Usually one of seven reasons, roughly in the order ChatGPT hits them: a blocked fetcher, a JavaScript-only page, an unclear entity, no third-party evidence, a page that doesn't answer first, no outside source naming you, or a roundup you're not in. Most businesses have checked none of the seven.
This is the fault-by-fault version of a question the get recommended by ChatGPT guide answers in five steps: each reason below gets its own five-minute check, in the order ChatGPT runs into it.
Access first, then readability, then who you are, then third-party backup, then page structure, then outside sourcing, then roundup inclusion. A problem early in that order hides every problem after it: if OAI-SearchBot can't fetch your page, it never gets far enough to judge your entity or your prose.
| # | Reason | Five-minute check |
|---|---|---|
| 1 | Fetcher blocked | robots.txt, CDN/WAF rules |
| 2 | JavaScript-only page | view the raw HTML |
| 3 | Entity unclear | Organization + sameAs in JSON-LD |
| 4 | No third-party evidence | reviews, mentions, directories |
| 5 | Page doesn't answer first | read your own first sentence |
| 6 | No source names you | ask ChatGPT, open its sources |
| 7 | Not in the roundup | ask the exact buyer question |
Check this first, because every other reason here is irrelevant if OAI-SearchBot or ChatGPT-User can't reach your page at all. OAI-SearchBot crawls to build ChatGPT's search index; ChatGPT-User fetches a page live when a question sends it there. A robots.txt that blocks either one, or a CDN rule (Cloudflare's bot-fight mode is the common culprit) that challenges them first, removes you from the pages ChatGPT can cite at all.
Proof: OpenAI — GPTBot & OAI-SearchBot documentation.
Check it: run the free AI crawler access check, or ask ChatGPT a buyer question with web search on and open every source it cites. If your domain never appears, start here.
If your content loads only after JavaScript runs in a browser, ChatGPT's fetcher sees an empty shell, because it reads the raw document and does not execute scripts. A React or Vue single-page app is the common version of this; Webflow, WordPress and most static site builders render text server-side by default and don't have the problem.
Check it: the free what AI sees tool shows the exact raw document a fetcher receives. If your text is missing there, it's missing for ChatGPT too.
If your site never states a clear, machine-readable identity, ChatGPT has no confident entity to attach a recommendation to, even once it has read the page. An Organization block in JSON-LD with a sameAs array linking your LinkedIn and, where one exists, your Wikidata item gives the model one unambiguous subject instead of a name it has to guess at.
Check it: run the free schema & entity check. If your Organization JSON-LD is missing or doesn't match your About page, that's reason three.
Your own claims about your own business carry less weight than the same claims on a site ChatGPT already trusts, a review site, a directory or press coverage. In Reflexa's own AI Visibility Index for project management (September 2026, 882 eligible answers), the category leader was mentioned in 51% of answers but had its own site cited as a source in only 18%, proof that most of the backup lives on other people's pages, not its own.
Check it: see do reviews change whether AI recommends you? for which review sites matter most for your category, and check whether you appear on any of them.
A page that opens with a brand story before stating what the business does and who it's for gives ChatGPT no self-contained passage to quote. Engines favor pages that answer a specific question plainly in the first sentence, because that's the passage that can be lifted into an answer without interpretation.
Check it: open your homepage or main service page and read only the first sentence out loud. If it doesn't state what you do and for whom, a model has to infer it, and inference is where it reaches for a competitor instead.
Ask ChatGPT the exact question a buyer would ask, with web search on, and open every source link in the answer, because that list is the honest answer to whether you're reachable at all. If your domain isn't in the sources for your category's buyer questions, no amount of on-page fixing helps until that changes.
Check it: run three or four buyer questions, not one, since answers vary between runs. If your name never appears across several tries, the gap is sourcing, not wording.
Some buyer questions ("best X for Y") pull their answer from a single roundup or comparison page that already lists a fixed set of names, and if you're not on that page you're not in the answer, no matter how good your own site is. It's the hardest of the seven to fix directly, since it depends on someone else's editorial page, but it's also why a technically perfect site can still be left out.
Check it: find which roundup the answer draws from (the sources panel usually shows it), and pursue getting listed there like any other press or directory placement.
Work the list in order, none of it costs anything:
Bottom line: fix the earliest blocker you find first. A page ChatGPT still can't fetch makes every later fix invisible.
Usually one of seven reasons, roughly in the order ChatGPT hits them: its fetcher is blocked, the page is JavaScript-only, your entity is unclear, nothing third-party backs you up, your page doesn't answer the question in its first sentence, no outside source names you, or the question is answered by a roundup you're not in.
No. Answers vary between runs, so one try tells you about that answer, not your visibility. Ask the same buyer question a few times, with web search on, and read the sources it opens each time.
No. Allowing OAI-SearchBot and ChatGPT-User only removes the reason furthest upstream. If your entity is still unclear or no outside source has ever named you, an unblocked crawler just reaches a page it still has no reason to recommend.
Usually because they clear more of the seven reasons above than you do, not because of one trick. A competitor with reviews on a site ChatGPT trusts, and a page that states who they're for in its first sentence, gives the model more to work with.
Each reason above links to a free, five-minute check. Or run the free check at dashboard.reflexatechnologies.com/audit, which tests your site against all seven at once and shows the evidence.
The free check tests your site against all seven, from outside, the way ChatGPT does it. 3 minutes, evidence included.