Schema does not cause AI citations. Ahrefs ran the test.

Between August 2025 and March 2026, Ahrefs ran the controlled test the schema industry does not quote. They tracked 1,885 pages that added JSON-LD structured data against 4,000 matched control pages. AI Overview citations moved −4.6%. AI Mode moved +2.4% and ChatGPT +2.2%, and neither of those was statistically significant (Ahrefs, 2026). Add schema anyway — it keeps your name, address and hours machine-readable. Just stop paying for it as an AI visibility service.

This matters because the shortlist moved. 45% of consumers have now used AI tools for local business recommendations, up from 6% a year earlier, which puts AI third among discovery tools behind Google and Facebook and ahead of Yelp and TripAdvisor, on US data; among those who use it, 63% trust what they get back (BrightLocal, 2026). When an assistant names two or three businesses in your category and city and yours is not among them, you were never in the running.

Rank is the wrong scoreboard

Ahrefs found in July 2025 that 76.10% of AI Overview citations came from pages ranking in Google's top 10, across 1.9 million citations (Ahrefs, 2025). The March 2026 re-run over 863,000 keywords put that at 38%, with 31.2% of cited pages ranking 11–100 and 31.0% not ranking at all (Search Engine Journal, 2026). BrightEdge, tracking nine industries, saw the overlap move the other way, from 32.3% to 54.5%, with healthcare at 75.3% and restaurants at 19.2% (BrightEdge, 2025). Two credible datasets pointing opposite ways means you hold the range instead of picking a number: ranking correlates with citation, it does not produce it.

So measure the thing you actually care about. Write 30 questions a customer would really ask — ten category-plus-city, ten constraint questions on price, hours, insurance, language and availability, five comparisons, five about your own name. Run all 30 on ChatGPT, Perplexity, Google AI Mode, Gemini and Claude in a signed-out session. That gives you 150 answers and two numbers: your citation rate, the share that name you, and your error rate, the share that state something false about you. Most local businesses start in single digits. That is normal, and it moves.

Mentions beat links, and most citations are already yours

Across 75,000 brands, Ahrefs measured which signals correlate with AI Overview visibility: branded web mentions 0.664, branded anchors 0.527, branded search volume 0.392, Domain Rating 0.326, and backlinks last at 0.218 (Ahrefs, 2025). Being named in plain text on somebody else's site outranks being linked from it. For a local business that means chamber pages, supplier pages, sponsorships, local press, association directories and "best of" round-ups.

Then the finding that reorders the whole plan. Yext analyzed 6.8 million AI citations across 1.6 million queries per model in mid-2025 and found 86% came from sources the brand controls: first-party websites 44%, listings 42%, reviews and social 8%, forums 2%. Gemini leaned hardest on websites at 52.1%, OpenAI on listings at 48.7% (Yext, 2025). You are not fighting Reddit for your own category. You are competing on assets you can edit this afternoon.

Two constraints sit on top of that. Uberall's May 2026 analysis reports effective rating floors before an assistant will recommend a business at all: ChatGPT prioritizes 4.3 stars and above, Perplexity 4.1+, Gemini 3.9+ (Uberall, 2026). Below those lines, page-level work does not rescue you. On content shape, the Princeton and IIT-Delhi generative engine optimization paper, tested on 10,000 queries across 25 domains, found quotations lifted position-adjusted visibility by 41%, statistics by 33% and citing sources by 28%, while keyword stuffing reduced it by 8–10% (Princeton / IIT Delhi, arXiv 2311.09735).

The two-minute check, and the file that does nothing

Open yoursite.com/robots.txt. OpenAI's crawler documentation is unusually direct: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links" (OpenAI crawler documentation). A block on OAI-SearchBot, or a blanket Disallow: / under User-agent: *, means you opted out of ChatGPT search results. GPTBot is training-only and ChatGPT-User is user-initiated; neither decides whether you appear. Many sites blocked all three in 2024 and never revisited it. Blocking GPTBot to stay out of training data is legitimate and does not require blocking OAI-SearchBot.

Now the file being sold to you. John Mueller of Google, in June 2025: "FWIW no AI system currently uses llms.txt" (Search Engine Roundtable, 2025). Google's own documentation says it more broadly: "You don't need to create new machine readable files, AI text files, or markup to appear in these features" (Google Search Central). Publishing an llms.txt costs nothing. Promising results from it is not defensible.

Picture a 40-seat restaurant

Picture a 40-seat neighborhood restaurant, one location, no agency. Uberall found 83% of restaurant locations entirely invisible in AI recommendations, while the top three brands in a category take 53.4% of share of voice (Uberall, 2026). The demand is there: 22% of US diners have used AI to choose a restaurant, rising to 61% among 25 to 34 year olds, and listings and review platforms account for over 41% of the sources AI cites for restaurants (Bloom Intelligence, 2026).

The audit comes back at four citations out of 150. The menu is a JPG, so no engine can read a single price. The events page says "ask us about private hire". The rating is 4.1, under the ChatGPT floor. Four fixes, in that order: rebuild the menu as HTML text with prices, allergens and dietary tags; publish a groups page with seated and standing capacity, minimum spend and lead time; answer the constraint questions in plain text — walk-ins on a Saturday, open past midnight, corkage, high chairs; and work the rating above 4.3 before anything else.

Why that order works is visible in an adjacent dataset. Across 120 gyms in 50 North American markets, AI assistants scored 48 out of 100 on understanding those businesses, with pricing at 0.10 out of 1.0, and hedged 64% of answers with lines such as "prices aren't listed, you'd want to check with them" (Courtyard, 2026). A hedged answer is a lost customer. Publishing a price range is a citation strategy.

What's in the playbook

• The five-phase method — entity audit, structured facts, citation corpus, cross-engine distribution, monitoring — each with the check that says it is done
• The 30-question audit template with ten examples to adapt, and the four fields you log per answer
• A 90-day plan in five rows, at roughly four hours a week, and what "done" looks like at each stage
• 14 industry playbooks — hotels, restaurants, health, beauty, general SMB, law, real estate, retail and gyms, automotive, education, events, financial services, home services, wine — each with its own numbers, five moves, the compliance caution and a 30-day checklist
• The entity fact sheet, the monthly monitoring log and the ten-point page-fact checklist
• Nine prompts for Claude, ChatGPT or Gemini, including the robots.txt audit and the page that answers a question you lost
• The Latin America step: city and country named together, local currency, the regional language tag, and WhatsApp above the fold
• 34 pages, 50 numbered sources with every figure linked, and both the English and Spanish PDF in one purchase

Do one thing before you read any more advice. Open an assistant you have never signed into, ask the one question your best customer would ask before choosing a business like yours, and write down who it names and which sources it cites. That is your starting position, and it takes four minutes. Get Cited by AI is at netwebmedia.com/playbooks, with the Spanish edition in the same purchase.

Questions owners ask

Does adding schema markup get my business cited by AI?

No, on the only controlled evidence available. Ahrefs tracked 1,885 pages that added JSON-LD against 4,000 matched controls: AI Overviews −4.6%, AI Mode +2.4%, ChatGPT +2.2%, the last two not statistically significant (Ahrefs, 2026). Add it as hygiene, not as a paid visibility service.

Should I publish an llms.txt file?

It costs nothing and does nothing today. John Mueller of Google said in June 2025: “FWIW no AI system currently uses llms.txt” (Search Engine Roundtable, 2025). Google's own documentation adds that no new machine-readable or AI text files are needed to appear in AI Overviews or AI Mode (Google Search Central).

What is the first thing to check?

Your robots.txt and your rating. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers (OpenAI crawler documentation). Then the floors: ChatGPT prioritizes 4.3 stars and above, Perplexity 4.1+, Gemini 3.9+ (Uberall, 2026).

← Back to all articles