ChatGPT recommends your competitor because it can verify them and it cannot verify you. An assistant names a business when it finds the same facts about it in several places it can read, and when those facts match the question. Break any link in that chain and it names someone else, even if you rank above them on Google.
Most of those broken links are fixable, and the first few can be fixed this week. Below are the seven I find most often, how to check each one yourself, and the order to fix them in.
How ChatGPT decides who to name
ChatGPT knows about your business in two ways. The first is its training snapshot, a picture of the web from months ago that changes slowly. The second is live search: when a question needs current or local information, it fetches pages during the conversation and writes its answer from them.
Live search is where most local and service recommendations come from. OpenAI runs a separate crawler for it, OAI-SearchBot, and blocking it quietly removes you from those answers. The difference is explained in GPTBot, OAI-SearchBot and ChatGPT-User: What Each One Actually Does.
This matters because the audience is enormous. ChatGPT passed 900 million weekly users in February 2026, and when one of them asks for a business, the answer usually holds two or three names.
Here is how that plays out, in a made-up but typical example. Two plumbing companies in Austin rank side by side on Google. One states "24/7 emergency service across Austin and Round Rock" in plain text, uses the same name and phone number on every listing, and has reviews mentioning burst pipes fixed at night. The other has a polished homepage, a service-area map saved as an image and an old phone number on Yelp. Ask ChatGPT for an emergency plumber in Austin, and only the first can be verified, so the first is far more likely to be named.
1. Its crawler cannot reach your site
The most common cause is also the most boring. A firewall rule, a "block AI bots" toggle in Cloudflare, or one line in robots.txt stops OAI-SearchBot at the door. An assistant cannot cite a page it was never allowed to read.
Check: run the four checks in How to Check Whether Your Site Is Blocking AI Crawlers. It takes under five minutes.
Fix: allow the search and user-triggered agents. A copy-and-paste version is in How to Allow AI Crawlers in robots.txt.
2. Your key facts live in images, PDFs or scripts
Your insurance list is a strip of logos. Your service area is a map image. Your menu is a PDF, and your hours load from a widget. People can see all of it; a model reading the page often cannot.
Check: open your page, press Ctrl+U to view the source, and search for your phone number, your city and one service name. If they are not in the raw HTML, assume the assistant cannot see them either.
Fix: write every fact a buyer filters by as plain text on the page: services, areas served, insurance or payment accepted, hours and price ranges.
3. You look like two different businesses
"Smith Family Dental" on Google, "Smith Dental PLLC" on Yelp, an old suite number on Facebook, a different phone number on the BBB. You know these are one business. A model sees conflicting records and trusts all of them a little less, while your competitor, with one name and one address everywhere, looks safer to recommend.
Check: search your business name and go through the first 20 results. Write down every version of your name, address, phone number and hours you find.
Fix: pick one canonical version and correct every listing to match it exactly. Then add Organization or LocalBusiness schema with a complete sameAs list, as described in The sameAs Property.
4. Nobody else is talking about you
Assistants trust what other people say about you more than what you say about yourself. Your competitor might appear in a local news story, a "best of" roundup, an association directory and a few Reddit threads. You appear only on your own site.
Check: search Google for your business name in quotes followed by
-site:yourdomain.com. Count the independent pages that mention you by name.
Fix: earn real mentions: local press, chamber of commerce and association listings, partner and supplier pages, podcast guest spots, sponsorships with a public page. This is the slow part, measured in months, and it is also the part a competitor cannot copy quickly.
5. Your site never states the answer to the question
Buyers rarely ask for "a good plumber". They ask for a plumber in a named city who takes emergency calls on weekends. If your site never says "24/7 emergency service in [city], including weekends", the model cannot match you to that question, however good you are.
Check: write down the five questions your best customers asked before hiring you. Search your own site for the exact condition in each one.
Fix: state each condition in one plain sentence near the top of the relevant page. Answer engines tend to lift the sentence that answers the question directly, so put that sentence first.
6. Your competitor appears in the lists the model reads
When someone asks for "the best" of anything, assistants lean on roundup articles, directories and comparison pages. If those lists name your competitor and not you, the answer follows the lists.
Check: ask ChatGPT your buyer's question and open the sources it cites. Note which lists and directories keep appearing.
Fix: get onto the credible ones honestly. Claim your directory profiles, pitch roundup authors with something genuinely new, and keep every listing identical to your site. Do not publish a fake "Top 10" that ranks you first; it is easy to spot and it undermines everything else you publish.
7. Your reviews are thin, old or vague
Reviews are third-party evidence, and the words inside them matter. "Great service!" tells a model nothing. A review saying the clinic fitted in a same-day emergency appointment gives it a fact to match against a question.
Check: read your last 20 Google reviews. How many mention a specific service, situation or result, and how old is the newest one?
Fix: ask every happy customer for a review and suggest they mention what you actually did for them. Never write reviews on a customer's behalf, pay for them or filter out unhappy customers before asking; fake and bought reviews break FTC rules and every platform's policy.
Which to fix first
Fix in order of speed. Live search reads your pages at the moment it answers, so access and on-page fixes can show within days. The model's background picture of you depends on references building up across the web, which takes three to six months.
| Reason | Who usually fixes it | Typical time to show up |
|---|---|---|
| Crawler blocked | Your web person, under an hour | Days |
| Site never states the answer | You, one page per question | Days to weeks |
| Facts hidden in images or scripts | Your web person, a day or two | Days to weeks |
| Inconsistent listings | You or an assistant, about a week | Weeks |
| Missing from lists and directories | You, ongoing | Weeks to months |
| Thin or vague reviews | Your front desk, ongoing | Months |
| No independent mentions | You, ongoing | Three to six months |
What not to do
- Do not buy links or mentions. Answer engines discount them, and they put your Google rankings at risk.
- Do not publish forty near-identical city pages with the city name swapped. Engines discount them almost immediately.
- Do not pay anyone who guarantees placement in ChatGPT. Nobody controls what a model outputs.
- Do not expect ads to fix it. ChatGPT began testing ads in the US in early 2026, but OpenAI says ads sit below the answer, carry a label and do not influence which businesses the answer names.
Frequently asked questions
If I rank first on Google, why doesn't ChatGPT mention me?
Ranking and recommending use different evidence. Google ranks pages; ChatGPT names businesses it can verify across several sources. A strong page with weak outside evidence can rank well and still be left out of the answer.
Can I pay ChatGPT to recommend my business?
Not inside the answer. OpenAI's stated ad principles say ads do not influence responses and are always shown separately, with a label.
Why does ChatGPT give a different answer every time I ask?
Answers vary between runs, accounts and locations, so one test tells you very little. Run the same question several times, in more than one assistant, before drawing conclusions. How to Check What AI Says About Your Business walks through a reliable method.
How long until ChatGPT starts recommending my business?
Fixes that affect live search can show within days. Changes to the model's background knowledge take three to six months, because they depend on new references appearing across the web.
Does this apply to Perplexity, Gemini and Google AI Overviews too?
Mostly, yes. Each engine retrieves sources differently, but all of them reward crawlable pages, consistent facts and independent mentions.
Find out which of the seven is costing you
You can work through this list yourself with the checks above. If you would rather see it done for your business, the $49 audit runs 25 real buyer questions across ChatGPT, Perplexity, Gemini and Google AI Overviews, sends you the dated transcripts, and lists the fixes in priority order. It is credited in full against your first month if you go ahead. Get the $49 audit.
Sources: ChatGPT weekly users, Wikipedia OpenAI ad testing and answer independence, CDO Magazine