Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- AI Search
ChatGPT names your brand but links to a competitor’s comparison page as the source. Is that a content gap, an entity problem, or just how retrieval works?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
In most cases this is retrieval working as designed, and the part you can act on is a content and coverage gap. It is the least likely to be an entity problem: if ChatGPT names your brand correctly and puts it in the right category, it already knows what you are. What it lacks is a document of yours (or an independent one about you) that matches the shape of the question well enough to be retrieved and cited.
The qualification is that “names you but cites a competitor” can hide three different situations, and they need different fixes. The fastest way to tell them apart is to look at which sentence the citation is attached to and what the competitor’s page actually says about you.
Why the mention and the citation come from different places
A ChatGPT answer that uses search is assembled from two sources of knowledge that are easy to conflate:
- What the model already knows. Brand names, categories and rough positioning learned during training. This is why a brand can be mentioned with no citation supporting that mention.
- What it retrieves for this prompt. Web pages found through search at answer time. Citations point to these documents, and they are attached to the statements those documents support.
OpenAI’s crawler documentation separates GPTBot (used for training) from OAI-SearchBot (used for ChatGPT search). OpenAI’s ChatGPT search help page says ChatGPT search sometimes partners with third-party search providers and, when it does, typically rewrites the prompt into one or more targeted queries, then may send more specific ones after reviewing the first results. OpenAI does not publish how it chooses which retrieved page to cite. That rewriting step explains the pattern well.
Think of it like a research assistant who already knows the market, is asked “which is better for a small team, you or Competitor X?”, and goes looking for a document to quote. The assistant can name you from memory, but the footnote goes to whatever page most directly compares the two. If the only such page was written by Competitor X, that is the footnote.
There is a second route worth checking. The competitor’s comparison page almost certainly names you in its title and body. In that case the page is not just the citation, it may be the actual source of your mention in that answer, and ChatGPT may be summarising the competitor’s framing of your pricing, limits or weaknesses.
Important distinction:
Being named is evidence of entity recognition. Being cited is evidence of retrieval and document fit. A brand can do well on the first and poorly on the second, and fixing one does not fix the other.
Which of the three situations you are in
What you see Most likely explanation How to confirm What to do The citation sits on a sentence comparing you with the competitor. Retrieval matched a comparison-shaped document. You have no retrievable page (yours or independent) that answers the comparison. Search the likely sub-queries (“[you] vs [them]”, “[you] alternatives”, “best [category] for [use case]”) in Google and Bing. Note who ranks. Publish a fair comparison or category page, and pursue inclusion in independent comparisons. The citation sits on a factual claim about you (price, features, limits) that comes from the competitor’s page. Your own facts are not stated plainly anywhere retrievable, or are out of date on the competitor’s page. Check whether your pricing and feature details are in server-rendered HTML text, not images, PDFs or client-side widgets. Check server logs for OAI-SearchBot requests to those pages. Publish clear, current fact pages. Allow OAI-SearchBot if it is blocked. ChatGPT describes you wrongly or confuses you with another company, whatever it cites. A genuine entity or consistency problem. Compare how you are described on your site, major profiles, directories and review sites. Align the core description (what you are, who for, where) across those places. Different citations on every run. Normal variability in generated answers and retrieval. Run the same prompt five to ten times in fresh temporary chats. Report citation share across runs, not a single screenshot. What the citation data says about comparison pages
The published evidence here is directional, not conclusive. One of the larger public datasets is Wix’s AI Search Lab study of about 1.06 million citations across 75,000 answers from ChatGPT, Google AI Mode and Perplexity. It found listicles were the most cited content type at 21.9%, while comparison and alternative pages made up less than 3% of citations across all verticals. In professional services, third-party listicles took 80.9% of listicle citations and self-promotional ones 19.1%. The study does not state its time period, and it kept transactional prompts non-branded, so it says little directly about branded “you vs them” prompts.
Two practical readings follow, both hypotheses rather than proven rules:
- Comparison pages are not a dominant citation type overall. When one wins, it is usually because the prompt was explicitly comparative and few documents matched it. That is a gap you can close.
- Where independent evaluations exist, they appear to be cited far more than brand-authored ones. A self-serving “us vs them” page may be retrieved and still lose the citation to a neutral review, which suggests independent coverage is at least as important as your own page.
Why this is worth fixing, and how much
The business risk is less about traffic volume and more about who controls the description of you at the moment of comparison. A user who clicks the citation lands on your competitor’s page, reading their version of your pricing and weaknesses. BrightEdge data puts ChatGPT at about 95.1% of tracked AI referral traffic, though AI remains a small share of total referrals and the methodology is not detailed. For a B2B product with a long consideration cycle, a handful of comparison-stage visits can matter more than the count suggests. For a low-consideration product, it may not justify much work.
A diagnostic sequence I would run first
- Build a panel of 20 to 40 prompts that real buyers would use: direct comparisons, “alternatives to”, “best for [use case]”, and pricing questions. Keep a control set of non-comparative prompts about your category.
- Run each prompt several times in logged-out or temporary sessions, since memory and personalisation can change results. Record whether you are named, which URLs are cited, and which sentences they support.
- Read the competitor’s page closely. List every claim it makes about you and check which are wrong or out of date.
- Check access: robots.txt rules for OAI-SearchBot, and log evidence that it fetches your key pages. Confirm the important facts are in the initial HTML.
- Search the likely sub-queries in Google and Bing. If you do not rank for “[you] vs [them]” or “[you] alternatives” in traditional search, it would be surprising for ChatGPT to find you through retrieval that draws on web search.
- Check GA4’s AI Assistant default channel group for ChatGPT sessions landing on your comparison-stage pages, remembering that referrer-less visits still fall into Direct.
What to build, in order of likely payoff
A fair comparison page
Name the competitor, state where they are the better fit, and give checkable facts (pricing tiers, limits, integrations) with a last-updated date. A page that admits trade-offs is more useful to readers and harder to dismiss as marketing.
Plain fact pages
Pricing, plans, limits and supported use cases in crawlable text. These give retrieval something of yours to cite for factual claims about you.
Independent coverage
Inclusion in neutral comparisons and category round-ups that already rank for the sub-queries. This is slower, but it matches the kind of source the citation data suggests is preferred.
Watch for this:
Do not chase this with tactics aimed at the model rather than the reader: seeded mentions, near-duplicate “vs” pages for every competitor, or hidden text. None of them has credible evidence behind it, and Google’s own AI guidance says seeking inauthentic mentions is not helpful. Nothing guarantees a citation in ChatGPT.
How to tell whether the fix worked
Treat it as a before and after test on the prompt panel. The hypothesis is that publishing a fair comparison page, once it is crawled and ranking for the sub-queries, raises the share of comparison prompts that cite your domain. The primary metric is citation share for your domain across repeated runs of the comparison prompts. Secondary metrics are mention accuracy, the share of citations going to competitor-owned pages, and AI Assistant sessions to the new page. Allow four to eight weeks, re-run the control prompts to catch model or retrieval changes that affect everything, and do not read much into a single week.
The result that should change your plan: if your page is indexed and ranking in traditional search for the sub-queries and ChatGPT still cites the competitor, the constraint is probably source preference rather than a missing document. At that point, independent coverage deserves more of the effort than further on-site pages.
My practical position: call it a retrieval outcome, treat it as a content and coverage gap, and only investigate entity signals if ChatGPT gets basic facts about you wrong. Start by reading exactly what the competitor’s page says about you, because that is what ChatGPT is currently quoting.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.