There is no slot to buy. So here is the work.
Assistants compose answers from what they can find written about a category. That makes this a coverage and description problem, not an advertising one — which is worse news for your budget and better news for your odds.
Start from the mechanism. An assistant asked for a recommendation runs searches, reads a handful of pages, and writes a paragraph naming two to five businesses. Nothing in that loop accepts payment. So every tactic below is about changing what there is to read.
One: be in the pages that decide the answer
Pull the cited sources for the questions you lose. In one run we did, a single roundup article was cited 126 times across the run, and the brand’s own website did not reach the most-read sources at all. That is your list, and it is finite.
Work it by citation frequency, not by how impressive the publication looks. Most roundups are maintained by one person who accepts submissions, and most category directories have a form.
Two: be describable in one sentence
Models quote sentences that are already clear. They do not do the work of turning a vague positioning into a specific one. If the shortest true description of you is "a platform for modern teams", there is nothing for an answer to carry.
The test: write the sentence you want an assistant to say about you. If it contains no category, no differentiator and no one you are better than, it is not a sentence an assistant needs.
Three: make your name resolvable
This is the one nobody expects and we keep finding. A brand that appears under several variants gets counted, and described, as several different things. In a recent run a restaurant appeared under three spellings of itself, and each one landed as a separate entity — so the answer that recommended it first for its speciality did not register against the brand at all.
- Pick one canonical name and use it in every directory, profile and listing.
- If your legal name and your trading name differ, make one page state both in the same sentence.
- If your name is also an ordinary word, expect noise and measure with whole-word matching or not at all.
Four: publish the thing nobody else will
The passages that get quoted are the ones with a real answer in them, including where you are the wrong choice. A comparison that concludes you should always be picked is discounted; a comparison that names the case for the other option is quotable, because it is the only kind that reads as information.
How long this takes
Months, on a schedule you do not control, because a source has to exist and then be re-read. Measure monthly and refuse to call a difference a change until it clears the interval — in one category we measured, ordinary variation ran about five points wide.
The short answers
How do I get ChatGPT to mention my business?
By changing what there is to read about your category. Assistants compose from pages they retrieve, so the levers are being present in the sources they already cite, being describable in one clear sentence, having a name that resolves to one entity, and publishing something genuinely useful about the buying decision.
Can I pay to appear in AI recommendations?
No. There is no inventory in a composed answer, and anyone quoting you a position is describing something they do not control.
What is the fastest thing I can do?
Get onto the specific pages that are already deciding answers in your category. Pull the cited sources for the questions where you are absent and sort by how often each domain appears — that list is usually short and often reachable.
Does my brand name matter?
More than most people expect. A brand that appears under several spellings is counted and described as several entities. We measured a business whose top recommendation in its own speciality did not register against its brand at all, because the answer used a different variant of its name.
Should I write content on my own site for AI?
It helps, but less than the third-party coverage does. Your site can say you are excellent; it cannot credibly say you are the best of five options, and a comparative question needs the second kind of sentence.
How long before anything changes?
Months. A source has to exist, then be re-read, and neither is on your schedule. Measure monthly and compare intervals rather than points.
How will I know it worked?
Same questions, same wording, at least three runs per reading, and a difference only counts when the confidence intervals stop overlapping. In one category we measured, ordinary variation was about five percentage points wide.
Find out in under ten minutes
One free check, no account and no card. You get the answers the assistants gave, not a score.
Every figure quoted here comes from a real question put to a real assistant, saved with the date and the assistant that answered it. This piece was written to answer the search "how to get mentioned by chatgpt".