You find out which competitors ChatGPT recommends by asking it the hiring questions your buyers ask, writing down every business it names, and then asking it to show you the sources behind those names. The second half is the part almost everyone skips, and it is the half that answers the "why." Twenty minutes gets you the list. Another twenty gets you the reasons.
A shortlist of three names looks like a scoreboard. It reads much better as a bibliography. Somebody asked a question, an assistant went looking, and it came back with the businesses it could point to something about. The names tell you who. The sources tell you why.
That reframe matters, because the sting of seeing a familiar competitor is real and the conclusion people draw from it usually isn't. Those names are not a ruling on who does better work. They are a ruling on who could be backed up at the moment somebody asked. Backing is buildable. Better is not something you need to become, because you already are it.
So here is the useful version. How to capture the competitor list properly, how to read the evidence trail behind each name, what the published research says about where those trails actually live, how to sort what you find into three things you can build, and where a free self-run reading stops being enough.
Write Down the Names, Then Ask Where They Came From
Start with the mechanics, quickly, because they are not the interesting part. Ask five questions a buyer would ask, in a chat with no memory of you, without naming your own business, and run each one in a fresh session. The full setup, including why a signed-in test flatters you and how many answers you need before believing any of it, is laid out in the self-check method. Use that article for the conditions. Use this one for what to do with the names.
Then build a sheet with four columns, and note that only the first is about you.
- Who got named. Every business, in the order they appeared. Directories, roundups and association pages count as names too, because a buyer clicks those.
- What the assistant said about them. Copy the sentence, not the gist. "Known for complex cross-border estates" and "another local option" are wildly different outcomes, and the sentence is the thing you will later try to earn.
- What it showed as sources. When an assistant browses, it usually links at least some of the pages it used, though how much it shows varies by product and keeps changing. If nothing is visible, ask plainly: which sources did you use for that answer. Record the domains.
- What kind of page each source was. Their own site, a directory listing, a review thread, a press mention, an industry body, a ranked roundup. This column is the one that turns a grudge into a plan.
It also helps to know that the assistant did research before it answered you. These systems run their own searches first, and those searches are visible when you capture them. In our own run on July 14, 2026, we recorded 129 of those research queries across three engines and sorted them into six clusters. It is a plain look at what a machine goes hunting for before it names anybody. Your competitor did not win an argument. They turned up in a search you never saw.
One honest caveat before you go further. A source shown next to an answer is evidence the assistant retrieved it, not proof that page caused the recommendation. Nobody outside these companies can see the full path. You are reading the visible trail, and the visible trail is still the most useful thing on the page.
The Reason Usually Lives on a Page They Do Not Own
Here is the finding that surprises most owners the first time they run this properly. When you follow the sources behind a competitor's name, you generally do not land on their website.
DerivateX studied this directly in May 2026, running ten test rounds across 40 B2B SaaS categories and capturing 233 ChatGPT recommendations made with web search on. Of the citations backing those recommendations, 11.6 percent pointed to the recommended company's own site and 88.4 percent pointed somewhere else: review platforms, comparison pages, editorial roundups, third-party write-ups. That is software rather than professional services, and it is one vendor's dataset, so treat it as a strong hint about how these answers get assembled rather than a measurement of your market.
The practical consequence is a change of target. If a rival keeps getting named, stop studying their homepage. Ask instead who published the page that named them, and whether that publisher could just as easily name you. Those are two different projects. Only the second one is available to you this quarter.
What counts as a naming page also varies enormously by category, which is why generic AI visibility advice ages badly. DeltaV Digital tracked 25,337 citations across 21,075 responses on five AI surfaces between April and July 2026, covering eight brands in different industries. Own-domain share ranged from 74.7 percent in higher education down to 7.4 percent in consumer automotive, and the B2B technology services cell came back empty. Ranked listicles carried 61 percent of citations in the B2B services set and none at all in the healthcare one. Eight brands is a very small sample, and an empty cell in a small sample is not proof that a category never cites owned pages. Read the study as a demonstration that the mix differs sharply by field, not as a table to copy.
So do not ask what AI rewards. Ask what your category's answers are actually built out of, then read your own capture for the pattern. If every named competitor traces back to two industry roundups and a professional association page, you have just found your next three months of work. If they all trace back to their own deep service pages, you have found something else entirely, and it is cheaper to fix.
This is also the point where owners with strong search rankings get a nasty surprise, because ranking well and being cited are not the same event. That gap deserves its own explanation, and we wrote one for firms whose SEO agency is already doing good work and cannot understand why the assistants disagree.
In Professional Services, Nobody Is Sitting on a Throne
Before you build a campaign around one rival, look at how wide the answer really is.
Parse analyzed 631,087 recorded brand recommendations on its own monitored panel, drawn from ChatGPT and Google AI Overviews across 10,839 buyer questions between October 2025 and April 2026. In consumer categories, the answers concentrated hard around a single winner. In professional services they did the opposite: the typical question named 13 different businesses, and the leading name held a median share of 26 percent. That is not a throne. That is a crowded shortlist where a quarter of the attention goes to whoever is most consistently checkable.
An academic reading points the same way. In a June 2026 preprint, researcher Dmitrij Żatuchin put 250 brand-free questions to three models across five industries, collecting 3,750 responses, and measured recommendation concentration at a Gini coefficient of 0.28, which he describes as moderate rather than winner-takes-all. Two other findings from that paper are worth carrying around. Competitive vacuums, meaning queries where no clear leader emerged, showed up in only 8.0 percent of cases, so the space is rarely standing wide open. And the top-recommended brand agreed across models only 41.6 percent of the time, which means the firm dominating ChatGPT in your category is often not the one dominating Gemini or Perplexity. It is a preprint, not settled consensus, and one author works in this industry, so weigh it accordingly.
Read those two datasets together and the job changes shape. You are not trying to dethrone one competitor. You are trying to become one of the names a machine can confidently include, in a list that already holds more than a dozen, on several engines that do not agree with each other about who belongs there. Inclusion is a far more winnable brief than displacement. It is also the honest one.
Sort Every Name Into One of Three Buckets
Now turn the sheet into a build list. For each competitor that got named, ask which of three things their trail demonstrates, because AI recommends businesses it can verify, that others vouch for, and that know something it did not already have.
The first bucket is verification. Their name, people, credentials, locations and profiles agree with each other everywhere a machine looks, so the assistant is confident they are one specific real business. If your capture shows competitors being named alongside their address, their principal and their credentials while you appear nowhere, you may be losing on identity rather than on evidence. That is the least glamorous of the three, and the fastest to fix.
The second bucket is corroboration. Somebody who is not them said they were good at something specific: a directory with real detail, an association page, a journalist, a genuine review that describes the situation rather than the mood. When you see the same two or three publishers behind several rivals, you are looking at your category's actual referral desk. Nothing about those pages is closed to you.
The third bucket is knowledge nobody else published. This is the competitor whose own page keeps getting quoted because they wrote down the hard case, the real sequence, the thing they tell clients on the phone. It is the slowest bucket to build and the only one a rival cannot copy from you by writing a check. It is also usually where an established expert's advantage has been sitting, unpublished, for fifteen years.
Order matters more than effort here. Publishing brilliant material while a machine still cannot confirm you are one business is a year of work poured into a name it will not risk. Fix what you are, then build who vouches for you, then say what only you know. The full readiness picture runs deeper than three sentences, but that sequence is the part worth writing on the wall.
What the Competitor List Cannot Tell You
Run this reading yourself, for free, as often as you like. It answers one question honestly: which businesses fill the space when your buyers ask, and what kind of evidence sits behind them. That is worth an afternoon of anybody's year.
What it cannot do is prove causation. A cited page might be central to the answer or barely relevant to it, and no visible trail shows you the ranking that happened inside the model. It cannot tell you how stable any of this is, because answers move with wording, week and whether the assistant browsed. It cannot tell you which of your own gaps to close first. And it cannot separate an assistant quoting your content from an assistant recommending you, which is worth understanding before you act, because being a useful source and being the chosen provider run on different evidence.
That is the gap the free Recommendation Check closes. It scores your business across 109 checkpoints in eight zones, hand-scored, and asks the buyer questions of the five engines we track: ChatGPT, Gemini, Perplexity, Claude and Grok. You get the competitor names saved word for word, the sources behind them, and a plain order of repair. There is no discovery call to sit through first, the report does the talking, and the method is published so you can check how it was scored.
If the reading shows the work is ongoing rather than a one-off repair, that is what The Answer exists to operate: the monthly cadence of published expertise, external validation and re-measurement that keeps a business explainable to a machine as its category moves. Either way the boundary stays fixed. AI can't recommend what it can't verify, and nobody controls what a model tells the next stranger who asks. What you control is whether the evidence about you is checkable, corroborated and worth quoting.
Start with the list. It is the cheapest honest information available to you, and it is sitting one question away.
Want to Learn More?
Probably Genius is a done-for-you operation for expert-led firms: we get what you know onto the public record in a form machines can verify, then measure what the engines say about you every month. If a familiar competitor keeps arriving in answers where your name should be, the problem is probably not the work. You're probably a genius at what you do. We make sure AI gets the memo.