There's a scene early in High Fidelity where three record-shop obsessives argue over a "top five". The owner offers his picks. One of his staff rips them apart for being too obvious, the sort of list anyone could have written. Nobody wins, because there's no right answer. It depends on how you frame the question, and who's answering.
I thought about that scene a lot while writing this, because the AI marketing world is having the same argument.
You'll have seen the adverts. "Does your business rank in ChatGPT?" "Get to the top of AI search." "We'll get you ranking in Gemini and Perplexity."
They're selling an answer to a question that doesn't make sense. There is no ranking in ChatGPT. No position one, no page two, no league table you can climb. I’m not saying its a snake-oil pitch - but its not very honest either.
That doesn't mean AI visibility isn't real. It is, and it matters. How AI assistants choose which businesses to recommend sets out the mechanism. This piece is about the other side: what you're being sold, and how to tell whether it means anything. If you buy AI visibility as a ranking, you'll be sold the wrong thing and measured against a number that doesn't exist.
Why there's no such thing as ranking in ChatGPT
A search engine results page is a list. It's ordered, it's fairly stable, and you can see where you sit on it. That's what "ranking" means.
An AI assistant doesn't return a list. It writes an answer, fresh, every time someone asks. Which businesses it mentions, if any, depends on:
- The exact wording of the question. "Best roofer near me" and "who should I call about a leaking flat roof" can produce completely different names.
- The conversation so far. What the person said earlier shapes what comes next.
- Whether it searches the web. Some answers come from what the model learned in training; others pull in live web pages. The sources differ, so the answers differ. Google now does the same inside its own results, with AI Overviews.
- Where the person is, and which assistant they use. ChatGPT, Gemini, Copilot, Perplexity and Claude all work differently.
- Chance. Ask the same question twice and you can get a different answer.
So when someone says you're "ranking third in ChatGPT", ask: for which question, worded how, asked by whom, on which day? There's no single answer, because there's no single list.
What the "AI ranking" tools actually measure
The tools behind these offers aren't useless. But it's worth knowing what they do.
Most of them pick a set of questions, run them through one or more AI assistants, and count how often your brand turns up. Some note where in the answer you appear, or whether you're linked as a source. The result is a percentage or a score.
That's a sample, not a ranking. It tells you how often you appeared for those questions, worded that way, on that run. Change the questions and the number changes. It's useful as a trend if the method stays consistent, but it is not a position you hold.
The problem isn't the measurement. It's the selling of it as "you rank number one", which implies a fixed place that someone can guarantee or move you up. Nobody can.
What actually influences whether AI mentions you
You can't control the answer. You can make it much more likely that you're part of it. Four things matter most:
- Be readable. When an assistant searches the web, it needs to be able to fetch and understand your pages. Content hidden behind scripts, logins or messy markup is easy to skip.
- Be the clearest answer. Pages that state plainly what you do, where, for whom and at what cost are easy to quote. Vague marketing copy isn't.
- Be talked about elsewhere. Reviews, directories, local press, trade bodies, industry roundups. What other people say about you shapes what a model "knows" far more than your own homepage does.
- Be consistent. The same business name, services, location and contact details everywhere, also stated in schema markup so a machine doesn't have to guess. Consistency helps a model connect every mention back to you, rather than splitting you across three half-identities.
If that list looks familiar, it should. Most of it is good SEO and good reputation-building, done properly. That's why SEO isn't dead, and why generative engine optimisation is less new than it sounds. The difference is that AI assistants are less forgiving of thin content and more reliant on what the rest of the web says about you.
Synthesised or cited: why saying something different matters
There's a difference between being used in an AI answer and being cited in one.
If your page says the same things, the same way, as a hundred other pages, an AI assistant doesn't need you specifically. It blends the common view into its own answer and moves on. Your content may have contributed, but nobody can see that, and nobody clicks through. (I've written about this before as the synthesis trap, and about why attribution is the new page one.)
Citation works differently. An assistant names and links a source when that source holds something it can't get elsewhere. That means:
- Proprietary information. Your own data, pricing, case studies, survey results, test findings or figures from real jobs.
- A distinct point of view. A clear opinion, argued well, that goes against the consensus or adds to it.
- A different angle on a familiar topic. The same subject, explained from experience, for a specific audience, or through a lens nobody else uses.
- Specifics that can be attributed. A named process, a named method or a clear claim that only makes sense credited to you.
None of this guarantees a citation. But generic content gets absorbed, and distinctive content gets attributed. If you want the link, give the assistant a reason to point to you rather than just borrow from you.
Or, to go back to the record shop: be the pick worth arguing about, not the one everybody names.
Visibility, not ranking: how we measure it at KOODOS
This is exactly why GroundWork, our SEO investigation, measures visibility and visibility opportunities. It never reports an AI "ranking", because reporting one would mean reporting something that isn't true. What you get is a Demand Map: what your market searches for and asks, who the assistants name and cite for those questions, and where you're missing. You can read a sample Demand Map before deciding anything.
| Selling "AI rankings" | Measuring visibility (GroundWork, then Coverage) | |
|---|---|---|
| What it reports | A position, as if fixed | How often and where you appear across a defined set of questions |
| What it implies | Someone can move you up a list | You can improve the odds of being included |
| How it handles variation | Hides it behind one number | Treats it as normal. GroundWork runs each question three to five times across several assistants and sets the baseline; Coverage re-runs the same question set to track the trend |
| What you get to act on | A score | The gaps: questions where competitors appear and you don't, sources AI cites that don't mention you, content that's missing or unclear |
The opportunities are the valuable part. Knowing you appeared in a share of sampled answers is mildly interesting. Knowing which questions you're absent from, who is showing up instead, and what they have that you don't, gives you something to do on Monday morning.
Five questions to ask anyone selling you AI rankings
- Which questions are you testing, and who chose them? If they picked the ones you already appear for, the score flatters you.
- How many times is each question run, and on which assistants? Answers change from run to run, so a single run is a coin toss, not a measurement. GroundWork runs each question three to five times, depending on the question's scope.
- What happens to my "ranking" if the question is worded differently? If they can't answer, the ranking doesn't mean much.
- What exactly will you change? Look for specifics: content, structure, technical fixes, reviews, citations. "Optimise for AI" isn't a plan.
- Are you guaranteeing a position? If yes, walk away. Nobody controls what an AI assistant says.
If you'd rather look for yourself first, Your second audience is an hour-long audit you can run on your own business with a notepad.
Build something worth citing
The businesses that show up in AI answers aren't the ones that bought a ranking. They're the ones with a website that clearly explains what they do, a reputation that other sites reflect, and facts that line up wherever you look.
That's not a trick. It's a well-built, well-maintained digital asset that you own, which is what it always should have been.
If you want to know where you're visible, where you're missing, and what to do about it, that's what GroundWork is for. See a sample Demand Map, or book a call.