For years, the playbook for local search was simple: rank at the top of Google, and customers would find you. That assumption is quietly falling apart. New industry research shows that the connection between traditional Google rankings and AI-generated answers — the summaries and recommendations now shown in Google AI Overviews, ChatGPT, Gemini, Copilot, and Perplexity — has weakened dramatically. In mid-2025, roughly three-quarters of AI Overview citations came from pages already in Google's top 10. By early 2026, that figure had dropped to under 40%. For trades, professional firms, and owner-operators who depend on being found online, the message is clear: ranking well is still necessary, but it is no longer sufficient.
Key points
- The share of AI Overview citations drawn from Google's top-10 results has fallen from about 76% in mid-2025 to roughly 38% by early 2026, confirming AI visibility is now its own discipline.
- Google's own May 2026 guidance confirms no special files or markup are needed for AI features — strong conventional SEO remains the entry fee, not a shortcut.
- AI search engines use a multi-stage retrieval process where content can be dropped at any point, which is why a #1 ranking doesn't guarantee an AI citation and vice versa.
- Community sources like Reddit, review platforms, and directories heavily influence AI recommendations — making reviews, citations, and Google Business Profile health more important than ever for local firms.
- A gold rush of unvetted 'GEO' freelancers and gimmick packages has created buyer confusion; business owners should avoid anyone promising guaranteed AI citations, since no vendor controls what AI engines cite.
Multiple studies throughout 2025 and 2026 have tracked how often AI answers cite pages that also rank in Google's top 10. The trend line is striking. In July 2025, analysis of AI Overview citations found about 76% overlap with top-10 results. By January 2026, that number had fallen to roughly 38%, and a large-scale study covering hundreds of thousands of keywords and millions of citation URLs landed at a similar figure. The pattern holds across engines, too — one analysis found that citations in ChatGPT, Gemini, and Copilot overlapped with Google's top 10 by only around 12%, while Perplexity, which runs its own independent retrieval pipeline, overlapped at roughly 29%. Perhaps most telling: about 28% of ChatGPT's most-cited pages don't rank for any meaningful keywords at all. In other words, AI systems are increasingly pulling answers from places traditional rank tracking never looks.
The reason lies in how AI-powered search actually works. When someone asks an AI assistant or sees a Google AI Overview, the system doesn't simply read the top result. It breaks the question into multiple hidden sub-queries, retrieves content from its own index, reranks the fragments it finds, and then generates an answer with citations. Content can fall out of the process at any of these stages — retrieval, reranking, or generation — even if it holds the top organic position. These sub-queries are invisible in standard analytics tools, and any tool claiming to display them exactly is offering a simulation, not reality. The practical takeaway for business owners: treat AI citations as a separate performance indicator from rankings and referral traffic. A page can rank first and never appear in an AI answer, and a well-structured page ranking fifth can be cited repeatedly.
Amid the hype, Google has been refreshingly direct. In May 2026, the company published official guidance on optimizing websites for its generative AI features, and its core message was that no special files, secret markup, or new tricks are required. Its AI features sit on top of the existing ranking systems, and the fundamentals — crawlable pages, useful content, clear business facts, and a technically sound site — remain what matters. Google engineers have also explicitly stated that the much-discussed llms.txt file is not something Google uses. While the file is harmless and can serve as a useful content exercise, some agencies have been selling it as an 'AI Overviews trick,' which it is not. The honest framing, echoed by credible vendors across the industry, is that AI search optimization extends conventional SEO rather than replacing it.
So what do AI systems draw on when they answer questions like 'best HVAC company near me' or 'who's a good accountant in my area'? Research into AI citation patterns points consistently to what might be called an evidence layer: the open-web footprint a business leaves behind. Community platforms dominate — Reddit is among the most-cited sources across major AI engines, and Wikipedia accounts for a significant share of ChatGPT's top citations. Review sites, local directories, consistent business listings, structured data, named authors with real expertise, and crawlable, genuinely useful content all feed what AI systems discover and reuse. Notably, only about a third of cited pages are ones a company can directly influence — which makes the pages you can control, like your website, your Google Business Profile, and your review presence, all the more valuable. For local businesses, this reframes reviews and citations not as ancillary tasks but as core inputs to how AI describes and recommends you.
Where attention flows, confusion follows. Recent marketplace analysis found that of hundreds of freelancer profiles advertising 'GEO' services on one major platform, the majority had no client reviews at all — and many appear to be SEO sellers who simply relabelled their offerings. Dubious packages like '500 citations and directories' are being marketed as AI optimization. Meanwhile, a serious measurement problem persists: AI visibility numbers vary enormously depending on which questions you ask, which engine you check, and how many times you run the prompt. A single AI answer is a sample, not a market. Responsible practitioners now recommend protocol-based tracking — repeated, controlled prompts across multiple engines — with results reported as directional trends rather than guarantees. That last word matters: no vendor, platform, or consultant controls what ChatGPT, Gemini, Claude, Perplexity, or Google's AI features choose to cite. Anyone promising guaranteed AI citations is selling something they cannot deliver.
The good news is that the highest-impact first steps are concrete and checkable. Confirm that AI crawlers can actually access your site — search bots like OAI-SearchBot, PerplexityBot, and Googlebot need access, and it's worth knowing that GPTBot is a separate, training-only crawler. Verify your robots.txt and any CDN security settings aren't accidentally blocking the bots that matter. Fix schema and structured data so your business facts — services, locations, hours, credentials — are unambiguous. Audit your review presence and local citations for consistency. And build genuinely useful content that answers the questions your customers actually ask. None of this is exotic; it's the fundamentals, applied with AI search in mind. A structured visibility audit — one that checks crawler access, entity consistency, content quality, and local signals, and reports findings honestly — is the sensible entry point before committing to any ongoing program.
The decoupling of rankings from AI citations is not a reason to panic — it's a reason to widen the lens. Strong conventional SEO remains the foundation that AI systems build on, but the answers customers now read increasingly depend on the evidence layer around your business: reviews, citations, community presence, and content AI can find, trust, and reuse. For small businesses and professional firms, the path forward is unglamorous but effective: keep the fundamentals tight, track AI visibility as its own directional metric, and be skeptical of anyone promising shortcuts or guaranteed placements in answers no one controls. The businesses that get this right in 2026 won't be the ones chasing tricks — they'll be the ones AI systems can confidently describe, verify, and recommend.