When people first learn they’re not showing up in AI recommendations, the instinctive response is to assume something is wrong with the algorithm. The engine is favouring established names, or penalising smaller businesses, or prioritising whoever paid somehow.
Most of the time that’s not what’s happening.
What’s happening is that the engine looked for sources it could build a confident recommendation from, and yours wasn’t clear enough, or consistent enough, or quotable enough to make the cut. That’s not a judgment about your quality. It’s a structural problem with how legible you are to the machine doing the recommending.
Understanding the mechanism makes the fix much cleaner.
I watched this play out with a client. Andre Obradovic is a men’s health coach with a professionally built, expensive, genuinely beautiful website, and the engines had never heard of him. The site converted the humans who reached it, but ChatGPT, Perplexity, Google AI and Grok ran the first round of judging and his name was never in it. After about six weeks of citeability work, fixing nothing about his expertise and everything about how legible it was, all four engines now name him when potential clients ask who to hire, first in most. His page views are up 51% and his email leads doubled month on month (GA4, May to June). Same coach, same quality of work. The only thing that changed was what the machines could read. The full case study is here.
How is an AI recommendation actually built?
The process, simplified, is this: retrieval, then synthesis.
When someone asks ChatGPT or Perplexity who to hire for a service, the engine retrieves relevant information from its training data and, in search-enabled modes, from live web sources. It finds pages, profiles, mentions, reviews, and any other public material associated with businesses or people in the relevant category.
Then it synthesises. It takes what it found, decides which sources to trust, and constructs a response. That response is usually a short list, with reasons attached to each name.
The critical thing to understand is that “retrieval” doesn’t mean “best result”. It means “clearest, most consistent, most quotable result”. The engine is not evaluating the quality of your client outcomes. It’s evaluating whether it can form a clear picture of you and express that picture confidently in the answer.
This is why a well-positioned but less accomplished business can outcompete a quietly excellent one. The well-positioned one gave the engine more to work with.
What three factors decide who AI recommends?
The engines don’t publish their recommendation algorithms in detail, and they vary across platforms. What follows is based on observable patterns across ChatGPT, Perplexity, and Google AI Overviews, the factors that consistently appear to separate cited businesses from uncited ones.
Legibility: can the engine tell what you do?
This is the first filter and the one that eliminates most businesses before the others even come into play.
Legibility means the engine can tell, without ambiguity or interpretation, what you do and who you do it for. Not a general category. A specific, repeatable answer it can include in a recommendation with confidence.
Vague positioning fails this test badly. “Strategic partner for growth-minded founders” means nothing to an engine trying to match you to a potential client who asked “who helps with AI search visibility?” The engine has no way to connect those two things because you didn’t give it the connection.
This is why founder-led service businesses are particularly exposed. Founders often position at the outcome level (“I help you scale with ease”) rather than the method level (“I improve your AI search visibility by building the legibility, consistency, and quotability signals the engines need”). The first is inspiring in a sales conversation. The second is what an AI can actually use.
Legibility lives primarily on your main web pages. Your homepage, your about page, your services pages. These are what the engine will read first. If none of them say plainly what you do, in terms that match what a potential client would search for, you fail the first filter.
Andre’s site failed exactly here. His credentials were real, ICF and Primal Health certifications among them, but nothing on the page declared them in a way a machine could verify, and his location page was titled “Location.” To a potential client, a polished site. To an engine, a blank. Wiring his name and his certifications back to the bodies that grant them was what let the engines confirm he was a real, credentialed person, and start naming him instead of someone they could place more confidently.
Consistency: do your sources agree?
If your homepage describes you one way and your LinkedIn another, and your Clutch listing says something different again, the engine has contradictory signals. It’s trying to form a unified picture of you from multiple sources, and when those sources disagree, the picture blurs.
A blurred picture is less likely to be cited than a clear one. Not because the engine punishes inconsistency explicitly, but because the confidence required to include someone in a recommendation requires a clear picture. If the engine is uncertain what you actually do, it will default to someone it’s less uncertain about.
This matters a lot after a pivot or repositioning. The old signals persist. Directory listings you forgot about still say what you used to do. An interview from three years ago still ranks for your name. Your LinkedIn summary hasn’t been updated in eighteen months. All of these contribute to an entity picture that reflects who you were rather than who you are.
Consistency work is unglamorous. It means going through every public profile and source that mentions you and making sure they all say a version of the same thing. But it moves the needle faster than almost any content work, because it removes the confusion before the engine gets to evaluate you.
This is the part of Andre’s work still running. We’re aligning every place that mentions him, starting with his Google Business Profile, so his name, his location, and what he does tell one story across the web. It is held, not finished, because the web keeps moving. When the sources agree, the engines stop hedging and start naming.
Entities, not keywords covers this concept in more depth, including why the entity picture matters and how to sharpen it.
Quotability: can the engine lift a clean answer?
Even if the engine can see you clearly and trust what it sees, it still needs something to quote.
An AI recommendation doesn’t just include your name. It includes a reason. “Jane is recommended for her structured approach to operations consulting, with published frameworks on reducing founder decision load.” The engine got that from something Jane wrote or something someone else wrote about her. It’s quotable. It’s specific. It’s self-contained.
If your content is good but not quotable, you’ll be present in the engine’s picture but absent from the recommendation. There’s nowhere to lift the reason from.
Quotable content has a few consistent features: answers that appear near the top of the page rather than buried in paragraphs, headings that are real questions rather than generic labels, definitions that are specific and complete enough to quote without the surrounding context, and FAQ sections that answer what clients actually ask rather than what you want them to ask.
Most service business content is written for human readers who are skimming to decide whether to engage. That’s a different reading goal from an engine pulling a specific answer. Good content for AI retrieval is often more direct, more structured, and more willing to give the answer up front rather than building to it.
What makes a page citeable has the structural specifics if you want to apply this to your pages.
What can you control about getting recommended by AI?
You can control all three ingredients. None of them require gimmicks, paid placement, or gaming anything.
What you can’t control: the exact queries that return your name, the timing of when changes to your site get picked up, or the specific preferences of each engine. The engines don’t publish their recommendation algorithms in detail, and they vary.
What that means practically: focus on the foundations. Clear positioning on key pages. Consistent signals across every public source. Content structured for quotability, not just for reading. These changes compound over time and they’re durable across algorithm changes, because they’re solving a real signal problem rather than chasing a specific implementation.
The place to start is knowing your current number. What does the engine actually say about you right now, across the queries your potential clients are running? That’s a much more productive starting point than guessing.
How to check if AI recommends your business walks you through the prompts. Or if you would rather see the exact answer the engines give right now, with a baseline score and a ranked set of fixes you can start the same day, that is the Citeability Snapshot.