AI Overviews Are Eating the Top of Search Results. Here's How to Show Up.

Two years ago, ranking in position one for "plumber Cottonwood Heights" was the whole game. You earned the click, you got the call, end of story. Today, when I run that same search on my phone, the top of the page is a Google-written paragraph answering the question, with three small source links tucked underneath it. The blue-link results don't start until I scroll past the fold.

Most home-services owners I audit haven't reckoned with this yet. Their site might still rank position three. They might still be getting traffic. But the slice of attention that used to sit at the top has been replaced by AI-generated answers, and if your site is not in the cited pool, you are invisible at the moment of decision. This post is about what's changed, who's choosing the cited sources, and what a local-business owner can do this month to start showing up.

What are AI Overviews and what do they look like?

AI Overviews are Google-written summary boxes that appear above the regular search results. They synthesize an answer from multiple websites and show three to five small source links beside the text. Google launched them in May 2024 and expanded them across most informational queries through 2025. ChatGPT Search and Perplexity follow a similar pattern: AI-written answer, citation links underneath.

If you have not seen one yet, search "how much does a roof inspection cost" on your phone. The first thing you see is not a website. It's a few paragraphs of text generated by Google's AI, with little circular icons next to it labeled with the source domains. Tap an icon and you get the page. Don't tap it and you still got the answer.

The same pattern shows up on ChatGPT Search (free for any ChatGPT user since late 2024) and on Perplexity. Both produce a written answer with inline numbered citations. ChatGPT picks roughly four sources per response. Perplexity often pulls from twice as many. This is the new top of search.

How does Google AI Overviews decide which sites to cite?

Google AI Overviews break the original query into multiple sub-queries (called "query fan-out"), then pick sources that show up across many of those sub-queries with clear, direct answers. Pages that rank position one are not automatically cited. Structured content, schema markup, and direct answer paragraphs near the top of the page beat raw ranking position in citation tests.

CXL analyzed 100 AI Overview citations and found that sites cited inside an AI Overview were not always the top-ranked organic result for that same query. Sometimes the cited page sat at position six or seven. What it had that position-one didn't: a clean question-and-answer structure that Google's model could pull a self-contained quote from without rewriting it.

Digital Applied's analysis of 1,000 AI Overviews found that 55% of cited content came from the top 30% of a page. The model reads the first chunk and grabs the answer that looks most quotable. If your home page has a video hero, a slide carousel, and three paragraphs of brand prose before any actual answer, there is nothing for the model to lift.

Schema markup (invisible code that tells search engines what kind of content you have, like "this is a service business" or "this is a frequently asked question") shows up repeatedly in citation studies. SchemaApp reports that pages with FAQ schema appear in AI Overviews 3.2x more often than pages without it. The deep-dive is in the next post at Schema Markup Without the Jargon. The short version: schema hands the AI a labeled map of your content. Without it, the model is guessing.

What about ChatGPT Search and Perplexity? Do they pick sources differently?

ChatGPT Search uses Microsoft Bing's index plus its own crawler (OAI-SearchBot) and tends to favor authoritative reference sites: Wikipedia gets cited in roughly 1 in 6 ChatGPT responses. Perplexity weights Reddit heavily (about 47% of its top citations), prioritizes news sources, and gives 3.2x more citations to content updated within the last 30 days. Same general pattern. Different source preferences.

The practical implication for a home-services owner: each AI search engine has a slightly different bias, but the underlying citation mechanics are similar across all three. Clean structure, direct answers, schema, freshness, and authority signals matter on every platform. You don't optimize separately for each one. You write content that works for all of them.

Perplexity is the most aggressive about freshness. Profound's citation pattern study found that Perplexity favors content updated in the last thirty days at 3.2x the rate of older material. ChatGPT is less aggressive about freshness but rewards original data: surveys, internal stats, real numbers from your own jobs.

What surprised me running my own site through these tools: ChatGPT was citing TruLight SLC for permanent-lighting questions in Salt Lake within a few months of the rebuild. The reason wasn't authority. It was structure. The site has clean question-headed sections and direct answer paragraphs. The model could lift them verbatim. That's the bar.

What about llms.txt? Should I add one to my site?

llms.txt is a proposed file (similar to robots.txt) that tells AI crawlers where the most important content on your site lives. Adoption is real among developer-focused brands like Anthropic, Vercel, and Stripe, but Google has publicly rejected the standard, and there is no statistical evidence that llms.txt files actually improve citation rates. Add one if it's free, but do not prioritize it over schema markup or content structure.

The honest summary: llms.txt is in a murky place as of mid-2026. An independent analysis found about 10% of large domains have one, but server-log studies show that AI crawlers are not actually fetching the file at meaningful rates, and Google's Gary Illyes publicly compared the standard to the long-dead "keywords" meta tag. Front Door Digital builds include an llms.txt because the cost is roughly five minutes and the downside is zero. Treat it as belt-and-suspenders, not as a citation lever.

The actual citation levers, in priority order, look like this:

1. Question-phrased H2 headings. Match the way a real customer would ask the question, not the way you would write a brochure.
2. 40-60 word direct answer immediately under the question. Self-contained. Quotable as a standalone paragraph.
3. Schema markup. At minimum: LocalBusiness on the home page, Service on each service page, FAQ on any page with question-and-answer content.
4. Page speed and mobile-first delivery. AI crawlers time out on slow pages just like humans do. More on that here.
5. Genuine expertise signals. Real author bio, real license numbers, real photos of real jobs. The AI models are getting better at telling the difference between actual experience and recycled content.

What can a home-services owner do this month to start showing up?

Start with five real customer questions and turn each into its own page section, using the four-block pattern: question-phrased H2, 40-60 word direct answer, expansion paragraph, then a bulleted list of details. Add FAQ schema markup to that section. Update your Google Business Profile with the same questions in the Q&A area. This is less than a day of work and starts moving the citation needle within weeks.

The hardest part for most owners is the first step: stop writing pages that talk about you and start writing pages that answer questions. Your "Services" page does not get cited because the AI model has nothing to extract from "Quality work since 1998." Your "Frequently Asked Questions" page does get cited because every H3 is already a question and every paragraph underneath is already an answer.

The four-block content pattern looks like this in practice:

• Block 1, the question H2. "How much does a sewer line replacement cost in Salt Lake?" Not "Sewer Line Services."
• Block 2, the direct answer (40-60 words). Self-contained paragraph that an AI model could pull as a standalone quote. State the answer in the first sentence, qualify it in the second, give the range in the third.
• Block 3, the expansion paragraph. Explain the variables that move the price. Add specifics. Cite real data if you have it ("our last twelve sewer-line jobs in Cottonwood Heights ranged from $4,800 to $12,300").
• Block 4, the bulleted detail. Five to seven bullets covering what's included, what's not, what to expect, what could change the quote.

Repeat this pattern for every real question your customers ask. Pull the questions from your phone notes, your email inbox, the "People Also Ask" box on Google, and Reddit threads where people in your city ask for contractor recommendations. Answer them on your own site, in your own voice.

Will AI Overviews actually replace traditional search results?

No, but they are reshaping what "ranking" means. Google has confirmed AI Overviews will continue rolling out across most informational queries. The blue links are not going away. They are getting pushed down the page and the click-through rate on positions one through three is dropping. The new prize is being one of the three to five sources cited inside the AI Overview itself, which is a different optimization target than the old "rank position one" game.

Search Engine Journal reported in late 2024 that AI Overview citations from top-ranking organic pages had dropped sharply. Translation: just being position one is no longer a guarantee that the AI model will pick you as a source. Sites with worse rankings are getting cited because their content is structured better. This is good news for small operators with clean sites and bad news for legacy domains coasting on backlinks.

The owners taking this seriously are not panicking. They are doing two things. First, auditing their existing top pages for the four-block pattern and rewriting where it's missing. Second, spinning up new pages built question-first, designed from the start to be quotable. A few hours per page, schema added once, then it compounds. Owners who wait twelve months will be competing against neighbors whose pages have been getting cited the whole time.

Why does this matter NOW even if AI Overviews aren't on every search yet?

Because the citation patterns AI models learn now will shape what they pull from for years. Google's models are constantly retraining, but the structural advantages a site builds today (clean H2 hierarchy, schema, FAQ self-containment, fresh content) compound. A competitor who waits eighteen months will be playing catch-up while your site is already in the cited pool. Early moving is genuinely cheap right now and will not stay cheap.

Same dynamic played out with mobile-first indexing in 2018-2020. Owners who rebuilt for mobile early ranked. Owners who waited until Google forced the issue spent more, scrambled harder, and never quite caught up. AI search is following the same curve, just compressed because the tools are evolving faster.

The companion technical move is schema markup, the next post in this series. Without schema, you are betting that the model parses your HTML correctly. With schema, you are handing the model a labeled map. The full breakdown is at Schema Markup Without the Jargon. Speed matters too: AI crawlers have their own timeouts, and a four-second mobile page rates lower than a sub-second one. We rebuilt my own TruLight site partly because the old site was getting deprioritized by every crawler that touched it. Receipts at our TruLight SLC case study.

Frequently asked questions

What is the difference between SEO and GEO?

SEO (Search Engine Optimization) is the practice of ranking high in traditional blue-link search results. GEO (Generative Engine Optimization) is the practice of getting cited inside AI-generated answers from systems like Google AI Overviews, ChatGPT Search, and Perplexity. They overlap heavily. Most GEO best practices (clean structure, schema markup, page speed, real expertise) also help traditional SEO. You do both.

How do I know if my site is appearing in AI Overviews?

Run searches your customers would actually type, on a phone, in an incognito window. If an AI Overview appears at the top of the results, check the small source icons next to the answer. Click through and confirm whether your site is one of them. Repeat this for ten to fifteen real customer queries. Tools like Profound, SE Ranking, and AlsoAsked also track AI citation share if you want to monitor it systematically.

Will adding FAQ schema get my site cited overnight?

No, but it is one of the highest-impact changes you can make. Studies from SchemaApp and others suggest pages with FAQ schema appear in AI Overviews two to three times more often than pages without it. The schema does not generate citations on its own. It makes your existing content easier for AI models to extract from, which only matters if the content is already structured as real questions and real answers.

Do I need to write content specifically for AI search, or will good SEO content work?

Good SEO content that is genuinely written to answer questions usually works for AI search too. The trap is "good SEO content" that is actually keyword-stuffed copy designed to please an algorithm rather than a reader. AI models are better than legacy search at telling the difference. Write the way you would explain something to a customer in your truck on the way to the job site, then add structure on top.

If you read this and immediately ran a search for your own service category to see if AI Overviews show up, that's the right reflex. The next reflex is to look at your own site and ask whether the content there could be cited by an AI model that has six seconds to skim it. If the answer is no, you have a fix list. If you want me to take a look and tell you exactly which sections to rewrite first, that's what the free Front Door Score is for.