You did the checklist. Schema markup on every page. An llms.txt file at the root. A few three-thousand-word guides. You waited. And ChatGPT still names a competitor when a buyer asks who to hire. This is the most common message we get, and the reason is not that you missed a step. Most of the advice being sold points at the wrong target.
Because most of it optimizes your own page, while AI decides who to name from evidence spread across the whole web plus whether it can quote you cleanly. Files and tags like llms.txt and schema are baseline hygiene. They help a machine read your site. They are not the switch that makes an assistant recommend you.
The advice that gets oversold
None of the tactics below are scams. Most are fine to do. The problem is how they are sold: as the lever that gets you into AI answers, when they are closer to keeping your shoes tied. Here is the honest version of each.
llms.txt files
A small text file that tells AI crawlers what your site is about. It is cheap and harmless to publish. But assistants almost never cite it when they build an answer, and Google's own guidance says not to count on it. Ship it if you want. Do not expect a single new mention from it.
Schema markup as the citation lever
Structured data helps machines understand what your page is and who you are. That is real and worth doing. What it does not do is flip a switch that makes ChatGPT recommend you. When Ahrefs tracked 1,885 pages that added schema in 2026, AI citations on pages that already got cited barely moved. Treat schema as a clean foundation, not the thing standing between you and the answer.
"Just rank on Google"
Ranking well helps Google's own AI surfaces, like Gemini and AI Overviews, because they sit on Google's index. It does not carry over to ChatGPT, Perplexity, or Claude. Each runs its own retrieval. When Ahrefs studied 15,000 queries in 2025, only about 12% of the URLs that AI cited ranked in Google's top ten. Page one of Google is worth having, and it still leaves three of the four big assistants unaddressed.
Three-thousand-word pillar pages
Length for its own sake does nothing. An assistant is not counting your words. It is looking for a clean answer to a real question. A page that answers what a buyer actually asks, in plain sentences near the top, beats a long one that buries it. Write for the question, not the word count.
The industry sells wrappers, files, tags, and templates, because a wrapper is easy to package and charge for. The thing that actually earns a mention is slower: being described clearly across the places AI already trusts. Nobody can sell you that as a one-week sprint, so it rarely makes the pitch.
What actually moves the answer
Two forces decide whether AI names you. The first is what it can quote from your site the moment it looks. The second is how you are described everywhere else it reads: reviews, maps, forums, local press, comparison pages. Your own page is one voice. The rest of the web is the crowd, and the crowd is louder.
The shift is not hypothetical. BrightLocal's 2026 survey found 45% of consumers already use AI to find and pick local services, and Gartner expects 25% of traditional search to move to AI chatbots in 2026. Those buyers never see your Google ranking. They see one answer with a couple of names in it.
So the work is not a file. It is a short list of moves that change what AI can quote and what the web says about you.
- Map the answers first. Ask your buyer's real questions in ChatGPT, Gemini, and Perplexity, and write down who gets named and which pages get cited. That list tells you where the answers come from in your market.
- Rewrite your top pages in the quotable shape: the direct answer in the first few lines, your service and city in plain words, a real comparison or price where you can, a visible date and a named person behind it.
- Show up where AI already looks. In most local markets that means Google reviews, a strong Business Profile, the odd Reddit or forum thread, and local news or association pages. Assistants read these more than they read your homepage.
- Publish one thing only you can: a small price comparison, a local stat, a real before-and-after. Original facts get quoted. Rewritten listicles do not.
- Keep your listings saying the same thing. Same name, address, phone, and category across your site, maps, and directories. Assistants trust sources that agree with each other.
- Measure named versus cited. Being named means AI wrote you into the answer. Being cited means it used your page but named someone else. Track both, because cited-without-named means your content is helping a rival get the call.
There is research behind that last one. A Princeton study presented at KDD 2024 found that adding original statistics, citations, and quotes to a page lifted how often generative engines surfaced it, by up to 40% for some methods. AI rewards facts it cannot find anywhere else.
Assistants are not ranking your homepage. They are triangulating across everything crawlable and naming whoever the evidence describes most clearly.
A ten-minute check you can run today
Open ChatGPT and ask the question your best customer would ask, in their words. Something like best roofer near me for a full replacement, or top family lawyer for a custody case. Then read the answer and check three things.
- Did it name your business? That is a win. Defend it.
- Did it cite your page but name someone else? Your content is doing a competitor's marketing. Fix the pages it quoted.
- Did it skip you entirely? You are invisible for that question. Start with the answer map above.
Turn web search off and ask again. If you still show up, you are in the model's training data, which is slow to change. If you vanish, you only exist through live retrieval, which moves in days. Either way, the answer changes when your evidence changes. We have watched scores move within weeks of a site fix.
Do it yourself, or hand it off
Here is the honest close. Everything above is doable on your own. Give it a few weekends: run the checks, rewrite your top pages, tidy your listings, publish one original thing. That alone puts you ahead of most local competitors, who are still buying llms.txt files.
If you would rather not, that is the job we do. We run the measurement across every assistant, week after week, and make the fixes that change what AI says. Same work, off your plate.
Our free check runs your buyer questions across the assistants, shows whether AI names you, only cites you, or skips you, and lists the fixes in priority order. No card, no call. It is the ten-minute check above, run on every question at once.