My competitors show up in ChatGPT but we don’t. How do I fix that?
Why an assistant names the company next to you and not you, how to find the exact pages it is reading, and the four things that change the answer, in the order that works. With figures from 1,931 buyer-question answers.
Written 14 October 2026 from answers collected 21 September – 3 October · Satayu Isariyaphorn
ChatGPT names your competitor and not you because, for that question, it read your competitor’s pages and did not read yours. With search on, ChatGPT builds a buyer answer from about five pages; Claude from twenty, Perplexity from eleven, Google’s AI Mode from four. If none of those pages names you, you are not in the answer. The fix is to be on the pages it reads for that question: one page on your own site that answers the question in its first paragraph, and the directory listings, comparison articles and forum threads it already cites. Check which pages those are before you write anything, and measure on three separate days, because the set of companies an assistant names is the same on two days only 55% of the time.
First, confirm it is a pattern and not one bad day
Ask the question a buyer would type, with your country in it, in a new chat with web search on. No follow-ups. Do it again on two more days with the same wording. Across 1,800 of our answers the same question on two separate days named the same set of companies 55% of the time overall, and only 40% on Google’s AI Mode (the figures). A competitor named once is noise. A competitor named in nine of nine answers while you are named in none is the pattern this page is about.
Why they are in and you are out
The assistants do not keep a ranking of software companies. When a buyer asks “what is the best rostering software for a café in Australia?”, each one searches, reads a handful of pages, and names the companies on those pages. Your competitor is named because it is on them. The usual reasons you are not:
- No page of yours answers that question. Your site describes your product. The question is about the buyer’s problem and place. If no page on your site says, in its first paragraph, which rostering software suits a small café in Australia and why, there is nothing for the assistant to lift. In our data the companies’ own pages are cited heavily (peakflo.co 35 times in 60 answers, rosterelf.com and vanta.com among Claude’s most-cited sites), but only the pages that read like an answer.
- You are missing from the directories Claude and Perplexity read. G2, Capterra, SourceForge, GetApp, Software Advice and TrustRadius drew 1,067 citations in our answers, almost all from Claude and Perplexity. A free listing with the category and country filled in puts you on the page they open; an empty or wrong-category listing does not.
- You are missing from the threads and videos Google reads. Google’s AI Mode cited Reddit 85 times and YouTube 69 times; the other three assistants cited Reddit nine times and YouTube never. If your competitor has been recommended in a Reddit thread that matches the question, Google will keep reading that thread.
- The comparison articles already exist, and you are not in them. “Deputy vs Planday” pages name Deputy and Planday. On head-to-head questions between two rivals, the company measured was named in 3 of 12 answers or fewer in most of our 21 reports. You cannot win a question you are not in; you can be in the “alternatives to X” pages that sit next to it.
Find the exact pages before you fix anything
Every assistant now shows its sources: chips and a “Sources” panel in ChatGPT, a list under the answer in Claude, numbered links in Perplexity, link cards beside Google’s AI Mode. Open them for each answer that named your competitor and not you, and write the pages down. After three days you have a short list of real pages: this G2 category, that Reddit thread, these two comparison articles, your competitor’s pricing page. That list is the whole job. What each assistant reads has the full breakdown by platform.
The four fixes, in the order that works
| Fix | Which assistant it moves | What it looks like |
|---|---|---|
| One page that answers the question | All four | The buyer’s question as the title, the answer in the first paragraph, your country named, a plain comparison with the two companies the assistants already name. Not a feature page. |
| Complete the directory listings being cited | Claude, Perplexity | G2, Capterra (the country site), SourceForge, GetApp: right category, country, pricing filled in, reviews only from real customers you ask yourselves. |
| Answer the thread Google is reading | Google AI Mode | Find the Reddit thread or YouTube video it cites, and answer the question there honestly, naming yourself once. Not a pitch. |
| Get into the comparison pages | ChatGPT, Perplexity | The independent “best X for Y” and “alternatives to X” articles it cites: offer the writer data or a customer story. Your own “alternatives to [leader]” page, honest about when the leader fits. |
Do the first one before the others. In every one of our 21 reports the first fix was a page shaped like the question the company was losing, because that is what all four assistants cite when it exists, and nothing else helps an assistant that has nothing of yours to read.
What not to do
Do not stuff your product name into pages that do not answer the question; the assistants cite pages for what they say, not for the names in them. Do not buy reviews or have them written: directories remove them and the assistants read the directory, not you. Do not ask the assistant “what about us?” as a follow-up and count that as being named; you have told it who you are. And do not measure once. Re-ask the same questions a month after the fixes, three days, same wording, and compare the count.
If you want it done
The free first report asks five buyer questions in your category to two assistants, 30 answers, and shows you which companies were named and which pages were read, so you can see the gap before deciding anything. The $390 audit does forty questions on all four assistants across three days and lists the fixes with the page behind each one. The method and counting rules are public, so you can run the check yourself with the steps on how to check Claude.