Someone asks ChatGPT which tool they should buy. It names three brands. If yours is not one of them, you did not lose a ranking, you lost the customer without ever appearing in a report.
That is the uncomfortable part of AI search. There is no position eleven to console yourself with, no impressions graph showing you were close. You are named or you are not.
Most advice on this topic jumps straight to tactics. I want to start one step earlier, because when I built a visibility checker and ran it on my own brand, it returned a flat zero while Ahrefs returned 11,236 mentions. I had been publishing for months and had no idea. Almost nobody knows their number before they start optimizing, which is how people spend six months on work they cannot evaluate.
- Ranking in ChatGPT means being cited, not placed. There is one answer and your brand is either named in it or absent.
- Measure before you optimize. Most brands start at zero, and without a baseline you cannot tell which change worked.
- Corroboration beats on-page work. Models name brands they have seen described consistently across several independent sources.
- Comparison content is the highest-leverage format, because a large share of buying questions asked to AI are comparisons.
”Ranking” in ChatGPT is the wrong mental model
Google gives you ten links and a click-through rate. You can be tenth and still get traffic. ChatGPT gives one answer, usually naming two to five options, and everything else is invisible. There is no long tail of consolation traffic.
That changes what you are optimizing for. You are not competing for a position on a page, you are competing to be one of a handful of brands a model is confident enough to name. Confidence is the operative word, and it is earned differently from rankings.
It also means the win condition is binary and easy to check, which is genuinely good news. You do not need a rank tracker to know if it worked. You ask the question and see if you are in the answer.
Check where you stand before you change anything
Almost every guide skips this and it is the step that saves the most wasted effort. You cannot tell whether a tactic worked if you never recorded the starting point.
The manual version is free and accurate. Write down the ten questions a buyer would actually ask before choosing something like yours. Not “best CRM software”, but the messy real ones: “what should I use instead of X”, “cheapest tool for Y that does Z”. Ask each in ChatGPT, Claude, and Perplexity. Log whether you are named, and which brands are named instead. That list of competitors is the most useful output, because those pages are your blueprint.
The fast version takes about ten seconds. Our free AI Visibility Checker scores 0 to 100 on how consistently ChatGPT, Google AI, Gemini, Perplexity, and Claude mention your brand, and names the brands they recommend in your place. Two checks, no signup.
One thing worth understanding before you read your own number. A zero does not mean AI has never mentioned you. It means you did not appear in a large sample of answers, so your mentions are too infrequent to register. Plenty of brands see a trickle of AI referral traffic in their analytics while scoring zero, and both facts are true at once. Occasional citations are luck. Consistent ones are the thing you are building.
How ChatGPT decides which brands to name
Two mechanisms, and they behave differently.
Model knowledge is what the model absorbed during training. Slow to change, heavily weighted toward things described consistently across many sources. This is why a brand mentioned in fifty comparison articles gets named and a brand with one excellent homepage does not.
Live retrieval is what happens when ChatGPT browses. Much faster, closer to classic SEO, and reachable within days of publishing. It favors pages that are crawlable, clearly structured, and directly answer the question asked.
You need both. Retrieval gets you into today’s answer. Model knowledge is what makes you a default six months from now. Most people optimize only for the first and wonder why the effect keeps evaporating.
The seven things that actually move it
1. Answer the question in the first two sentences: Models lift the clearest self-contained answer they can find. A three-paragraph preamble means the extractable part is your competitor’s page.
2. Write sentences that survive being quoted out of context: Test it directly: pull any key sentence out of the page and read it alone. If it needs the paragraph above to make sense, rewrite it. Define the term, attach the number, name the source.
3. Earn mentions on sources models already trust: This is the highest-leverage item and the one nobody wants to hear, because it is not on-page work. Reddit threads, comparison roundups, forum answers, and independent reviews carry weight out of proportion to their traffic, because they let a model see the same claim corroborated in more than one place.
4. Be unambiguous about who you are: Consistent brand naming, Organization schema, a real About page, a named author with credentials. Models cite entities they can identify confidently, and hedge on ones they cannot.
5. Own the comparison queries: A large share of buying questions put to AI are comparisons, alternatives, and “which should I pick”. If you have no honest comparison content, the model uses somebody else’s, and theirs will not be flattering to you.
6. Keep it fresh and crawlable: Real published and updated dates, no JavaScript wall in front of the content, fast enough to fetch. Answer engines favor recently updated sources, and a page they cannot render is a page that does not exist.
7. Re-measure monthly: AI answers move in days. Check the same questions on a schedule, otherwise you are reading noise as signal.
What does not work
Keyword stuffing aimed at models: They are trained on natural language and repetition reads as low quality. It hurts the extractability you are trying to build.
Treating llms.txt as a switch: It is a reasonable convention and costs nothing, but published research found the overwhelming majority of these files receive effectively no real traffic, and most requests to them come from bots. Add it if you like. Do not expect it to change your visibility on its own.
Buying an AI optimization package before you have a baseline: If you do not know your starting number, you cannot evaluate what you bought, which is exactly the position most people are in when they buy.
How to tell whether it is working
Two signals, and they answer different questions.
Visibility is how consistently models name you across many answers. Measure it monthly with the same question set so the comparison is honest. Rising visibility means the corroboration work is landing.
Referrals are people arriving from an AI tool, visible in your analytics as traffic from ChatGPT, Perplexity, or Claude. This measures conversion of visibility into visitors.
They move independently, and the gap between them is informative rather than contradictory. Referrals without visibility means you got lucky in a handful of answers. Visibility without referrals usually means you are being cited for questions that do not lead anywhere commercial, which is a signal to change the questions you are targeting rather than to work harder on the same ones.
Start with the number
The tactics above are not exotic. Answer clearly, be quotable, get mentioned elsewhere, be identifiable, cover the comparisons. Most of it is work you already know how to do, aimed slightly differently.
What separates people who make progress from people who stay invisible is almost never the tactics. It is that one group knows their number and the other is guessing. Get the baseline, pick the two items above that are weakest for you, and check again in a month.
A zero today is not a verdict. It is an open lane, and it will not stay open.