You typed your own service into ChatGPT the way a customer would. Best commercial roofer in your area. Who should I hire to handle payroll for a 40 person company. It gave you three names, wrote a friendly little paragraph about each one, and none of them were you. One of them you had never heard of.
That is a strange thing to sit with. There is no ranking to look up, no page two to scroll to, and no report that explains the decision. The businesses it named are not paying for the spot. They did not find a setting you missed.
Here is what is going on. ChatGPT is not choosing winners off a leaderboard. It is repeating what the internet has said about your category, over and over, for years, and then shaping that into a sentence. Your competitor is not better at a new game. They have more evidence attached to their name.
That distinction matters, because it changes what is worth doing about it. Below is how a recommendation actually gets assembled, which parts of it you can move, which parts you cannot, and where the effort pays.
Key Takeaways
- ChatGPT tends to name businesses it already knows, then finds sources that support the answer it has drafted.
- The strongest signals sit on other people’s websites, not on yours.
- OpenAI runs three separate crawlers with three different jobs, and blocking the wrong one keeps you out of ChatGPT’s search answers entirely.
- Asking once proves nothing. Answers shift between phrasings and between sessions.
- Nobody sells placement inside ChatGPT’s answers. Any agency promising it is describing a system that does not exist.
What Happens When Someone Asks ChatGPT for a Recommendation
Most advice on this subject jumps straight to a checklist. Fill out your Google profile, get more reviews, add schema markup. The advice is mostly right, but handed over as a list it gives you no way to judge what matters for your business. So start with the machine.
The Model Answers From What It Already Knows
A large language model is not a search index with a chat window bolted on. It was trained by reading an enormous amount of the public internet, and what it retains is patterns: which businesses come up in which contexts, what problems they solve, who they compete with, what people say about them afterward.
When someone asks for a recommendation, the model reaches for the names that pattern already contains. Research from Seer Interactive, which analyzed 541,213 responses across several AI platforms, points to something useful here. Their read is that the model drafts its answer first, choosing names from what it knows, and only then looks for sources to support those choices. In their data, a brand was cited 53.1% of the time when it appeared in the response and 10.6% of the time when it did not.
Put plainly: the citations look like the reasoning, but they behave more like a bibliography assembled afterward. The names came first.
Then It Searches, Sometimes
ChatGPT does not stop at memory. When a question calls for current information, it rewrites that question into web searches and pulls live pages, which is why you sometimes see links in the answer and sometimes do not.
So there are two separate jobs to win. Being known, which is slow and cumulative. And being findable right now, which is closer to the search work you may already be paying for.
Search rankings help with the second job, but they do not decide it. Ahrefs found that only 38% of citations in Google’s AI Overviews came from pages sitting in the top 10 organic results. Ranking well and getting cited are related. They are not the same achievement.
Why the Answer Changes Every Time You Ask
Ask the same question twice and you may get two different lists. Change three words in the prompt and the list changes again. Log out and it changes once more.
That is normal behavior, not a glitch and not a signal about your business. It also means a single screenshot proves very little in either direction. We come back to this later, because it shapes how you should check where you stand.
A Quick Definition
Generative search is search that hands you a written answer instead of a list of blue links. ChatGPT does it, Google’s AI Overviews do it, Perplexity does it. The work of improving how you show up in those answers gets called generative engine optimization, or GEO, and sometimes answer engine optimization, or AEO. Different labels, same underlying job: making sure the systems writing those answers know who you are and have reason to trust it.
The Signals That Decide Who Gets Named
Mentions Beat Links
Ahrefs studied 75,000 brands to see which measurable factors lined up with appearing in AI answers. Branded web mentions, meaning your name written on other people’s sites whether or not they link to you, showed the strongest relationship at 0.664. Backlinks, the currency of traditional search for two decades, came in at 0.218. A later cut of the same research put mentions in YouTube titles, descriptions, and transcripts higher still, around 0.737.
The three strongest factors were all things happening away from your website.
For an owner, that is worth translating. Being talked about counts for more than being linked to. That makes this look less like the technical SEO you may have bought before and more like public relations: getting your name into the places where people already discuss your category.
The Caveat Most Articles Drop
You will see 0.664 quoted all over the internet as though it were a lever you can pull. Ahrefs published the warning themselves, and most pages repeating the number leave it out.
Correlation is not causation. What the data shows is that businesses with high AI visibility also tend to have broad presence across the web. It does not show that starting a YouTube channel causes citations to appear. The underlying thing being measured is something closer to general prominence, and prominence is not a switch.
We think these numbers are still useful. They tell you where to look. They do not tell you which button to press, and we would rather say that than sell you a correlation as a formula.
Somebody Else’s List Beats Your Own
When Ahrefs looked at the pages ChatGPT cites for early-stage questions like “best web design agencies in London,” recently updated “best of” lists were the most prominent page type by a wide margin. There was also a relationship between placing highly on those third party lists and turning up in the answers themselves.
A detail from the same research explains why the sources sometimes look unfamiliar: 28% of ChatGPT’s most cited pages have no organic search visibility at all. These are not always the polished authority sites you would expect. Which is frustrating, and also an opening, because getting onto a list is a conversation with a publisher rather than a two year link building program.
The uncomfortable part is that your own “best providers in our industry” page, with you at the top, carries almost none of this weight. Third party validation works precisely because you did not write it.
Reviews Carry More Weight Than You Think
For service businesses, reviews are often the highest leverage thing on this entire list, and the most under-managed.
What matters is not only the star average. Volume signals that real people use you. Recency signals that they still do. Detail signals what you are actually good at, and this is where most businesses leave value sitting on the table. A review that says “great service” tells a language model nothing. A review that says the crew arrived when they said they would, found the leak in the flat roof section within an hour, and billed what they quoted tells it everything: what you do, how you work, and who should hire you.
If you are asking customers for reviews with a generic template, you are getting generic reviews. Ask them what problem you solved and you will get answers that do more work.
1
Web Mentions
Your name appearing in context on sites you do not control. The strongest measured signal, and the slowest to build.
2
Third Party Lists
Inclusion in the “best of” and roundup pages that already exist for your category and location.
3
Reviews
Volume, recency, and specific detail across the platforms that matter for your industry.
4
Your Own Site
The floor, not the ceiling. It supplies the facts everything else has to agree with.
The Part Nobody Wants to Say Out Loud
Some of this is fixed, at least for now.
A model’s training reflects a version of the web that already exists. If a competitor has been written about, listed, reviewed, quoted, and linked for eight years and you opened your doors last spring, they hold an advantage that no amount of on-site work erases this quarter. That is not a failure on your part. It is arithmetic.
It also compounds. Being recommended generates coverage, and coverage feeds the next round of recommendations. The gap between the known and the unknown tends to widen on its own.
And there is no paid shortcut through it. There is no ad slot inside ChatGPT’s organic recommendation, no submission form, no placement to buy. You can pay for things that influence the inputs, sponsoring content, running PR, advertising on platforms that get cited. You cannot pay for the output.
None of that is a reason to sit it out. It is a reason to judge this work by whether it builds real presence rather than by whether it promises a shortcut. The businesses winning here started before it was obvious, and the same will be true a year from now.
Be Careful Who You Buy This From
If a vendor offers guaranteed placement in ChatGPT’s answers, a “number one AI ranking,” or a fixed timeline to get there, they are describing a mechanism that does not exist. Ask them instead what they will do to earn mentions on sites you do not own, how they will measure it, and what they will show you in month three. Those questions have real answers, and the answers are worth paying for.
Make Sure ChatGPT Can Read Your Site
Before any of the harder work is worth doing, check that the door is unlocked. This is where we find genuine, fixable problems more often than you would expect.
Three Crawlers, Three Different Jobs
OpenAI runs three separate automated visitors, and they are not interchangeable:
- GPTBot gathers content that may be used to train future models. Blocking it is a decision about training use.
- OAI-SearchBot is the one that governs whether your pages can appear in ChatGPT’s search answers. OpenAI states that sites opted out of it will not be shown there.
- ChatGPT-User fetches a page live when someone’s question sends ChatGPT to look at it.
The common mistake is a well-meaning one. A business decides it does not want its content training an AI model, tells its developer to block the AI bots, and the developer blocks all three. Training use is now declined and so is any chance of appearing in ChatGPT’s search results. These settings are independent. You can allow search visibility while declining training use, and for most businesses that is the sensible default.
Changes take roughly a day to register on OpenAI’s side, so do not judge the result the same afternoon.
The Quieter Blockers
Crawler permissions are not the only way to be unreadable:
- Security rules and rate limits that return errors to crawlers your robots.txt file technically welcomes. This is a frequent and invisible cause of missing citations.
- Key facts trapped in images, PDFs, or scripts instead of plain readable text on the page.
- Services, coverage areas, and pricing described so vaguely that no system could repeat them with any confidence. If a person cannot tell from your site whether you serve their county, neither can a model.
Our free AI Meta Elements Optimizer is a decent place to start on the last one, since titles and descriptions are usually the first place vagueness shows up.
How to Tell If Anything Is Reaching You
Two checks worth running. ChatGPT tags the visits it sends you with a referral marker, so traffic arriving from it can be identified in your analytics rather than guessed at. And your server logs will show which crawlers are actually getting through, which is the only way to catch a security rule quietly turning them away.
How to Check Where You Stand
Asking ChatGPT whether it has heard of you is not a test. It will usually say something agreeable. Here is a check that produces something you can act on.
Write Down Five Real Customer Questions
In their words, not yours. Not “commercial HVAC preventative maintenance contracts” but “who can service the rooftop units on my building.”
Ask Each One in a Fresh Session, Logged Out
Never mention your business name. You want the answer a stranger gets, not the answer shaped by your own chat history.
Run Each Question Three or Four Times
Record who gets named and how often. One appearance is noise. A name showing up in every run is the pattern you are competing against.
Ask What Supported the Answer
Follow up by asking which sources back up the recommendation, and note the domains. These are the pages that matter in your category.
Compare That List Against Where You Appear
The pages that keep coming up, on which you are absent, are your work list. It is usually shorter and more specific than you expect.
Read the result honestly. Appearing occasionally means you are known but not strongly associated with that particular question. Never appearing while the same three competitors always do is the clearest answer this exercise gives you, and it is better to know it than to assume.
Where the Work Pays Off
Ordered by leverage, not by ease. The first two are within reach for most businesses without outside help.
1
Fix Your Own Facts First
State what you do, who you serve, where you serve them, and how you price, in plain readable text. Everything downstream depends on this. If your own site is ambiguous about whether you handle emergency callouts, no amount of external mention work will fix the gap.
2
Build Review Depth, Not Just Review Count
Ask recent customers what problem you solved and how it went, on the platforms that carry weight in your industry. Keep them coming steadily rather than in bursts. Detailed and current beats a high average earned three years ago.
3
Get Named on Other People’s Pages
Find the roundups and “best of” lists that already rank for your category, then pitch inclusion with something concrete to offer. Trade association directories, local and industry publications, supplier partner pages, podcasts, event sponsorships. This is the slow one and the one that compounds hardest.
4
Answer the Questions People Actually Ask
Content built around a real buying constraint outperforms content built around a keyword. The page that answers “can you do this in an occupied building without shutting the floor down” is the page a model can match to a real question.
5
Keep Your Details Consistent Everywhere
Conflicting hours, addresses, and service descriptions across your profiles give every system a reason to hesitate before naming you. This is dull, cheap, and worth doing once properly.
Most of this is the same work that improves how you show up in ordinary search, which is the point. Our generative search services work sits alongside organic search rather than replacing it, because the foundations overlap almost completely.
How Long This Takes
We are not going to give you a number, and we would question anyone who does.
The mechanism explains why. Mentions accumulate at the pace other people publish. Review depth builds at the pace you serve customers. Lists get refreshed on their editors’ schedules, not yours. And the model’s underlying knowledge updates when the next version trains, on a timetable nobody outside OpenAI sets.
What we can tell you is the shape of it. Technical fixes can register within days. Content and review work starts showing up in the narrower, more specific questions before the broad competitive ones. The high value prompts, the ones where three established names have owned the answer for years, take the longest and may not fully turn over. Anyone quoting you a fixed timeline is quoting you a sales number.
See how AI describes your business today.
Get a free visibility check and a clear, plain-English plan to start showing up when AI answers for your customers.
Call or Text
What This Comes Down To
The competitor ChatGPT named is not better at prompting than you are. They have not found a setting. More of the internet vouches for them, in more places, over more years, and the model is repeating what it read.
That is buildable. Slowly, without shortcuts, and with no guarantee attached, but buildable. And the work that builds it, clear facts on your own site, real reviews from real customers, your name in places you do not control, is work that pays whether or not anyone ever asks a chatbot about your industry.
If you want a straight read on where you sit and what would move first, book a free consultation and we will tell you what we see. Including if the answer is that your money belongs somewhere else this year.
