Our team will be out of office on Friday, May 1, 2026. We’ll be back and ready to assist you starting Monday, May 4th.

Why Doesn’t ChatGPT Recommend My Business, and How Do I Fix It?

Contents

Three AI engines were just asked the same 100 questions, they agreed on almost nothing, and this article answers the question that follows: what does a small business actually have to do to get named when someone asks an AI who to hire.


Tracewell Gordon, who builds AI visibility software for local service businesses, said the part out loud that most marketers will not. "More and more people skip Google and ask ChatGPT who to hire. The answer names two or three businesses. If yours is not one of them, you never hear about the job you lost."

Read that again. You never hear about it.

There is no bounce rate for a job you were never considered for. No abandoned cart. No form to review on Monday. The customer asked, got three names, picked one, and you never knew it happened. That is a brand new way to lose money, and almost nobody has instrumentation for it.

So here is the direct answer. ChatGPT probably does not recommend your business for two reasons, and neither one is what you have been sold. First, on most questions it is not searching the live web at all, so there is nothing to win. Second, when it does search, the pages it credits are overwhelmingly product manuals, developer docs and comparison roundups, not the blog posts you were told to publish. You fix the first by understanding which questions actually trigger a search. You fix the second by making sure that when a search does happen, the engines can read your site, and what your site says about you is specific enough to be quoted. That is the whole thesis. The rest of this is evidence and a process.

Key Takeaways

Note for publishing: render this section as a shaded callout box.

  • ChatGPT searched the web on only 34.5 percent of queries as of February 2026, so most answers never touch your website.
  • Three engines asked the same 100 questions shared all three of the same sources only seven times, so there are almost no incumbents to displace.
  • Being named earlier in an answer barely matters to buyers; being described clearly matters roughly twice as much.
  • The highest-leverage fix is the least glamorous: confirm AI crawlers can reach your site and your copy is in the HTML.
  • AI visibility is a snapshot with a date on it, so this is a monthly check, not an annual project.

The Problem: You Are Being Sold a Map of a Country Nobody Has Surveyed

Candidly, I think the AI visibility conversation right now is 10 percent evidence and 90 percent confidence.

Scroll LinkedIn for four minutes and you will find a carousel promising it got a client cited in ChatGPT in 14 days. Read it closely and you will notice what it never includes: the data. No methodology. No question list. No record of which engine, on which date. Julia McCoy made that observation, then did something about it. On November 11 and 12, 2026, she ran 100 plain-language how-to questions about getting work done with AI through ChatGPT, Claude and Perplexity. Fresh session every time. Her own brand deliberately absent. She logged, for each answer, whether the engine searched the web and which sites it linked.

I want to be honest about my own part in this, because I got it wrong in a way that cost me real money.

[PLACEHOLDER, JONATHAN TO REPLACE WITH A REAL INCIDENT: a specific time you optimized a page or a property for AI search and got the order of operations wrong, ideally one where you did the visible work before checking the plumbing. Include roughly when, what you spent in money or calendar time, and the moment you found out. Suggested shape: "In [month year] I had my team rebuild [page] to be AI friendly. We [what you did]. Call it [cost] and [time]. What we never did was [the basic check you skipped]."]

We were decorating a storefront without confirming the street was open. That is the most common mistake I now see in the businesses I coach, and the reason it is so common is that the visible work feels like progress and the plumbing check feels like admin.

Here is the reframe. The absence of evidence is not the problem. It is the opportunity. If the engines had settled on a stable list of trusted sources, you would be facing a decade of catching up to incumbents. They have not settled. McCoy's own summary was blunt: "There are almost no incumbents outside the platforms' own manuals." That is a sentence about a window, and windows close.

The Evidence: Five Numbers That Change What You Should Do Monday

1. Most of the time, the engine is not even looking at the web. Semrush analyzed more than a billion lines of United States clickstream data from October 2024 through February 2026 and found ChatGPT enabled web search on just 34.5 percent of queries, down from 46 percent in late 2024. McCoy found the same pattern from the other direction: ChatGPT skipped searching on 24 of the 73 questions it answered, and the skipped ones were the open-advice questions. Design a logo. Write a YouTube script. Write a sales email that does not feel pushy. Those came from memory. Every question that named a specific tool, platform or rule triggered a search. You do not win advice questions with content. You win the questions where someone is comparing named options.

2. The engines do not agree on who to trust, and the gap is enormous. ChatGPT and Claude both cited sources on 31 of the same questions, yet shared even one website on only six of them. Across all 100 questions, 35 sites appeared on at least two of the three lists. Only seven appeared on all three. Three engines, one question set, three almost unrelated answers about who is credible.

3. Clarity beats position, and it is not close. Semrush surveyed 1,030 United States shoppers in December 2025. Only 20 percent said a brand stood out because it appeared higher or earlier in the AI's answer. But 43 percent said a clearer or more detailed explanation made a brand stand out, 39 percent were moved by price or value context, and 37 percent by a description that fit their specific situation. Stop fighting for first position. Fight to be the business the AI can describe precisely.

4. The academic work says the same thing in colder language. A controlled study titled "What Gets Cited: Competitive GEO in AI Answer Engines" fed exactly two competing sources into a model's context and measured which got cited first, with brands anonymized and source order counterbalanced to separate content effects from position bias. Across six models the researchers ran 252,000 trials. The result: topical relevance and list position were the biggest drivers, while explicit price information and a recent timestamp also helped consistently. Formatting-only edits had almost no effect. Put plainly, an engine picks the page that is obviously about the exact question, says what things cost, and shows a recent date. In McCoy's data the credited pages had a median age of 64 days, and pages opened mid-search skewed under 30 days.

5. And the tactic everybody sold you does essentially nothing. Ahrefs analyzed 137,210 domains. Twenty-eight percent publish an llms.txt file, and 97 percent of those files received zero requests in May 2026. Nothing fetched them at all. Among the remainder that were fetched, AI retrieval bots like OAI-SearchBot and PerplexityBot accounted for 1.1 percent of requests, and on sites without the file, the number of AI bots that came looking was zero. Google's John Mueller called llms.txt "not done for search," a temporary crutch for AI coding tools. Plenty of people paid for one this year. You do not need to.

The demand is not theoretical. Forty-three percent of those surveyed shoppers have already discovered a new brand through an AI tool, and 50 percent have made a purchase after using AI during research. The only question is whether your site is legible when the engine finally goes looking.

The Solution: Access, Answer, Audit

What changed for me was giving up the idea that this is a content problem. It is three problems in a strict order, and most people start on the third. I run it as Access, Answer, Audit.

Access comes first because nothing else matters if it fails. The engines reach your site through named crawlers: GPTBot and OAI-SearchBot for OpenAI, ClaudeBot for Anthropic, PerplexityBot for Perplexity, Google-Extended as Google's opt-out token. If your robots file, firewall or security plugin turns any of them away, every hour you spend on copy is wasted. The other half of access is whether your words are in the HTML or trapped in JavaScript. A business owner commenting on TruLata's launch as Digi Dinos put it better than any consultant has: "crawler access is the part small sites never think to look at." That same owner reported that ChatGPT now sends most of their site visits while Bing barely knows they exist.

Answer is where the research pays off. Based on the citation study and the shopper survey, the page that gets quoted does four things at once. It is unmistakably about one specific question. It states real numbers, including price, instead of "contact us for pricing." It carries a visible, recent date. And it describes who the thing is for in concrete terms. This is the opposite of brand copy. It is closer to writing a product manual about yourself, which, given that 71 of the 182 pages ChatGPT credited were help centers and developer docs, should tell you something.

Audit is what makes this a system instead of a project. You cannot infer your visibility from one engine, because they disagree, or from one month, because the credited pages skew fresh. McCoy told readers to treat her numbers as "a snapshot with a date on it." Build the monthly check now, while it is cheap.

The tool market grew up around this almost overnight. TruLata's AI Visibility offers a free scan that reads your site the way an answer engine does, checking crawler access, whether your copy is in the HTML, and structured data, then asking ChatGPT, Gemini and Claude six buyer-intent questions for your market and showing you all 18 answers. Its paid tier runs weekly across nine engines and scores you on the share of answers that name you, from $399 a month. RankSpot 2.0 connects through an MCP server so Claude or ChatGPT reads Google's top five for a keyword, writes a page to beat them, publishes it, and pitches editors of already-ranking pages from your own mailbox, priced per action at roughly $3 an article and $0.09 a pitch. And OpenAI has opened a Businesses in ChatGPT beta whose Business Hub dashboard shows how often ChatGPT cites your site and how many clicks it sends.

Practical Steps: Do These In This Order

  1. Check crawler access before you touch a single word of copy. Pull up your robots.txt file and your security plugin settings and confirm GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended are not blocked. Then view the page source of your most important page and search for a sentence you know is on it. If it is not in the source, your copy is trapped in JavaScript. Fifteen minutes, and it is the highest-value fifteen minutes in this article.

  2. Run a free baseline scan today, then put a recurring reminder on the calendar. TruLata's free scan takes about 10 seconds and shows you all 18 answers from ChatGPT, Gemini and Claude for six buyer-intent questions in your market. Know its limits: the founder admitted publicly that it does not check Bing indexing today. Use it as a starting line, not a verdict.

  3. Claim your business in ChatGPT this week. OpenAI's Business Hub lets you verify through your business phone, business email, your domain or your Shopify store, then pass an ID check for a verified badge. Once verified you control your description, logo, category and return policy, and stores can push a product feed with live prices and stock. OpenAI lists no fee. Jeff J Hunter's framing is right: "Until last week, what it told them about you was whatever it found lying around."

  4. Rewrite your three most commercially important pages as manuals, not brochures. One page, one specific question, answered in the first paragraph. Put your actual prices on it and a visible last-updated date. Name exactly who it is for and who it is not for. You are optimizing for a machine that rewards topical precision, explicit price and recency, so give it all three.

  5. Use AI to write the description of yourself that AI can quote. Most owners are too close to their business to describe it plainly. Hand that job over, using this structure:

[The Job]
Write the single clearest possible description of my business for someone
who has just asked an AI assistant for a recommendation in my category
and will see only two or three names.

[The Background]
My business is [BUSINESS NAME], a [TYPE OF BUSINESS] serving
[GEOGRAPHIC AREA OR MARKET]. We have been operating since [YEAR].
The specific problems customers hire us to solve are [PROBLEM 1],
[PROBLEM 2] and [PROBLEM 3]. Our pricing works like this: [PRICING].
The customer we serve best is [IDEAL CUSTOMER]. The customer we are
genuinely wrong for is [POOR FIT]. What we do differently from the
other options in our market is [DIFFERENTIATOR].

[The Deliverable]
A description of 120 words or fewer, in plain English, that names a
real price range, states who it is for and who it is not for, and
avoids every superlative and marketing adjective. Written from the
perspective of a [TECHNICAL WRITER] speaking to [A BUSY BUYER
COMPARING THREE OPTIONS]. Then give me a second version of 40 words
or fewer for use as a meta description.

[The Questions]
Ask me any questions you have.
  1. Pick the questions you will track and write them down. Not keywords. Questions, phrased the way a customer would type them, including the comparison questions that reliably trigger a search. Six to ten is plenty. Store them in a document so every monthly check uses the identical list, because the moment the list drifts your data is worthless.

  2. Run those questions across at least three engines every month and log the names that come back. Since only seven sites appeared on all three of McCoy's lists, checking one engine tells you almost nothing about the other two. Track who gets named instead of you. That is your real competitive set in this channel, and it will not match the one in your head.

Frequently Asked Questions

Does claiming my business in ChatGPT make it recommend me?
No, and anyone telling you otherwise is selling something. Jeff J Hunter, who reported on the Business Hub beta, is explicit that claiming "doesn't guarantee ChatGPT recommends you." What it guarantees is that the description, logo, category and, for stores, the prices and stock levels ChatGPT shows all come from you rather than from whatever it scraped.

Do I actually need an llms.txt file?
Almost certainly not. Ahrefs analyzed 137,210 domains and found 97 percent of published llms.txt files received zero requests in May 2026, with AI retrieval bots accounting for just 1.1 percent of requests that did arrive. Google's John Mueller called it a temporary crutch for coding tools. Spend that hour on crawler access instead.

How often should I check my AI visibility?
Monthly. The credited pages in Julia McCoy's 100-question study had a median age of 64 days, and pages the engines opened mid-search skewed under 30 days old. Freshness is weighted heavily enough that a quarterly check will miss most of what happened. McCoy's own advice was to re-run the test monthly, not annually.

Should I stop doing traditional SEO?
Absolutely not, and the data here is unambiguous. Semrush found that over 21 percent of all of ChatGPT's outbound referral traffic goes to Google. Of surveyed shoppers, 77 percent use AI and search engines together, and 68 percent name Google as their primary channel for verifying what an AI told them. Your Google presence is what survives the verification step.

Do I have to buy a tool for this?
No. Start with the free scan, the free Business Hub claim, and a monthly check you run yourself in three browser tabs. Paid tools buy scale and consistency, not access: TruLata runs weekly across nine engines from $399 a month, and RankSpot 2.0 charges per action at roughly $3 an article. Earn your way into those.

The Close

Go back to the thing Gordon said. The answer names two or three businesses, and if yours is not one of them, you never hear about the job you lost.

That is the actual stake. Not traffic. Not rankings. Not a dashboard number. Jobs you will never know you were in the running for, going to someone whose website happened to be legible to a machine on the day a customer asked.

Now sit with this. Three of the most capable AI systems on earth were handed the same 100 questions, and they shared all three of the same trusted sources exactly seven times. Seven. They have not decided who the authorities are in your category, or in almost any category. Forbes never appeared in ChatGPT's results. Wikipedia never appeared. HubSpot showed up three times.

That is a land rush with no incumbents, in a channel where half the country is already researching purchases, and almost every competitor you have is either ignoring it or buying an llms.txt file because someone on LinkedIn told them to. Nobody owns these lists yet. That will not be true in a year.

It will not stay this way. It cannot. Lists stabilize, platforms consolidate, winners calcify, and twelve months from now the businesses that got named in November 2026 will be the defaults the engines reach for out of habit. I have watched this movie with directories, with Google, with app stores, with social platforms. The early window always feels too uncertain to act on, and then it is gone, and the people who moved during the uncertainty own the ground.

So do not outsource this one. Go check whether GPTBot can reach your website. Then ask ChatGPT who the best business in your category is, and look at the names that are not yours.

There are almost no incumbents in AI search right now. This is the last time that will ever be true.


About the Author

Jonathan Mast is the founder of White Beard Strategies, an AI coaching and mentorship company for entrepreneurs, where he helps non-technical business owners turn AI from a novelty into working infrastructure. He is the creator of the Perfect Prompt Framework 2.0 and a frequent speaker on practical AI implementation for small business. He spends most of his week looking at what AI tools actually do versus what they are sold as doing, which is why the first step in this article is the unglamorous one.

About the Author