You ask ChatGPT the question your customers ask and it names a competitor. Not a bigger one, necessarily. Sometimes a smaller one with a worse product. The instinct is to assume the model is biased or random. It is neither. It is working from what it could read, and the answer usually tells you what went wrong if you know where to look.
Here are the seven causes we see most, roughly in the order you should rule them out.
Key takeaways
Read the answer like a detective: which pages did it cite, what reason did it attach to the rival, and did it mention you at all? Each pattern points to one of seven causes. Three of them are technical and fix in a day. Four are about content and sources and take a few weeks to show.
First, read the answer properly
Ask the question in a fresh chat with web search on. Then record three things before you do anything else:
- The citations. Click every source. Note the domain and the kind of page: a roundup, a review site, a forum thread, the rival's own page.
- The reason. The phrase attached to the rival. "Great for teams", "cheapest option", "open on Sundays". That phrase came from one of the cited pages.
- Your absence, or your caveat. Were you not mentioned, mentioned with a "but", or mentioned with a wrong fact?
Those three observations are enough to diagnose almost every case.
Cause 1: ChatGPT cannot read your site
The tell: you are not mentioned at all, on any question, including "what do you know about [your name]". The rival's own pages are among the citations. Yours never are.
Your robots.txt blocks OAI-SearchBot or ChatGPT-User, your firewall serves a challenge page to bots, or your pages render their content with JavaScript and arrive empty. This is the most common cause on sites built in the last three years, and the easiest to miss because everything looks fine in a browser.
The fix: run the AI bots checker and the citability checker. Allow the documented bots, and make sure the HTML contains the text. Our robots.txt guide covers the rules.
Cause 2: You have no page for that question
The tell: you are mentioned for branded questions but not for "best X" or "X for Y". The citations are roundups and comparison pages, none of which are yours.
ChatGPT searches, then reads the best matches. Your homepage matches nothing specific. The rival has a page titled "Invoicing for freelancers" and you have a homepage that says "Get paid faster".
The fix: one page per customer question. Title it with the question in the plain form a stranger would type, answer it in the first screen, and state the facts as sentences. The citation checklist covers how to build the page.
Cause 3: The fact behind the reason is missing from your site
The tell: the rival is named with a specific reason, and that reason is something you also do, but the model did not know it.
"Ledgerly is a strong choice for Stripe users" means a cited page said Ledgerly syncs with Stripe. If you also sync with Stripe but it is only shown as a logo on an integrations page, the model never read it as a fact.
The fix: write the fact as a sentence on the page that matches the question. "Syncs with Stripe, PayPal and GoCardless; payments reconcile automatically." Then do the same for every reason attached to every rival in your answers. That list of reasons is the most useful output of a scan, and our post on the facts AI can't find shows how to state each type.
Cause 4: The rival is on the sources, and you are not
The tell: the citations are third-party pages, the same two or three domains across many questions, and the rival appears on them.
Most "best of" answers are written from roundups, review platforms and forum threads. If a category has one strong roundup and the rival is in it, the rival is in the answer. Nothing on your own site changes that.
The fix: get onto those pages. For review platforms, claim your profile and ask recent customers for reviews. For roundups, contact the author with a factual update; most maintained lists add credible entries. For forum threads, answer the actual question as the owner with your facts and no pitch. For directories, get listed. The sources that decide who gets recommended explains how to find the ones that matter for your market, and the TangentFlow directory is one built to be read by assistants.
Cause 5: A wrong fact is disqualifying you
The tell: you are mentioned, but with a caveat that is not true. "Northline is an option, though it does not offer a free tier." You do offer a free tier.
The model read an old page, a stale review, or a third-party listing you never updated. Wrong facts on trusted sources outrank right facts on your own site if yours are hard to find.
The fix: find the source of the wrong fact by clicking the citations, then fix it at the source: update the listing, reply to the review, ask for the correction. At the same time, state the correct fact plainly on your own matching page so the model has a second, current source. The mirror question in a scan ("what do you know about us") lists these caveats with the page each came from.
Cause 6: Your names do not match
The tell: you are described accurately in a branded answer, but for category questions the model names you under a different name, or splits you into two.
"Hartley & Co" on the site, "Hartley and Company Accountants" on Google, "Hartley Co." on a review platform. To a model matching names across sources, that is three weak businesses instead of one.
The fix: choose one name and one location string. Use it on the site, in schema, on your Google profile and on every listing. It takes an afternoon and removes a surprising amount of noise.
Cause 7: The rival is simply better documented
The tell: the rival is cited from its own site as often as from third parties. Its pricing page, docs, comparison pages and case studies show up as sources. Yours never do.
Some companies have quietly done the work: a comparison page for every rival, a docs site that answers "does X integrate with Y", a pricing page that states every number. The model cites them because they are the best answer to the question, and no third party is needed.
The fix: this is the long game and the one with the most durable payoff. Build the comparison pages, the "X for Y" pages, the plain pricing page, the FAQ that answers real questions. Publish on a schedule and keep them updated. TangentFlow's Autopilot drafts these from your own facts and puts them on a calendar, but the pattern works with any tool, or none.
Which cause is it?
| What you see in the answer | Most likely cause |
|---|---|
| Never mentioned, even by name | 1: cannot be read |
| Named on branded questions only | 2: no page for the question |
| Rival named with a reason you also qualify for | 3: fact missing |
| Same third-party domains cited everywhere | 4: not on the sources |
| Named with an untrue caveat | 5: wrong fact |
| Named under a variant name | 6: name mismatch |
| Rival's own pages cited repeatedly | 7: better documented |
Most sites have two of these at once, usually one technical and one content. Fix the technical one first; the content fix cannot land while the model cannot read you.
How long until the answer changes
Technical fixes show within a crawl cycle, often a week or two. Content and source fixes take longer: the new page needs indexing, the review platform needs the reviews, the roundup author needs to publish. In our experience the first question usually flips three to six weeks after the fix, and the rest follow as the pages get read.
The only way to know is to ask again, in a fresh chat, on a schedule. A free scan records the citations, reasons and caveats for twelve of your customers' questions so the diagnosis above takes minutes, and paid plans re-ask the same questions every week and email you when one flips.
