AI Search Has Changed. Here's How to Optimize Your Website for AI Overviews.

Search traffic is down. If you've been watching your analytics and wondering what happened, the answer is the same for most websites right now: AI Overviews.

Google started rolling out AI Overviews — those AI-generated answer blocks at the top of search results — in May 2024. Since then, the search result page has been reorganized around them. The 10 blue links are still there. They're just lower, and they're getting fewer clicks.

Here's what site owners need to understand: this isn't a penalty. It's not something you did wrong. It's a structural change in how search works — and it has different rules than the SEO you're used to.

The good news is those rules aren't complicated. The bad news is most websites aren't following them, which means most websites are invisible to AI search.

Let's fix that.

The short version: AI Overviews appear in roughly 16% of all Google searches and have cut organic click-through rates by 61%. Sites *cited inside* the AI Overview earn 35% more clicks than sites that aren't. Getting cited requires answer-first content structure, question-format headings, FAQ schema markup, and content that cites credible sources — not more backlinks or keyword density.

What's Actually Happening to Search Right Now

AI Overviews appeared in just 6.5% of all Google searches in January 2025. By July, that number hit 24.6%. Google pulled it back after some high-profile AI errors and it settled at 15.7% by November (Semrush, 2025). As of early 2026, they appear in 59% of desktop queries where they trigger at all, and 84% of all AI Overview traffic is in the US (Conductor, 2025).

The coverage will keep fluctuating. The direction won't change — these things aren't going away.

The impact on click-through rates is real. When an AI Overview appears on a results page, organic click-through rates drop 61% compared to pre-AIO baselines — from 1.76% down to 0.61% (Seer Interactive, 2025). Users get their answer from the AI box and don't click anything.

Here's the counterintuitive part: if your website is *cited inside* that AI Overview, your organic clicks go up 35% compared to sites that appear on the same results page but aren't cited. Paid clicks go up 91%.

There are effectively three tiers now:

1. Cited in the AI Overview — you benefit
2. Not cited, AIO is present — you lose traffic
3. No AIO on the page — you're in the old normal

The entire goal of GEO (generative engine optimization) is to land in tier one. That's what the rest of this article covers.

Why Most Websites Are Invisible to AI Overviews

AI doesn't read your website the way Google's crawlers have for the last 25 years.

Traditional SEO rewarded pages that were authoritative, well-linked, and optimized for specific keywords. Google's crawler would assess all of that, determine relevance, and rank your page accordingly. You showed up in the top 10 or you didn't.

Generative AI works differently. When someone types a question into Google now, the AI assembles an answer from across multiple sources. It's looking for passages that directly answer the question — short, clear, structured, self-contained. Not long-form narrative. Not pages that eventually get to the point after three paragraphs of context.

Most websites are built to tell a story. AI wants the answer first.

There's another wrinkle worth understanding, and it's one most guides miss. Early data suggested 76% of AI Overview citations came from pages already ranking in Google's top 10. By early 2026, that number had dropped to 38% (Ahrefs, via Search Engine Journal, 2026). What's happening is called "fan-out" — the AI searches for sub-questions related to the main query, not just pulling from the top-ranked page for the main keyword.

A site ranking #40 for a main keyword can be cited in the AI Overview if one of its pages answers a related sub-question particularly well. This is a real opening for smaller sites. You don't have to outrank an established competitor on their main keyword. You have to answer a related question better than they do.

And finally: only 12.4% of websites use any structured data markup (Frase.io, 2025). Structured data helps AI systems understand what your content is and what it's saying. If 87.6% of sites aren't using it, that's not a gap — that's a doorway.

What AI Overviews Actually Pull From

A peer-reviewed study by researchers at Princeton, IIT Delhi, and Cornell (GEO study, arXiv / KDD 2024) tested ten different content strategies against generative AI systems to see which improved citation rates. Three outperformed everything else:

• Adding statistics — AI visibility improved up to 40%
• Adding citations (citing other credible sources within your content) — up to 40% improvement
• Adding expert quotations — up to 30% improvement

The other seven strategies, including general rewrites and keyword optimization, had minimal effect.

This is worth sitting with. The way to get AI to cite your content is to have content that itself cites credible sources. The AI is making a judgment about epistemic quality — does this source show its work? Does it reference real data? Does it quote people who know what they're talking about?

Websites that are opinion without evidence don't get cited. Websites that show their sources do.

Beyond the data signals, the structural patterns that AI Overviews consistently pull from are:

• Question-formatted headings — H2s and H3s that mirror how a user would actually ask the question ("How do I get cited in Google AI Overviews?" not "AI Overview Optimization Tips")
• Answer-first paragraphs — the first sentence of each section answers the question directly, before adding context
• Numbered steps and bullet lists — AI Overviews are displayed as structured, labeled sections; pages that already use this format translate more cleanly into that output
• Self-contained passages — short paragraphs (40–80 words) that make sense in isolation, without needing the surrounding text for context

How to Actually Optimize Your Website for AI Overviews

Here are the specific changes that move the needle.

Restructure your content to answer first

Look at every major section of your service pages, blog posts, and FAQ content. Does each section open with the direct answer to the question implied by the heading, or does it start with background and buildup?

If you write: *"Historically, businesses have approached lead generation in many ways. Customer acquisition costs have varied widely depending on the industry..."* — that's context-first writing. AI skips it.

If you write: *"Referral-based outreach converts at 3–5x the rate of cold outreach for service businesses (HubSpot, 2025)."* — that's answer-first. AI can extract and cite that.

Go through your highest-traffic pages and rewrite the opening paragraph of each H2 to answer the implied question directly. One sentence. Specific. If you can include a statistic, include one. This single change is the highest-leverage thing you can do.

Add FAQ schema to your important pages

FAQPage schema markup tells Google — and the AI — that a section of your page is structured as question-and-answer pairs. Pages with this markup are 3.2x more likely to appear in AI Overviews (Frase.io, 2025).

If your site runs on WordPress, plugins like Rank Math and Yoast handle this without touching code. On a custom site, it's a JSON-LD snippet added to the page — about two hours of developer work per batch of pages. Every FAQ answer should be 40–60 words, self-contained, and answer the question directly. "Great question!" is not an answer.

Write with citations

If you publish blog content, start citing your claims. Link to studies, government data, well-known industry publications. Not just because it helps E-E-A-T signals (though it does), but because — as the Princeton/Cornell study showed — cited content gets cited in return.

This feels counterintuitive to a lot of site owners I work with. "Why would I link people away from my site?" Because the AI is judging your content's credibility partly by whether it demonstrates awareness of the broader body of evidence. A blog post that makes claims without sourcing them reads — to both AI and Google — like opinion, not expertise. You can link out and still close the loop by making the insight or takeaway yours.

Target sub-questions, not just head keywords

The fan-out behavior changes how you should think about your content strategy.

Instead of writing one long "everything about X" article and hoping it ranks for the main keyword, think about what the five or ten related questions are that people ask alongside that main topic. Write sections — or separate posts — that answer those sub-questions in depth.

For example: if your main topic is "small business bookkeeping," you're probably not going to outrank QuickBooks for that keyword. But "how to separate personal and business expenses as a sole proprietor" is a sub-question that a smaller, more focused site can own — and that can get cited when someone asks a broader bookkeeping question.

This is also why page structure matters. A single page that has clearly defined subsections (with H3 headings posing specific questions) gives the AI multiple potential citation targets, not just one.

Use structured formatting throughout

Numbered lists, bullet points, H3 subheadings within sections, tables when comparing options. Not because it looks better (though it does), but because it mirrors the format the AI uses when it assembles answers. AI Overviews are rendered as structured, labeled outputs. Pages that already look like that translate more cleanly.

If your content is currently long-form paragraphs — three to five sentences, flowing narrative — start breaking it up. One main idea per paragraph. Bullet lists for steps or criteria. Numbered sequences for processes.

Check that your pages are actually accessible to AI crawlers

This one's quick. Google's AI systems use the same crawlers as its search index. If your pages are indexed in Google Search, they're accessible. But if you've ever added rules to your `robots.txt` to block certain crawlers, verify you haven't accidentally blocked AI crawlers. Some security plugins add blanket rules that interfere.

If you want to go further, an `llms.txt` file tells AI assistants how to interpret your site's content. It's not yet an official standard, but adoption is growing and early movers have an edge.

GEO vs. SEO: It's Not a Replacement

GEO (generative engine optimization) is not a replacement for SEO. You still need your pages indexed, you still need reasonable domain authority, and ranking in the top 10 still gives you a 57.9% probability of being cited in an AI Overview versus 13.8% at position 50 (Originality.AI, 2025).

What GEO changes is the optimization target. Traditional SEO optimizes for ranking position. GEO optimizes for citation — being pulled into the AI's assembled answer. Those goals overlap significantly, but they're not identical.

The clearest way to explain it: SEO asks "can Google find and rank my page?" GEO asks "when AI assembles an answer to this question, will it use my content?"

The tactics that answer the second question — answer-first structure, question headings, cited evidence, FAQ schema, sub-question targeting — don't hurt your traditional SEO. Most of them help it. They're additive, not a trade-off.

What GEO does require is letting go of the idea that long-form narrative writing is inherently more authoritative. AI doesn't reward length. It rewards directness, structure, and demonstrated rigor.

If your site is the kind that's been doing solid SEO for years — ranking reasonably well, with quality content — you're probably closer than you think. The changes are more about restructuring what's already there than starting over. If you're not sure what's actually working on your site right now, that's a good place to start. Your Website Is Leaving Money on the Table

A Note on Paid Search

If you're running Google Ads and wondering why your costs have been creeping up even as traffic comes down, AI Overviews are part of that story.

Paid click-through rates on queries where an AI Overview is present dropped significantly from pre-AIO baselines. When the AI answers the question at the top of the page, fewer people click on anything — including ads. This is a real cost for businesses running paid campaigns on informational queries.

The practical response is to shift spend toward transactional queries — specific product searches, service plus location queries, bottom-of-funnel terms — where AI Overviews appear less often and users have already decided they want to contact someone or buy something. Informational keywords are increasingly expensive to buy clicks on when AI has already answered the question for free.

What to Do With This

Search is not going back to the way it was. The AI Overview box is at the top of the results page because users engage with it. Google will keep expanding and refining it.

The sites that figure this out now — that restructure their content for answer-first delivery, add schema markup, and start citing credible sources — will have a meaningful advantage over the next 12–24 months. The gap between optimized and unoptimized sites is widening.

None of this requires a major rebuild. It requires rethinking how you write. The goal is no longer just to rank a page. It's to be the source AI trusts when it assembles an answer to a question your customers are asking.

Start with your most important pages. Rewrite the opening of each major section. Add FAQ schema. Cite your data. And if you're not sure where to start or want someone to look at what's specifically going on with your site, that's exactly the kind of thing worth getting a second opinion on before you spend more on ads or content that won't get seen. 5 Questions to Ask Before You Hire an AI Consultant