Why ChatGPT names your competitor: data from 7,915 AI answers
TL;DR. ChatGPT names the brands that appear on the pages its own search returns, and also brands the model already knows, according to a study of 158,099 answers. In 7,915 AI answers that Zaraftis collected for 80 brands between May and October 2026, 46% named at least one competitor and not the brand. For 59 of the 80 brands (74%), AI cited at least one third-party site in five or more answers that did not name the brand, at least one of which named a competitor. The most common type of such site was another company’s own website: it appeared for 48 of the 59 brands and made up 58% of all such sites. Reddit appeared for 34% of the brands, YouTube for 27% and G2 for at most 15%. The engines differ: YouTube showed up next to a competitor on Google AI Overviews for 16 brands and on ChatGPT for none.
Why does ChatGPT recommend your competitor and not you?
ChatGPT names the brands it finds on the pages its own search returns, and it also names brands it already knows. A study of 158,099 AI answers and 468,037 citations found that every page ChatGPT cited was one its own search had retrieved. Which pages the search retrieved explained 32% of the variation in citations; the properties of the pages themselves explained 2.8%. The same study found that ChatGPT writes brands it knows into its searches and sometimes names a brand with no supporting page at all.
So when your competitor is named and you are not, there are two places to look. One is the model’s memory, which is slower to change. A study of six AI models found that the brands named first follow search interest and online conversation about the brand more than brand popularity. The other is the set of pages the engine reads at the moment of the question. Those pages you can find, and some of them you can change.
One answer is not enough to see this. AI lists of brands change from one run to the next: in SparkToro’s test of 2,961 runs, there was less than a 1 in 100 chance that ChatGPT or Google’s AI gave the same list twice. The pattern only shows across many answers, which is why the rest of this article uses 7,915 of them. The mechanism itself is covered in how AI engines decide which brands to cite.
For 3 in 4 brands, AI cites a site where the competitor shows up and the brand does not
In 46% of the 7,915 answers, AI named at least one competitor and did not name the brand. The question is what those answers cite.
For every scan, Zaraftis records these as source gaps: a third-party site that AI cited in at least five answers, where none of those answers named the brand and at least one named a competitor. 59 of the 80 brands (74%) had at least one. Across them there were 544 source gaps.
| Type of site | Brands with this gap (of 59) | Share of all 544 gaps |
|---|---|---|
| Other companies’ own sites | 48 (81%) | 58% |
| Publishers and blogs | 35 (59%) | 12% |
| Forums (mostly Reddit) | 22 (37%) | 4% |
| Marketplaces and stores | 20 (34%) | 12% |
| Video (YouTube) | 16 (27%) | 3% |
| Social networks | 12 (20%) | 3% |
| Review sites | 10 (17%) | 3% |
| Reference and documentation | 9 (15%) | 2% |
| Government and academic | 9 (15%) | 2% |
80 brands, the latest Zaraftis scan for each, run between 18 May and 28 October 2026. 7,915 AI answers with 73,030 citations: ChatGPT 2,302, Google AI Overviews 2,135, Gemini 945, Copilot 759, Google AI Mode 241, and 1,533 Perplexity answers from scans before Zaraftis stopped tracking Perplexity. A site counts as a source gap when AI cited it in at least five answers of a scan, the brand was named in none of those answers, and at least one competitor was named. Each scan keeps its 15 largest gaps, so brands with many gaps are undercounted. Sites are grouped by domain rules for well-known platforms and by a model’s classification for the rest. Most brands are B2B software companies; some are retail and services brands in the UAE. A source gap shows that a site was cited in answers that named a competitor; it does not prove that the site caused the mention.
Other companies’ pages carry the competitor most often
The largest group of source gaps was other companies’ websites. 48 of 59 brands had at least one, and they made up 58% of all gaps.
The most frequent was zapier.com, a source gap for 7 brands, more than any other company site in the data. Next were zendesk.com, elementor.com and wifitalents.com (3 brands each), then monday.com, mixmax.com and pipeline.zoominfo.com (2 each). Most of the rest appeared for one brand only. In the rows we read one by one, many were direct competitors that the brand had not added to its list of tracked competitors, and others were vendors in a nearby category whose pages compare several products.
This fits outside research. In Onely’s study of 31 “best agencies for X” questions, the named agencies’ own sites were 57% of the pages ChatGPT cited, and lists of the best agencies were 22%.
It also means a short competitor list hides part of the problem. If a company that AI treats as your competitor is not on your list, its pages look like neutral third-party sites in a report that sorts sites by type.
Reddit and review sites come lower than the advice says
Much of the advice on this question points to Reddit and review sites; in October 2026, Google’s own AI Overview for “how to get mentioned in ChatGPT” recommended Reddit and cited Reddit, Quora and YouTube. In our data, each of these platforms was a source gap for a minority of brands: Reddit was a source gap for 20 of 59 brands (34%), YouTube for 16 (27%), LinkedIn for 8 (14%) and G2 for at most 9 (15%).
Large studies that count all citations put these platforms near the top. A sponsored study of 107 million answers looked at each engine’s ten most cited sources in IT services. Community sites such as Reddit and YouTube made up about half of them, and review directories such as G2 and Clutch much of the rest. A 2025 study of 6.8 million citations across retail, finance, healthcare and food service found forums were only 2% of them.
The difference comes from what each study counts. A list of an industry’s ten most cited sites will mostly be big shared platforms, because those are cited for everyone. A source gap is counted for one brand, in the answers where its competitor appears. Most of the sites in our data were a gap for one brand only, so they would never reach an industry’s top ten. For your brand, the sites that matter most may be ones that no industry study names.
YouTube counts on Google AI Overviews, not on ChatGPT
The engines cite different kinds of sites next to a competitor. The table shows the share of brands with each type of gap, counting for each engine only the brands that had at least one gap there:
| Type of site | ChatGPT (55 brands) | Google AI Overviews (55 brands) | Copilot (22 brands) |
|---|---|---|---|
| Other companies’ own sites | 78% | 85% | 73% |
| Publishers and blogs | 44% | 49% | 36% |
| Forums | 22% | 38% | 0% |
| Video | 0% | 29% | 9% |
| Review sites | 16% | 15% | 9% |
YouTube was a source gap on Google AI Overviews for 16 brands and on ChatGPT for none. Forums, almost all Reddit, were a gap for 21 brands on AI Overviews and 12 on ChatGPT. This matches the agency study above, where ChatGPT’s own searches returned Reddit threads for every one of the 31 questions, yet it cited a Reddit thread in one answer.
Copilot named a competitor and not the brand in 71% of its answers, the highest rate of any engine. Most Copilot scans in the data are from September and October 2026 and cover a different set of brands, so this number is not directly comparable with the others.
A fix aimed at one engine can miss another. In our data, YouTube never appeared next to a competitor on ChatGPT, so a video made for Google AI Overviews may not reach ChatGPT answers at all. This is one reason why a view of one engine alone misleads.
What should you fix first?
Start with the pages AI cites in the answers where your competitor is named and you are not, and work on the ones you can change.
- Run your buyers’ questions on the engines they use, several times each. One answer is a sample of one.
- List the third-party sites cited in the answers that name a competitor and not you, and count how many answers cite each one. The Zaraftis scan report does steps 1 and 2 for you: it runs your buyers’ questions across the engines and lists these sites as source gaps, with the competitors named and the engines that cited them.
- Check whether each site belongs to a competitor. If AI treats a company as your competitor, add it to the list you track, so its pages stop looking like neutral sources.
- Sort the rest by what you can change: a vendor’s comparison page that accepts submissions, a review profile you have not completed, a publisher you can give data to, a forum thread where you can answer.
- Make your own pages say who you are for, in your buyers’ words. In the agency study, ChatGPT cited the agencies’ own sites more than any other kind of page. The AI-readability checklist covers what to fix on the page.
Start with step 3. It costs the least: it needs only a check of who owns each site, and in the rows we read, many of the sites that looked neutral belonged to competitors the brand had never tracked.
See which sites name your competitors and not you.
The Zaraftis scan report lists your source gaps: the sites AI cites in answers that name a competitor and not you, with the competitors named and the engines that cited each one.