Usually because your competitor is on the pages the assistant reads for your buyer’s question and you are not. When ChatGPT searches, it names companies from review sites, lists, forum threads and company pages that match the words the buyer used. When it answers from memory, it names who the web wrote about most clearly before the model was trained.
That is good news, in a way. It means the gap has a cause you can find, and the cause is rarely that the assistant prefers them. It is that the assistant could read about them, in the right words, in the right places, and could not read about you.
This note gives the four usual reasons, a twenty-minute test to tell which one applies, and what we saw when we asked 120 buyer questions across eight B2B categories.
What are the four usual reasons?
Almost every case comes down to words, pages, memory or access.
Work through them in this order. The first two are the most common and the fastest to fix.
- Different words. Your buyer asks for “compliance automation for a 50-person startup” and your site says “trust infrastructure”. The assistant matches the buyer’s words. Your competitor uses them, so their pages fit the question and yours do not.
- Missing from the cited pages. The assistant searched and read a review site, two roundup articles and a forum thread. Your competitor is on three of them. You are on none.
- A thin footprint in memory. With search off, the model answers from what it learned in training. A competitor written about for years, in many places, is easier to remember than a company with one website and a launch post.
- Pages it cannot reach or understand. Your robots.txt or firewall blocks the search crawler, or the answer to the buyer’s question sits in an image, a PDF or a script the crawler does not read.
None of these is about the assistant liking your competitor more. Each is about what it can read. That is why the fix is always on a page, and never a message to the assistant.
How do you tell which reason applies to you?
Ask the buyer’s question twice, and read the citations.
Twenty minutes, one question at a time.
- Write the question in your buyer’s words, with no company names. Use the question where you see the competitor named.
- Ask it in a fresh ChatGPT chat with web search on. Record who is named and every cited page.
- Ask it again with search off, or in a temporary chat with personalisation off, so the answer leans on what the model learned. Record who is named.
- Open each cited page. Note whether your competitor is on it, whether you are, and what words the page uses for the category.
| What you see | Most likely reason | First fix |
|---|---|---|
| Competitor named with search, cited pages mention them but not you | Missing from the cited pages | Get onto the pages that keep being cited |
| Cited pages use words your site never uses | Different words | Rewrite one page in the buyer’s words |
| Competitor named from memory, you are not, search answers vary | Thin footprint in memory | Say the same true thing about yourself in many places |
| Your own pages never appear, even for questions only you answer | Pages it cannot reach or understand | Check robots.txt, bot protection and text on the page |
Answers vary from one run to the next, so ask the same question on two different days before deciding. A pattern that holds on both days is worth acting on.
What did we see across 120 buyer questions?
No category we checked had a settled shortlist.
On 17 October 2026 we asked 120 buyer questions across eight B2B categories, fifteen per category, from finance operations and legal tech to privacy and cybersecurity. Each was asked once to Claude with web search on, through the API. A company counted only if it appeared in the text of the answer itself.
The answers were crowded and scattered. Each category’s fifteen answers named between 112 and 248 different companies. Across all eight, four in five of those names (1,247 of 1,558) appeared in only one answer, and in every category it was more than two in three. No company was named in more than ten of its category’s fifteen.
When an answer singled out a first pick, the reason it gave was usually fit. Of the 75 answers that named a first pick, 59 gave fit or features as a reason, 24 gave size or reputation, and only 8 pointed to reviews or analyst ratings. An answer can give more than one reason, and a second model sorted the reasons each answer gave into types.
One assistant, one answer per question, one day. It is a baseline, not a ranking. But it says something useful to anyone missing from an answer today. The list is not closed. A competitor named in one answer is often absent from the next, and the reason given for a first pick is usually that the company fit the question as asked.
Across all eight, four in five of those names appeared in only one answer.
What should you fix first?
The reason you found decides the order.
If the words are different, rewrite the one page that should answer the question, using the buyer’s words in the heading and the first sentence. Say who it is for, what it does and how it differs, in text. This is the fastest fix, and it often helps with search answers within weeks.
If you are missing from the cited pages, list those pages and work through them by type. Claim and complete review profiles. Ask real customers for reviews. Pitch the author of the roundup that appears most. Answer questions in the forum under your own name. Our note How to get your business recommended by ChatGPT walks through each type.
If your footprint in memory is thin, the work is slow. Describe yourself the same way on your site, your profiles, your talks and the pages of the people who write about your category. It pays out when a newer model is trained, on a schedule you do not set.
If your pages cannot be reached, fix that today. OpenAI says sites that opt out of its search crawler, OAI-SearchBot, will not be shown in ChatGPT search answers.
The assistant is not choosing your competitor over you. It is naming who it could read about. Make yourself readable in the places it reads.
See the use case for defense and dual-use software: How does a defense or dual-use software company get named when a program office asks an assistant what to pilot?
See the check page for ChatGPT: Does ChatGPT name you when buyers ask?
Questions people ask next
Why does ChatGPT recommend my competitor instead of my company?
Usually because your competitor appears on the pages ChatGPT reads for your buyer’s question, in the words the buyer uses, and you do not. Ask the question with web search on and read the cited pages to find out which pages matter.
Can I ask ChatGPT to stop recommending a competitor?
No. There is no setting or request for that. What you can change is whether you appear on the pages it reads and whether your own pages answer the buyer’s question plainly.
Is it worth fixing if answers change every time?
Yes, because the pages behind the answers change more slowly than the answers do. Being on the pages that keep getting cited raises your odds across many runs, not one.
How long does it take to start being named?
Changes to pages the assistant already cites can show up once those pages are read again. Changes to what a model remembers wait for a newer model. Ask the same questions monthly and compare.
Can Throughline find out why a competitor is named instead of us?
Yes. Request an AI visibility check. A growth strategist reviews how AI assistants see your company and replies by email within two business days.