporting evidence — no data, no cited sources, no named specifics — read as low-authority and rarely enter the candidate pool in the first place.

• Thin or generic content — nothing an engine can't already assemble from a dozen other pages.
• Keyword-stuffed copy — the same classifier that scores helpfulness reads it as low quality.
• Neglected, un-refreshed pages — stale facts get out-competed by anything current and credible.
• No citations or specifics — passages with no numbers, sources, or named entities look unsupported and get skipped.
• The answer buried under a marketing intro — even a strong page loses if the definition isn't near the top.

What should you not do?

Don't stuff keywords. Don't write 12-bullet feature lists that all start the same way (it's an AI-tell and gets demoted). Don't bury the definition under 400 words of intro. Don't use Speakable schema (mostly dead). Don't add FAQ schema without matching the on-page H2 structure (Google can detect mismatch). Don't try to game by adding fake author bios — Google's E-E-A-T detection is much sharper than it was 2 years ago.

• Avoid generic openers ("In today's digital landscape...") — the synthesis layer skips these.
• Avoid 4-item bulleted lists with identical sentence rhythms — these read as AI-generated and get demoted.
• Don't gate definitions behind email forms — the crawler can't follow them and you exclude yourself from extraction.
• Don't use JS-rendered content for primary copy — Google's Vertex AI retrieval can read it but the latency hurts you.
• Don't republish thin AI-spun content under your domain. Google's helpful-content updates aggressively demote it.

What's the 30-day quick-win plan?

Pick your top 10 commercial pages by current organic traffic. Rewrite the first paragraph under each H2 to follow the inverted-pyramid pattern. Add Article + FAQPage compound schema. Refresh dateModified on the Article schema. Resubmit the pages in Google Search Console. Then track AIO presence weekly. Because these pages already have organic authority, clean rewrites give the synthesis layer something extractable to pull — and existing, already-indexed pages are where any presence lift tends to show up fastest.

If you want this run for you — including the audit, the rewrites, and the monthly tracking — book a strategy call and we'll quote the work. For the dental and HVAC playbook specifically, see dental SEO. For the full GEO program, our GEO service covers the whole stack.