“We don’t show up when people ask ChatGPT” — why, and which cause is yours

If my business is not showing up in ChatGPT, the cause is usually not the website. Across 57 buyer prompts polled on both ChatGPT and Gemini between 3 July and 11 September 2026, the largest group — 39% — was prompts where our domain was findable on the open web and cited by neither engine.

• The symptom is not the diagnosis. “We don’t appear” has five distinct causes and they need different fixes.
• Most common cause: nothing the engine reads mentions you. It is a sourcing gap, not a site gap.
• Cheapest thing to rule out first: whether you are blocking OAI-SearchBot in robots.txt. Five minutes.
• The test that separates the causes is free and takes about 20 minutes across three engines.
• One check proves nothing. ChatGPT held a citation in a median 45% of a prompt’s polls after first appearing (n=10 ever-cited prompts, 3 July – 11 September 2026).

Why is my business not showing up in ChatGPT? The five causes, ranked by frequency

We run a citation monitor against a registry of real buyer prompts in our own niche. Between 3 July and 11 September 2026 it logged 1,371 answered polls across 71 prompts on ChatGPT (browse-enabled) and Gemini (search-grounded); 57 of those prompts were polled on both engines and on DuckDuckGo answer pages, so every prompt in the table below has all three surfaces observed. Each prompt falls into exactly one state, and the states are ranked by how often they occur — not by how bad they are.

Two-thirds of the absences in our own log (38 of 57 prompts) are cases where no AI engine cites us at all, and only 16% are ChatGPT being fussier than Gemini. One domain, one niche (lead generation and appointment setting) — directional, not universal. Method and the earlier 77-day window: our first-party dataset of 5,051 AI-search citation polls.

The 20-minute test that tells you which cause is yours

Do this before you buy anything. It is the only free way to tell a sourcing problem from a content problem from a flicker.

1. Write five buyer prompts, not brand prompts. “Best appointment setting agency for mortgage brokers”, not “what is [your company]”. Nobody who has not heard of you asks about you by name.
2. Run all five in a logged-out ChatGPT session and record cited / not cited. Roughly 5 minutes.
3. Run the same five in Gemini and in one answer-page search engine. Another 10 minutes. Now you have the three surfaces the table above uses.
4. Open every source the engine cited — not the answer, the sources. Count how many are pages you own. If the answer is zero and the rest are directories, forums and third-party listicles, you are in row 1.
5. Re-run the ChatGPT five in seven days. A prompt that misses twice is an absence; a prompt that hits once and misses once is flicker, which is row 4 and a different job entirely.

Step 4 is the step people skip and the one that carries the diagnosis. To turn this into a standing instrument, our guide to tracking your own AI search visibility DIY has the prompt-registry and logging method we use.

Cause 1 (39%): nothing you do not own mentions you

This is the biggest bucket and the least intuitive, because the instinct when you are absent is to rewrite the website. The engines are mostly not reading your website. Across 774 answered ChatGPT polls on 60 tracked prompts (3 July – 11 September 2026) our monitor logged 7,220 cited-source rows, and leadsnow.ai accounted for 105 of them — 1.5%. On Gemini it was 304 of 5,069 rows, or 6.0%. The other 94–98% of what the engines read to answer our buyers’ questions are pages we do not control.

Which pages? Across both engines in that window the monitor recorded 2,692 distinct cited domains. reddit.com was cited on 32 of the 71 tracked prompts and clutch.co on 17 — more distinct prompts than any vendor’s own website except our own (19). If a forum thread and a directory profile are the most broadly-read sources in your category and neither names you, on-site work will not put you in the answer. The fix is presence where the engines already read: our note on Reddit as a source of AI search citations for B2B covers the mechanics; directory profiles, earned media and third-party comparison pages do the same job.

Cause 2 (28%): you have no page that answers the literal question

Sixteen of the 57 prompts returned nothing of ours on any of the three surfaces. In every case we checked, we had a page about the topic and no page that answered the sentence. A services page titled “Lead Generation” is not a candidate answer to “which agency can optimise my business for ChatGPT”; it does not contain the question, so there is nothing for retrieval to match and nothing extractable to quote.

The test is mechanical: paste the exact prompt into your site search. If no page carries that sentence or a close paraphrase in a heading, and none answers it in its first 50 words, retrieval has nothing to match. The extraction mechanics are in how to be in the 15% of pages ChatGPT actually cites.

Cause 3 (16%): you are in AI answers, just not the one you checked

Nine of the 57 prompts cited us on Gemini and never on ChatGPT; exactly one went the other way. Ever-cited rates over the same window: ChatGPT 10 of 60 prompts (17%, 774 answered polls), Gemini 18 of 68 (26%, 597 polls). ChatGPT is the harder door, and it is also the one everybody screenshots.

If your entire evidence of failure is a single ChatGPT session, roughly one in six of our own “absences” would have been a false alarm. Check at least two engines before you conclude anything, and report them separately — a blended AI visibility number hides the only useful signal in it.

Cause 4 (16%): you were cited, and the citation did not hold

A citation is not a ranking. Measured on our own prompt set between 3 July and 11 September 2026, ChatGPT kept citing us in a median 45% of a prompt’s polls counted from the first citation onward (n=10 ever-cited prompts), against 94% for Gemini (n=18). ChatGPT citations flicker; Gemini citations stick. Our earlier 19 May – 3 August window put the same two medians at roughly 44% and 70%, so the Gemini figure has moved and the ChatGPT figure has not.

At 45% persistence, a single ChatGPT check is closer to a coin flip than a status report. Two misses a week apart is evidence; one miss is not.

The one cause our data cannot rank: you are blocking the crawler that does the citing

Our monitor cannot rank this because we block nothing, so it never appears in the 57. Check it first anyway: it is the only cause with a five-minute fix. OpenAI documents four separate agents, and the one that matters is OAI-SearchBot, used to “surface websites in search results in ChatGPT’s search features”; its documentation states that sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. Blocking GPTBot, the training crawler, is a different decision — and plenty of sites have blocked both by pasting a snippet from a blog post. Open yourdomain.com/robots.txt and read it. If OAI-SearchBot is disallowed, that is your answer.

“ChatGPT crawls my site every day and still never mentions me”

Being fetched and being cited are separate events, and the gap between them is the most misread signal in this whole category. Between 1 and 11 September 2026, OpenAI’s two answer-time agents fetched leadsnow.ai 1,761 times — ChatGPT-User 1,200 and OAI-SearchBot 561, counted from our own server bot log. Over the same eleven days our monitor ran 78 answered ChatGPT polls across 36 tracked prompts and our domain appeared on 5 of them (14%). Roughly 1,761 fetches converted into five prompts’ worth of visibility. Crawl logs measure interest; only a poll series measures citation.

Which means the honest cost of doing this properly is a standing measurement habit plus whatever the fix for your row actually requires. Here is where the arithmetic turns:

The first three rows are genuinely do-it-yourself and most businesses should start at row two. The crossover is not a page count, it is the fix: rows 3 and 4 of the causes table are content and cadence work you can do; row 1 — getting named on sources you do not own — is outreach, and outreach is where the hours stop being spare. That is the part we sell as AI SEO and AEO services, and equally the part you can run in-house given someone who will do sustained outreach for months.

Frequently asked questions

Why is my business not showing up in ChatGPT even though I rank on Google?

Because ranking and citation are separate systems. Google’s own documentation for AI features says that to be eligible as a supporting link in AI Overviews or AI Mode “a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements” and that there are no additional technical requirements — eligibility is the floor, not the outcome. In our own log, 22 of 57 tracked prompts (39%, 3 July – 11 September 2026) were findable on the open web and cited by neither ChatGPT nor Gemini.

Does blocking AI crawlers stop me appearing in ChatGPT?

One of them does. OpenAI documents OAI-SearchBot as the agent used to surface websites in ChatGPT’s search features, and states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links. GPTBot is the model-training crawler and is a separate decision. Check your robots.txt for both names before you spend money on anything else.

How do I check whether ChatGPT mentions my business?

Run five buyer-intent prompts — not your brand name — in a logged-out session, record cited or not cited, then repeat in seven days. One check cannot distinguish absence from flicker: on our prompt set ChatGPT held a citation in a median 45% of a prompt’s polls after first appearing (n=10 ever-cited prompts, 3 July – 11 September 2026).

Which agency can optimise my business for ChatGPT, and what should I ask them?

Ask for their own poll log before you ask for a case study: which engines, over what window, across how many prompts, and what the persistence figure was. Anyone selling this should be able to answer in numbers, because the measurement is cheap and the absence of it is a choice. A screenshot of one ChatGPT answer is not evidence at 45% persistence.

How many prompts do I need to track before the number means anything?

Ten to fifteen, polled weekly, is enough to see a trend within about three weeks. Our own registry sat at 71 prompts across ChatGPT and Gemini for the window in this article; past roughly 70 you add polling cost faster than insight. Freeze the list once written, or week one and week eight are not comparable.