Nobody knows how AI engines pick what to cite. Here's what's durable.

Owners ask me how to get ChatGPT to recommend their business. The honest answer is that nobody outside the companies running these engines knows the exact citation logic, and the engines themselves change it from month to month. Anyone selling “the seven hacks to get cited by AI” is guessing. Some of the guesses are educated. Most are repackaged SEO advice with new vocabulary on top.

There’s a layer underneath the guesswork that doesn’t shift with the next model release. That layer comes from how these engines actually work, not from the behavior of any one model. If you invest there, you’re not chasing a moving target. You’re building something that holds whether ChatGPT, Gemini, or Claude is on top this quarter.

1. If you aren’t indexed, you can’t be cited.

Every AI engine that cites web sources works the same basic way. The model doesn’t pull facts out of its training data. It searches a live index for passages, then writes an answer around what it found. Each engine sources that index differently, and the arrangements change from year to year. Google’s AI Overviews use Google’s own search index. Others run their own crawlers or pay for access to someone else’s. The part that matters to you isn’t which engine uses which index. It’s whether you’re in the index at all.

If your site isn’t in those indexes, the model has nothing to cite when somebody asks for a roofer in your town. You can have detailed, accurate content on the local web and still be invisible.

Indexability is the oldest discipline in search. Submit a sitemap, and check that no robots.txt rule from years ago is blocking the crawlers. The pages describing your business need to return HTML in the response body, not a blank shell that needs JavaScript to render. None of this is new. It’s the floor every other improvement sits on.

2. The engine has to know what your business is.

Before an AI engine can cite you, it has to be sure which business you are. Two businesses with similar names. A phone number that changed five years ago and still shows up on old directory pages. To the engine, that ambiguity is a problem, and it drops the sources it isn’t confident about.

The fix is unglamorous. Use the same business name everywhere your business appears online, and make sure the address and phone on your website match the Google Business Profile and any directory listings you have. The structured-data markup that does the heavy lifting here is LocalBusiness and Organization, which give the engine a machine-readable card with your name, address, phone, and category.

When that markup is missing or wrong, the engine has to infer what kind of business you are from prose. It sometimes guesses correctly. More often it just doesn’t cite you.

3. Content the engine can extract gets quoted more than content it can’t.

These models read your page looking for sentences they can lift into an answer. A page that’s one long paragraph of marketing prose is harder to extract from than a page that says “I serve Hendersonville, Gallatin, and Mt. Juliet” in a clean factual sentence. A page with a clear question-and-answer block is easier to quote than a page where the same answer is buried in the middle of a hero pitch.

This isn’t about dumbing the page down. It’s about giving the engine clear, declarative facts it can pull verbatim. If a customer asks ChatGPT “what are the hours of [business] in [city],” the model wants to find a sentence on your site that literally states the hours. If your site only shows hours inside a designed widget that doesn’t render as text, the model has nothing to quote.

A paper presented at KDD 2024 (Aggarwal et al., “GEO: Generative Engine Optimization”) tested how different content changes affected whether AI engines cited a source. Adding inline citations to authoritative sources and using verbatim quotations measurably improved visibility, with the headline finding that targeted content changes lifted source visibility by up to 40% across the queries they tested. The engines have shifted since the paper landed, but the direction is durable. Source-backed factual content gets cited. Marketing prose doesn’t.

4. Sources that are themselves cited by others get preferred.

The fourth thing is the oldest. Search engines have always ranked sites partly on how often other sites point at them, and AI engines inherit that signal. A page that’s been linked to by a local newspaper or by another business in town carries more authority than a page that exists only as a brochure for its owner.

This is the same trust-and-authority logic that’s driven search rankings for over a decade. AI engines don’t reinvent that wheel. They use the same signals because the signals work. A new domain with no inbound links has to earn that authority over time, and there’s no shortcut. Anyone selling one is selling link-buying schemes that have always carried risk.

What this means if you own a local business

The temptation right now is to chase whatever AI-SEO tactic is on a podcast this month. I’d skip it. The vendors selling “AI SEO” packages mostly bundle the four things above with new branding. The price is higher than the work justifies, and the durable part of what they do is just the four fundamentals.

If you want to invest where the work pays, invest in those fundamentals. Make sure your site is indexed and crawlable. Make your business identity consistent across the web with proper structured data. Write content the engine can extract verbatim. Earn citations from sources that already have authority.

That’s the same list I work on with retainer clients. None of it is exotic. None of it requires a rebuild for most small businesses. It’s the same fix-the-basics work I described in the Hendersonville report and the audit explainer.

If you’d like a written audit on where your site sits against those four, I do them free. Send your URL through the contact form and I’ll look.