ChatGPT usually recommends a competitor over you for one of three kinds of reason. It answered from memory, without searching. It searched but couldn't reach, read or find your pages. Or the pages it found, often other sites' lists and reviews, favour your competitor. The fixes below are most of how to get ChatGPT to recommend your business.
Start with the first one, because the brands in ChatGPT's answers change between runs, and your competitor may not be as far ahead as one answer made it look.
I read six pages that Google ranks or cites for this question on 25 September 2026 (US, desktop). None says how strong the evidence behind its reasons is, and none mentions that the answers vary between runs. So each reason here is graded by its evidence: the engine maker's own documentation, a published measurement, a correlation, a lab test, or a lead that makes sense but that I haven't seen tested. The list is built from the diagnostics in Shruwd, the tool we are building, reordered for checking by hand.
Find out what the answers say about you.
One measurement, your own crawler data, and the three access findings that matter most. Free, no card, about two minutes.
Start freeThe 10 reasons at a glance
| # | Reason | How to check | Evidence |
|---|---|---|---|
| 1 | It may name them less often than it seemed | Ask the same question in 10 new chats and count | Measured |
| 2 | It answered from memory, without searching | Look for links to sources in the answer | Documented |
| 3 | OpenAI's search crawler can't get in | Read your robots.txt and your CDN's bot settings | Documented |
| 4 | Your page is blank until JavaScript runs | View the page source and search it for a price | Measured, in 2024 |
| 5 | Your page doesn't rank for the searches it runs | Search the question, and narrower versions of it, in Bing | Measured, small sample |
| 6 | The pages it reads list them, not you | Search each cited page for both names | Correlation |
| 7 | It reads your page and names them anyway | Look for your own site among the sources | A lead |
| 8 | Their page answers more of the question | Compare their section headings with yours | A lead |
| 9 | Your page looks older than theirs | Compare the dates on both pages | Correlation |
| 10 | Their page is easier to quote | Look for a direct answer near the top | Lab test only |
The first five take a few minutes each. Below them, the evidence is mostly correlations and leads.
When I checked, Google's AI Overview for this question suggested asking ChatGPT why it recommends them. The reply is another generated answer, and it can sound like a reason without being one. When Anthropic planted hints in quiz questions in 2025, and the model used a hint, Claude 3.7 Sonnet's reasoning mentioned it 25% of the time and DeepSeek R1's 39%. That was other models, on multiple-choice quizzes, in reasoning written while answering, not ChatGPT explaining a recommendation afterwards. Treat its explanation as a list of things to check, not a diagnosis.
1. It may name them less often than it seemed
One answer is a sample of one. In late 2025, SparkToro and Gumshoe had volunteers, each on their own account, run the same 12 prompts 2,961 times across ChatGPT, Claude and Google's AI. For ChatGPT or Google's AI, they put the chance of any two responses listing the same brands at under 1 in 100. Some of that spread may come from differences between the accounts.
To check: ask your question in ten new chats and write down every company named. Ten runs is still a small sample. Say your competitor comes up in 6 of the 10 and you come up in 3. The ranges those counts support are about 31% to 83% for them and 11% to 60% for you. They overlap, so counts like these can't show that ChatGPT prefers them. Even 0 of 10 doesn't rule out a true rate near 28%.
The fix: if their range still sits above yours, with no overlap, after 20 or 30 runs, carry on down this list. If no company comes up in most answers, nobody owns the question yet, and that's an opening more than a loss. How to measure AI visibility works through how many runs it takes before a number means something.
2. It answered from memory, without searching
ChatGPT doesn't search the web for every question. OpenAI's 2024 announcement of ChatGPT search put it this way: "ChatGPT will choose to search the web based on what you ask." Profound, which sells an AI visibility tracker, put the share of ChatGPT conversations with at least one web search at about 18% in February 2026.
To check: look at the answer. OpenAI says an answer built from a search carries links to its sources. If there are none, it probably answered from what it absorbed about your market in training, where your competitor likely came up more often than you.
The fix: probably the slowest on this list, because an answer from training can't see what you changed this week. What seems to count is how often other sites mention you, which is reason 6 again, over a longer time. LLM SEO goes through the studies behind that, and why a correlation isn't a promise.
3. OpenAI's search crawler can't get in
A block is quick to rule out. OpenAI's crawler documentation says a site that shuts out OAI-SearchBot, the crawler that collects pages for ChatGPT's search, drops out of ChatGPT's search answers, apart from navigational links. Its training crawler, GPTBot, is set separately.
To check: load yoursite.com/robots.txt and read the group that names OAI-SearchBot, or the User-agent: * group if none does, for a Disallow that covers your key pages. The free crawler access check on Shruwd's site reads the file for OpenAI's, Anthropic's, Perplexity's and Google's AI crawlers, and tells a block on a training crawler, often deliberate, from one that costs you citations. It reads the rules for your homepage only. The AI crawler list says what each of those crawlers is for.
Then look at the CDN or firewall in front of your site, which can refuse a crawler that robots.txt allows. On Cloudflare, AI search crawlers have had a setting of their own since 15 September 2026, and Cloudflare Block AI Bots shows where to see what it turns away.
The fix: undo the rule or setting you didn't intend. OpenAI says its systems can need about 24 hours to adjust to a robots.txt change.
4. Your page is blank until JavaScript runs
If your page builds its content in the browser, OpenAI's crawlers may get an empty frame. In Vercel and MERJ's analysis of AI crawler requests on Vercel's hosting network, published in December 2024, OpenAI's crawlers fetched JavaScript files without executing them.
To check: open the page source (Ctrl+U) and search it for a price or a plan name, not only the headline. A page that's blank without JavaScript is easy to spot. The partial version is easier to miss: most of the page is in the source, but the part a buying question needs, such as prices, plan limits, a comparison table or reviews, loads afterwards. That page passes a blank-page test and still leaves out the section that mattered.
The fix: render that content on the server, or generate the page ahead of time. The free check above counts the text your homepage carries before any JavaScript runs. That catches a blank homepage, not the partial case.
5. Your page doesn't rank for the searches it runs
If your page isn't in what ChatGPT's searches return, the answer is unlikely to cite it. One question can become several searches: OpenAI's documentation for its API describes the model searching, reading the results and deciding "whether to keep searching", and Google calls the idea query fan-out.
To check: search your buyer's question, then the narrower versions a model might run: your category with "pricing", "for small teams", "alternatives", or "vs" and your competitor's name. Run them in Bing, not only Google. Of the pages ChatGPT cited for 100 questions in February 2025, Seer Interactive found 87% among Bing's first 20 results for the same question, and called the result directional.
The fix: if your page is missing from the first two pages for most of those searches, work on ranking for them, starting with whether Bing has indexed it. There's more on where ChatGPT searches in How to rank in ChatGPT.
6. The pages it reads list them, not you
Search for the best tool for a job, and the pages that come back are often lists and reviews on other people's sites. If your competitor is on those pages and you aren't, an answer built from them passes that on.
Some of those lists are written by competitors. Of the 110 top-10 places across 11 buying searches in my own category, run once each on Google on 18 September 2026, 50 went to "best tools" lists on vendors' blogs. That was Google, not ChatGPT, and one category. It still shows how much of what a search returns can come from companies selling against you.
To check: ask the question, open every source the answer links to, and search each page for your name and your competitor's. For a long list of sources, this PowerShell counts both names on each page:
$you = "YourBrand"
$them = "TheirBrand"
$urls = @(
"https://example.com/best-tools-for-x",
"https://example.org/x-reviews"
)
foreach ($u in $urls) {
try {
$html = (Invoke-WebRequest -Uri $u -UseBasicParsing -TimeoutSec 30).Content
$text = $html -replace '(?s)<script.*?</script>|<style.*?</style>', ' ' -replace '<[^>]+>', ' '
$text = [System.Net.WebUtility]::HtmlDecode($text)
$counts = foreach ($name in $you, $them) {
"{0}: {1}" -f $name, [regex]::Matches($text, "\b$([regex]::Escape($name))\b", 'IgnoreCase').Count
}
"{0}`n {1}" -f $u, ($counts -join " ")
} catch {
"{0}`n couldn't load: {1}" -f $u, $_.Exception.Message
}
}
It counts names in the HTML each site sends. A 0 can mean the name isn't there, or that the page builds its list with JavaScript. Some review sites refuse scripts outright, which shows as "couldn't load". Open those in a browser. A low count can also be a mention in a sidebar or footer, which isn't a listing, so open those too. And if your brand name is also an everyday word, the count includes those uses.
The fix: start with the pages that name them and not you. Ask the editor to include you, submit to the review sites, and take part in the forum threads that keep getting cited. Ahrefs' study of 75,000 brands, ChatGPT included, found that mentions on other sites went along with AI visibility far more closely than links did. That's a correlation, and I haven't found a controlled test showing that getting onto the cited pages changes the answer.
7. It reads your page and names them anyway
Your own site can be among the sources while the answer names someone else. Look for your domain in the answer's links. Two patterns can cause it.
The passage ChatGPT used gives advice without naming your product. The answer takes the advice and names companies from its other sources. The fix is on that page: in the passage closest to the question, say plainly what your product does about it, by name. When ChatGPT already reads your page, this is where I would start to get it to mention your business.
Or it cited the wrong page. A how-to post of yours gets cited for "which tool should I use", while the page that answers that question, your pricing or comparison page, doesn't. The fix: make the right page answer the question well, and link to it from the page being cited.
Neither applies to "what is X" questions, where naming a vendor would make the answer worse. The symptom shows in the answer itself. Whether the fix moves the answer hasn't been tested, as far as I know.
8. Their page answers more of the question
To check: open the competitor page that ChatGPT cites and list its section headings. Do the same for your closest page. Pricing, integrations, who it's for, what it doesn't do: each question their page answers and yours skips is a gap. Read the sections as well as the headings, because two pages can cover the same ground under different words.
Why it might matter: when one question turns into several searches (reason 5), a page that covers more of them can turn up in more results. I haven't seen it measured.
The fix: answer the missing questions on your page, with your own facts.
9. Your page looks older than theirs
Ahrefs analysed about 17 million cited URLs in 2025. Going by when its crawler first saw each page, pages cited by AI assistants were about a year newer on average than the pages in Google's organic results, and ChatGPT's citations showed the widest gap. Google's AI Overviews were the exception. The cited pages were still 2.9 years old on average, and Ahrefs calls freshness "one among many factors".
Seer Interactive looked at pages cited by ChatGPT, Gemini and Perplexity for four brands in 2026. 75% had been updated in the last year, and Seer reads that freshness as coming from updates, not from new pages. Both studies are observational, so neither can show that a date causes a citation.
To check: find the published or "last updated" date on your page and on the competitor pages being cited.
The fix: if yours is years older and theirs are recent, refresh it. Update the facts and the date together. A new date on old facts leaves the facts an answer would quote as old as they were.
10. Their page is easier to quote
This comes last because its main evidence is one lab test.
To check: look at the first 1,200 characters of your page, roughly the first two paragraphs. That's the window Shruwd's own check uses, not a known ChatGPT limit. Is there a heading in the question's own words, with a direct answer of a few sentences under it? Does the page give prices, numbers, dates and names, or general claims? Compare it with the page that gets cited.
The lab test is a 2024 research paper, the one that named "generative engine optimization". In a test engine the authors built, pages rewritten by a language model to add statistics and quotations made up more of the answers. Google says its AI features don't call for any special writing style. Schema markup doesn't look like the missing piece either, going by the tests covered in LLM SEO.
The fix: put the answer first and use real facts. Do it for the person reading, and treat any effect on ChatGPT as a bonus.
How to get ChatGPT to recommend your business
Work down the list and deal with the first reason that fits before moving on. By the three kinds of reason at the top:
- First, make sure the gap is real, with 20 or more runs of the question.
- If it couldn't reach, read or find you: undo any block on
OAI-SearchBot, put your content in the HTML your pages send, and rank for the searches it runs, on Bing and not only Google. - If the pages it found favour your competitor: get onto the lists and reviews it cites, and name your product where your own cited pages answer the question.
- If it answered from memory: expect that part to move last, as mentions of you build up across the web.
Then rerun the same questions, as many times as before, and compare the ranges.
What Shruwd checks
Shruwd runs a fixed set of the questions your buyers ask, on Google AI Overviews and on ChatGPT, where it uses OpenAI's API in place of the app. It shows how often you and each competitor you track come up, as a range, never a single score. It also reads your server logs for AI crawler visits and checks them against the IP ranges the vendors publish. Most of the ten reasons have a matching finding, which names the page and the change to make, and Shruwd measures again after you mark it done.
Frequently asked questions
Why does ChatGPT recommend my competitor and not me?
Usually for one of three kinds of reason. It answered without searching the web, so what it learned in training decided the answer. It searched but couldn't reach, read or find your pages. Or the pages it read, often lists and reviews on other sites, favour your competitor. Ask the question several times before deciding which, because the brands in ChatGPT's answers change between runs.
How do I get ChatGPT to mention my business?
Get onto the pages it cites for your buyers' questions, and make sure your own pages can be read without JavaScript and name your product where they answer the question. If your page is already among the sources and you're still not named, the passage it uses probably doesn't say what your product does.
Can you pay ChatGPT to recommend your business?
Not into its answers, according to OpenAI. ChatGPT does carry ads, which OpenAI says are labelled and kept apart from the answers, and its ads announcement states: "Advertising does not influence the answers ChatGPT provides."
How many times should I ask ChatGPT before I trust the result?
More than ten. Ten runs of one question gives ranges up to about 50 points wide, so 6 in 10 against 3 in 10 can still be chance. Twenty or thirty runs narrows it. Use a new chat each time.
