Key takeaways
- Only 38% of AI Overview citations come from pages that rank in the top 10, so traditional ranking alone won't get you cited.
- Google's AI Overview selection rotates 45.5% of the time per refresh -- meaning you need to monitor citations continuously, not quarterly.
- Pages that combine text, images, and short-form video see a 317% higher selection rate for AI Overviews than text-only pages.
- The 8 on-page signals below are the clearest predictors of citation -- most can be audited and improved without a developer.
- Tools like Promptwatch can show you exactly which prompts you're being cited for (and which competitors are winning where you're not).
Here's something that should make you rethink your SEO strategy: a page ranking #1 for a keyword is now less likely to appear in Google's AI Overview than a page ranking #7 that answers the question more directly.
That's not a hypothetical. Seer Interactive's data shows that only 38% of AI Overview citations come from top-10-ranked pages. The rest come from pages Google's AI considers more useful for the specific query -- regardless of their position.
When Google AI Overviews appear, users click traditional results only 8% of the time, down from 15% without them (Pew Research, 2025). But brands cited inside an AI Overview earn approximately 35% more organic clicks and 91% more paid clicks than uncited competitors on the same page.
So the question isn't "how do I rank higher?" anymore. It's "what does my page need to look like for Google's AI to trust it enough to cite it?"
These 8 on-page signals are your answer.
Signal 1: Answer-first structure in the opening paragraph
Google's AI doesn't read your whole page before deciding whether to cite it. It looks for a direct, usable answer near the top -- typically within the first 20 to 30 words of a section.
This is the single biggest structural change most content teams need to make. The old SEO habit was to build up context before delivering the answer: introduce the topic, explain why it matters, then finally get to the point. AI Overview selection inverts that completely.
If someone searches "how long does it take to rank on Google," your page should open that section with something like: "Most new pages take 3 to 6 months to rank on Google, depending on domain authority and competition." Then explain why. Not the other way around.
Practically, this means auditing your H2 and H3 sections and checking whether the first sentence after each heading actually answers the implied question. If it doesn't, rewrite it so it does. This one change alone can move the needle on citation rates.
Signal 2: E-E-A-T signals baked into the page
Google uses Experience, Expertise, Authoritativeness, and Trustworthiness (E-E-A-T) as a filter against generic content -- and this matters more for AI Overviews than it does for traditional rankings.
The reason is straightforward: AI Overviews are Google's public face. If the AI cites a page that turns out to be wrong or misleading, that's a trust problem for Google, not just for the publisher. So the algorithm is conservative. It leans toward pages that signal real human expertise.
What does that look like on-page?
- Author bylines with credentials or relevant experience (not just a name)
- First-person observations or specific data points that couldn't come from a generic AI writing tool
- Citations to primary sources (studies, official documentation, named experts)
- "Last updated" dates that show the content is maintained
One thing worth noting: E-E-A-T isn't a checkbox. A page that has an author bio but reads like it was written by someone who's never actually done the thing they're describing won't fool the system. The signals need to be consistent throughout the content.
Signal 3: Semantic completeness and topical depth
AI models don't just look for the answer to the specific query. They look for whether your page covers the topic thoroughly enough to be considered a reliable source.
This is what "topical authority" actually means in practice. A page about "best running shoes for flat feet" that only covers shoe recommendations -- without touching pronation, arch support mechanics, or fitting considerations -- is less likely to be cited than a page that covers all of those angles, even if the shorter page ranks higher.
The practical implication: before publishing or updating a page, map out the full set of questions a user might have about that topic. Not just the primary query, but the follow-up questions, the related concerns, the "what about X?" questions. If your page answers most of them, it reads as authoritative to both humans and AI.
Tools like Frase are useful here for identifying semantic gaps in your content against what's already ranking.
Signal 4: Structured data and schema markup
Schema markup doesn't directly cause AI Overview citations. But it does something almost as valuable: it makes your content machine-readable in a way that reduces ambiguity.
When Google's AI is parsing thousands of pages to construct an overview, structured data is a shortcut. It tells the system "this is a how-to," "this is a FAQ," "this is a product review" -- without requiring the AI to infer it from context.
The schema types most relevant to AI Overview selection:
FAQPage-- directly maps to the question-answer format AI Overviews preferHowTo-- signals step-by-step instructional contentArticlewithauthoranddatePublished-- supports E-E-A-T signalsBreadcrumbList-- helps Google understand your site's topical structure
If you're not sure which pages on your site have schema and which don't, a crawler like Screaming Frog can audit this quickly.
Signal 5: Multimodal content (text + images + video)
This one surprised a lot of content teams when the data came out. Pages combining text, high-quality images, and short-form video see a 317% higher selection rate for AI Overviews compared to text-only pages.
That's a massive difference, and it makes sense when you think about how Google's AI constructs overviews. AI Overviews often include images alongside text. Pages that already have relevant, well-labeled images are easier to pull from -- the AI doesn't have to go find visuals elsewhere.
"Well-labeled" is doing a lot of work in that sentence. The signals that matter:
- Descriptive file names (not
IMG_4521.jpgbutflat-foot-running-shoe-support.jpg) - Alt text that describes the image content accurately
- Captions where appropriate
- Images that are actually relevant to the surrounding text (not stock photos dropped in for visual break)
Short-form video embedded on the page -- even a 60-second explainer -- adds another layer. It signals that the content is comprehensive and that a real person invested in communicating the topic well.
Signal 6: Clear, specific headings that match query intent
Your H2 and H3 headings are essentially the table of contents that AI systems use to navigate your page. If your headings are vague or clever rather than descriptive, the AI has a harder time matching your content to specific queries.
Compare these two heading approaches for an article about email marketing:
Vague: "Making Your Emails Work Harder" Specific: "How to improve email open rates: 6 tested tactics"
The second version maps directly to how someone would phrase a search query. That alignment matters because AI Overviews are built to answer specific questions -- and they're much more likely to pull from a section whose heading already signals it answers that question.
A practical audit: take your top 10 pages by traffic and read through the H2/H3 headings. Ask whether each one could stand alone as a search query. If not, rewrite it so it can.
Signal 7: Page speed and mobile experience
This is the "table stakes" signal -- it won't get you cited on its own, but poor performance here can disqualify you even if everything else is right.
Google's AI Overview selection process still runs through Google's infrastructure, which means Core Web Vitals still matter. A page that loads slowly on mobile, or that has layout shifts that make it hard to read, sends signals that the content isn't being maintained well. That's a trust problem.
The specific metrics to watch:
- Largest Contentful Paint (LCP): under 2.5 seconds
- Cumulative Layout Shift (CLS): under 0.1
- Interaction to Next Paint (INP): under 200 milliseconds
These aren't new requirements, but they're worth re-checking if you haven't audited your Core Web Vitals recently. Google's PageSpeed Insights gives you a free baseline, and tools like Sitebulb can give you a more detailed crawl-level view.
Signal 8: Internal linking that demonstrates topical authority
The last signal is one that's easy to overlook because it's not about the page itself -- it's about how the page connects to the rest of your site.
Google's AI doesn't evaluate pages in isolation. It looks at the broader context: does this domain have other pages on this topic? Does the site structure suggest genuine expertise in this area, or is this a one-off page?
Strong internal linking does two things. First, it passes context -- a page about "email deliverability" that links to related pages about "SPF records," "email list hygiene," and "bounce rate benchmarks" signals that the site covers this topic comprehensively. Second, it helps Google's crawler (and AI crawlers) discover and index all of your relevant content.
The practical fix: for any page you want to optimize for AI Overviews, make sure it links to at least 3 to 5 closely related pages on your site, and that those pages link back to it where relevant. It doesn't need to be complicated -- just intentional.
How these signals work together
None of these signals operates in isolation. A page with perfect schema but vague headings and no author credentials is still a weak candidate. A page with great E-E-A-T signals but slow load times might get passed over for a slightly less authoritative competitor that loads instantly.
The way to think about it: each signal reduces a different type of risk for Google's AI. Answer-first structure reduces the risk of citing something that doesn't actually answer the question. E-E-A-T reduces the risk of citing misinformation. Schema reduces the risk of misclassifying the content. Speed reduces the risk of citing a page that frustrates users.
Stack enough of these signals together, and you become the low-risk, high-quality choice.
Tracking whether your optimizations are working
Here's the frustrating part: AI Overview citations rotate 45.5% of the time per refresh. A page that gets cited this morning might not be cited this afternoon. This makes it genuinely hard to know whether your changes are having an effect without systematic tracking.
Manual spot-checking doesn't scale. You'd need to run hundreds of queries, record which pages get cited, and do it repeatedly over time to see trends. That's not realistic for most teams.
This is where dedicated AI visibility tracking tools become necessary rather than optional. Promptwatch tracks your citations across Google AI Overviews and 9 other AI models, shows you which pages are being cited and how often, and -- importantly -- shows you which prompts competitors are winning where you're not. That last part is what turns tracking into action.
For teams that want to monitor AI Overview appearances specifically, tools like Thruuu and SE Ranking also offer AI Overview tracking as part of their feature sets.
A quick comparison of on-page signal impact
| Signal | Difficulty to implement | Speed of impact | Risk if missing |
|---|---|---|---|
| Answer-first structure | Low | Fast (days to weeks) | High -- AI can't extract a clean answer |
| E-E-A-T signals | Medium | Moderate (weeks) | High -- content gets filtered as generic |
| Semantic completeness | Medium | Moderate (weeks) | Medium -- may lose to more thorough competitors |
| Schema markup | Medium | Fast (days) | Medium -- reduces machine-readability |
| Multimodal content | High | Moderate (weeks) | Medium -- lower selection rate without it |
| Specific headings | Low | Fast (days to weeks) | Medium -- harder for AI to match to queries |
| Page speed / CWV | High | Slow (weeks to months) | Low-Medium -- disqualifier at extremes |
| Internal linking | Low | Moderate (weeks) | Medium -- weakens topical authority signals |
The signals in the "low difficulty, fast impact" category -- answer-first structure and specific headings -- are where most teams should start. They require no technical changes, just editorial discipline, and the effect on citation rates can show up within weeks of a content refresh.
Where to start
If you're looking at this list and feeling overwhelmed, pick one page that matters to your business and run it through all 8 signals as a diagnostic. Does the opening paragraph answer the question directly? Are the headings query-shaped? Does the author have visible credentials? Is there schema? Does it load in under 2.5 seconds on mobile?
Fix what's broken on that one page, track its citation rate for 30 days, and use what you learn to build a repeatable process for the rest of your site.
The brands winning in AI Overviews right now aren't doing anything exotic. They're doing the fundamentals well, consistently, and they're measuring the results closely enough to know what's working.