AI Overviews and Local Search: What Changed in 2026
AI Overviews are taking real share of local-intent search results. What that means for service businesses, GBP optimization, and how to get cited by AI search.
By Chase Weiser
Through the back half of 2025 the share of local-intent searches that surface an AI Overview climbed sharply, and that climb has continued through early 2026. Live SERP samples we ran on commercial-intent local queries at the time of writing show AI Overviews appearing on a clear minority of queries, with prevalence varying widely by vertical (high in some service categories, near zero in others, including some Aegis client verticals where they barely appear today). The direction of travel is the part to plan around. When the AI Overview does fire on a query in your category, the first thing the searcher sees is not the map pack. It is a synthesized answer with three to five citation sources, and your job is to be one of those sources.
I spent the last three months auditing how AI Overviews pick citations for local-intent queries and the patterns are clearer than most agencies will admit. There are maybe six things that move the needle, and the businesses doing them well show meaningfully higher citation rates than peers with similar GBP profiles and review counts.
What the citation looks like in 2026
If you run a query like “best pool service West Palm Beach” today, you will see something close to this structure. A 60-to-90-word synthesized paragraph at the top, followed by three to five citation cards with the business name, a one-sentence quote pulled from the source page, and the source URL. The map pack is below that. Organic results are below the map pack.
Google pulls citations from a mix of the businesses’ own websites, local directories like Yelp and BBB, and third-party “best of” listicles. In the AI Overviews we have been able to dissect, the site-direct slice consistently dominates, with directory citations a distant second and listicle citations a smaller share again. The site-direct citations are also where the real growth is. Yelp gets cited because Yelp has been there forever. You cannot rank a brand-new pool service on Yelp. You can rank it on its own website if the page is built right.
The pages getting cited share six characteristics, and the order matters.
The six things AI Overviews are looking for
Passage-level answer paragraphs
The single biggest factor I see is whether the page has a tight 40-to-80-word paragraph that directly answers the question Google rephrased into the AI Overview. Not the whole page. One paragraph. If a user asks “what does pool service in Jupiter cost,” the paragraph that gets cited reads something like: “Weekly pool service in Jupiter, FL typically runs $130 to $180 per month for residential pools under 20,000 gallons. Saltwater pools and screen-enclosed pools usually price 10 to 20% higher because of additional cleaning time. Most local services include chemicals, basket cleaning, vacuuming, and equipment checks in the base rate.”
That paragraph is citable because it is specific, it answers the implicit question, it includes numbers, and it is exactly the length Google wants to quote. Pages that bury that information inside a 600-word service description rarely get cited.
FAQ schema with question-format H2s
Most of the cited pages I logged had FAQPage schema or FAQ-formatted H2 headings. The pattern looks like an H2 phrased as a question (“How much does pool service cost in Florida?”) followed immediately by a 40-to-80-word answer. Then JSON-LD FAQPage schema wrapping the same content.
Important caveat on FAQPage schema in 2026: Google scaled back FAQ rich results in late 2023 and continued tightening through 2026, so FAQPage is no longer a reliable Google rich-result win for service businesses. The reason to ship it now is AI search. Perplexity and ChatGPT lean on FAQPage heavily, and pages with clean question-and-answer structure get surfaced for long-tail question variants in those engines even when Google itself does not render the rich result.
A clean llms.txt at the site root
This one is newer and most agencies are not doing it yet. An llms.txt file at the site root is a markdown manifest that tells AI agents what the site is about, what the key pages are, and how to find structured data. It is an emerging proposed convention (Jeremy Howard at Answer.AI introduced it in September 2024), not a formally adopted standard. It is also not required by any AI search engine. In our own logs, sites that ship a published llms.txt show higher AI citation rates than peers without one, especially in Perplexity and ChatGPT. The file takes about 30 minutes to write. There is no good reason not to have one.
NAP consistency across the citation graph
Name, address, phone consistency across GBP, Yelp, BBB, Apple Maps, Bing Places, and the top 30 to 50 industry directories is still the foundation. AI Overviews cross-reference NAP data the same way Google’s local algorithm does, and inconsistencies suppress citations. If your phone number on GBP ends in -4550 and your phone number on the BBB listing ends in -1234, you are competing with yourself.
Authoritative sourcing in the body copy
Pages that link to authoritative sources (industry standards bodies, government data, peer-reviewed studies where relevant) get cited at higher rates. For a pool service page, that might mean linking to the Florida Department of Health pool chemical guidelines or the Association of Pool & Spa Professionals safety standards. The AI is reading the page and using outbound link quality as a proxy for trustworthiness.
Fresh content with explicit dates
The cited pages had a “last updated” date or a publication date inside the body, not just in the schema. Pages that read as fresh in 2026 rather than scraped from a 2021 template are getting cited; static pages that have not been touched in two years are mostly invisible to AI Overviews regardless of how well they used to rank.
ChatGPT and Perplexity cite differently
Google AI Overviews are the biggest target, but ChatGPT search and Perplexity now drive a meaningful share of high-intent service queries, and they cite differently.
ChatGPT’s local-search results pull more heavily from Yelp and Google Business Profile data than from the businesses’ own websites. The citation cards are usually three to five GBP listings with the cosmetic of a chat answer wrapped around them. Optimizing for ChatGPT local search is mostly GBP optimization: review velocity, photo count, services list completeness, and Q&A activity.
Perplexity is the opposite. It strongly prefers structured site content over directory data, and it surfaces author and date information aggressively. A blog post on a service business site with clear authorship and a recent date will outrank that same business’s GBP listing in a Perplexity answer about half the time. If you are running SEO in Jupiter, FL for a service business, Perplexity is where blog content earns its keep.
What to actually do this quarter
If you are starting from a baseline GBP and a serviceable website, three moves cover most of the AI Overview opportunity. First, pick the top 15 questions your customers actually ask, write a 40-to-80-word answer paragraph for each, and publish them as H2 questions across your service pages with FAQPage schema. Second, publish an llms.txt file. Third, audit your NAP consistency across the 30 directories that matter for your industry and fix the breaks. In our audits most service businesses turn up between 4 and 11 NAP inconsistencies, and the cleanup typically takes a week.
The businesses doing this in early 2026 are the ones who will be cited by default through 2027.
