AI SEO: How to Get Your Business Cited by ChatGPT

To get cited by AI search — ChatGPT, Google's AI Overviews, Perplexity — you need three things: the crawlers have to be allowed in, your pages have to answer specific questions in plain, extractable language, and the facts about your business have to be consistent everywhere they appear. That's most of it. AI SEO isn't a separate discipline you buy; it's ordinary good structure done deliberately.

Here's what actually moves the needle, and what's being sold to you as magic.

What's actually different about AI search?

Traditional search gives you ten links and lets the user choose. You could rank seventh and still get clicks.

AI search gives one answer and cites a handful of sources. That changes the math in both directions. Ranking seventh is now worth close to nothing — the user never scrolls to a list that isn't there. But being one of the three sources an assistant cites when someone asks "who's a good web developer near me" is worth far more than a seventh-place blue link ever was.

So the game shifts from rank on a page of results to be the thing the machine quotes. Those are related but not identical problems, and the second one has clearer mechanics than most people assume.

How does an AI assistant decide what to cite?

Strip away the mystique and there are four gates. You have to clear all of them.

Most small business sites fail at gate one or gate two, which is good news — those are the two you can fix fastest.

Step 1: let the crawlers in

Different bots do different jobs. Search discovery, model training, and visits requested by a user have separate controls. Check the purpose before changing a rule:

You can opt out of training while allowing search access. The independent controls are documented by OpenAI and Anthropic. Perplexity documents its search crawler separately.

For Google, use Googlebot and Search controls for AI Overviews and AI Mode. Google-Extended covers training and grounding in specified Gemini products. Allowing any crawler makes access possible; it does not promise indexing or a citation.

Check yours right now — it's a public file:

curl -s https://yourdomain.com/robots.txt

If a rule blocks a search crawler from a public page you want found, investigate it. A training opt-out is a separate decision. Keep intentional restrictions on private or staging pages, and check your hosting or security settings too.

The other access killer: if your site renders its content entirely in the browser with JavaScript, many crawlers fetch an effectively empty page. This is the same failure that keeps sites out of normal Google results, and I walked through how to diagnose it in why your website isn't showing up on Google. If a crawler sees a blank shell, no amount of AI optimization matters.

Step 2: answer the question in the first two sentences

This is the highest-leverage writing change you can make, and it costs nothing.

AI systems don't cite pages so much as passages. The system is looking for a self-contained chunk of text that directly answers the question, which it can lift with confidence. Copy that circles the point for three paragraphs before arriving at it never gets extracted, because there's no clean chunk to take.

So invert the usual structure. State the answer, then explain it.

Doesn't get cited:

In today's competitive market, pricing a website involves many factors. Every business is unique, and we pride ourselves on tailored solutions that meet your needs...

Gets cited:

A small business website costs $899–$2,500 from a freelancer in 2026. DIY builders run $16–$49/month; agencies typically start at $5,000.

The second one is quotable. It contains a specific claim, in one sentence, that answers the exact question someone typed. That's the entire trick.

Practical version:

• Phrase your H2s as real questions — the ones people actually ask.
• Put a direct answer immediately under each H2, before context or caveats.
• Use numbers and specifics. "Fast" isn't extractable; "loads in under two seconds" is.
• Keep one idea per paragraph. Long compound paragraphs are hard to lift cleanly.
• Add a real FAQ section with genuine questions and complete standalone answers.

A well-built set of small business pages does this naturally — every page has a clear job and answers one question properly.

Step 3: hand machines the facts in a format they can't misread

Structured data is JSON-LD markup that states your business facts unambiguously instead of hoping a parser infers them from your layout. It doesn't make you rank. It removes ambiguity, which matters more when a machine is deciding whether to repeat something as fact.

The types worth having:

• LocalBusiness — name, address, phone, hours, service area, price range.
• FAQPage — your questions and answers, explicitly paired.
• Article — for blog posts: author, publish date, topic.
• BreadcrumbList — how the page sits in your site.

Google publishes the canonical reference for structured data, and schema.org documents the vocabulary itself. You can validate what you've got with Google's Rich Results Test.

Use structured data when it accurately describes the visible page. Google says no special schema is required for AI Overviews or AI Mode. Helpful markup can clarify your content, but it is not a requirement for every AI citation or a guarantee of one.

Step 4: be consistent everywhere

This is the gate people ignore, and it's the one that quietly disqualifies you.

Before an AI system states something as fact, it generally wants to see that fact corroborated in more than one place. If your website says you're open until 6, your Google Business Profile says 5, and an old directory listing says 7, the system has three conflicting claims and low confidence in all of them. The safe move — from the model's perspective — is to cite a business whose details agree with themselves.

So make these identical everywhere they appear:

• Business name, exactly — including whether it's "&" or "and"
• Street address, formatted the same way
• Phone number
• Hours
• Services offered, described in the same words

"Everywhere" means your site, your Google Business Profile, Bing Places, Apple Business Connect, Yelp, and any directory that has you listed. Stale listings you forgot about are actively working against you.

What about llms.txt?

llms.txt is a proposed convention — a plain-text file at your domain root pointing AI systems at your most important content, roughly analogous to a sitemap. The proposal lives at llmstxt.org.

My honest read: no major provider has committed to honoring it, and there's no evidence it affects citations today. I have one on this site because it takes ten minutes and costs nothing if the convention catches on. That's the right level of investment. If someone quotes you a monthly retainer for "llms.txt optimization," you're being sold weather.

What doesn't work

The AI SEO space has filled up with repackaged nonsense fast. Things to walk away from:

• "Guaranteed AI Overview placement." Nobody can guarantee this. AI answers vary by phrasing, location, session, and model version — ask the same question twice and you can get different sources. Guarantees are impossible by construction.
• Keyword stuffing "for AI." Language models handle synonyms and paraphrase natively. Cramming exact-match phrases makes your copy worse to read and no easier to extract.
• Bulk AI-generated articles to "feed the machines." This produces exactly the interchangeable, thin content that both search engines and AI systems are getting better at discounting. I went into why in why most AI-built websites fail.
• "GEO packages" that turn out to be schema markup plus a blog post per month. That's ordinary SEO with a new sticker and a higher price.
• Treating every AI bot as the same thing. A training opt-out can coexist with search access. Review each bot's purpose before applying a blanket rule.

The checklist, in priority order

If you do these in order, you'll clear the gates that actually matter:

1. Check search-crawler access to the public pages you want discovered. Keep training preferences and private-page restrictions separate.
2. Confirm your content exists in the HTML, not only after JavaScript runs. View source and search for a sentence from your page.
3. Rewrite your top three pages so each section answers its question in the first two sentences.
4. Add a real FAQ to your main service pages, with complete standalone answers.
5. Use relevant structured data where it matches the visible content, then validate it. Special AI markup is not required.
6. Audit your listings for name, address, phone, and hours consistency. Fix every mismatch you find.
7. Add llms.txt if you have ten spare minutes. Skip if you don't.

Steps 1–4 are free and account for most of the result. Steps 5–6 are the ones people skip and then wonder why nothing changed.

The uncomfortable part

All of this assumes there's something worth citing on your site. An AI assistant recommending a web developer needs a reason to pick one — a page that explains the work, real prices, actual examples. If your site is four sentences and a contact form, there is no passage to extract and no fact to corroborate. The optimization has nothing to optimize.

That's the real work, and it's the same work that made sites rank before any of this existed: say specific, true, useful things about what you do, in a structure a machine can parse. AI search rewards that more directly than the old system did, because vagueness now fails visibly instead of just ranking poorly.

If you want someone to go through your site and fix the gates in order — crawlers, structure, schema, consistency — that's part of how I build and rebuild sites. Send me your domain and I'll tell you which gate you're failing, for free. Usually it's the first one.