Why is my brand not in AI answers?

We are not writing this from the outside. Our own audited visibility score is zero — named in 0 of 8 answers, 0 of 90 citations — so this page is the diagnosis we ran on ourselves.

Cause 1: nobody is asking the question you think they are

The most common cause, and the cheapest to check. People tracking “AI visibility” for their brand name are measuring a question almost nobody asks. A buyer who already knows your name does not need an assistant to recommend you.

The check: write down the ten questions a buyer asks before they have heard of you. Not “is Acme any good” — “what is the best X for a small team”. If your tracked prompts are mostly your own brand name, you are measuring the wrong thing. See choosing the questions to track.

Cause 2: your page cannot be fetched

An engine that cannot retrieve the page cannot cite it, and this failure is silent — you see an absence, not an error. The usual culprits:

• robots.txt blocks the AI crawler, often added years ago or by a plugin default.
• A bot-protection rule at the CDN challenges or blocks non-browser user agents.
• The content only exists after JavaScript runs. A crawler that does not execute scripts sees an empty shell.
• The page is behind a login, an email wall or a cookie consent gate.
• The facts live in a PDF, an image or a video. Nothing can quote those.

The check: fetch your own page the way a crawler would — no browser, no JavaScript — and read what comes back. If your key sentence is not in that text, no engine has ever seen it. The crawler names and the robots.txt decision are on AI crawlers and robots.txt, which also covers what each crawler is for and how to decide what to allow.

Cause 3: no page of yours answers that exact question

This is the cause for most sites that pass the first two, and it was ours. The evidence from our own audit: of the 74 domains Perplexity cited in run A across our eight questions, 69 were cited for exactly one question, and 74 of the 90 cited URLs in run A carried a word from the question in the path itself (75 of 90 in run B). Not one citation went to a homepage, in either run.

A homepage, a product page and a features page are pages about you. The engine is looking for a page about the question. If the closest thing you have is “Our platform”, you are not in the running.

The check: for each of your ten questions, name the URL on your site that answers it directly. Where you cannot name one, that is the page to write. The method is on writing a page an engine will quote.

Cause 4: the answer is on the page but not in a quotable form

You have the page, the engine fetched it, and it still cited someone else. Usually because the answer is buried, hedged or not actually stated. Three specific versions:

• The answer is in paragraph nine, after eight paragraphs about why the topic matters.
• The specific is missing. “Pricing starts from an affordable monthly fee” cannot be cited by anything. A figure can.
• The page hedges every sentence, so there is no self-contained claim to lift.

The check: delete everything but your first two sentences. Do they answer the title? If not, rewrite the opening before you touch anything else.

Cause 5: nothing outside your own site mentions you

Being cited and being named are different things, and they fail for different reasons. Your own pages can win citations. Getting named as a recommendation in an answer usually depends on other people’s pages — roundups, comparisons, directories, coverage — because those are what the engine reads when it is asked “who are the best X”.

The check: ask the engine a “best X” question in your category and read the sources it cites. Those specific pages are your target list. Not “build backlinks” — get onto the handful of pages that are actually being read for your question. In our own category that meant a set of roundup posts on domains most people have never heard of, which is a much cheaper target than it sounds.

Note the asymmetry this creates, which is explained fully on citation share: you can be cited constantly and named never, or named constantly and cited never. Work out which one is failing before you spend anything.

Cause 6: you are measuring it wrong

The commonest false alarm. Before concluding you are invisible, rule out these:

• One run is not a measurement. Engines are non-deterministic. Our two runs a minute apart produced identical headline numbers but moved per-domain counts by one or two — peec.ai from 4 citations to 2. Check twice.
• A failed API call is not an absence. If your tool records a timeout as “not mentioned”, your score is fiction. Ours excludes failures from every metric and marks the run incomplete.
• Changing the question set breaks the series. Different questions, different denominator, no comparison.
• Analytics referral data is not citation data. A citation that nobody clicked still shaped the answer, and it leaves no trace in your analytics.

The check: our method, including how failures are handled, is on how we measure AI visibility.

The order to work through it

1. Are you tracking questions a buyer asks before they know you? (Cause 1)
2. Can a crawler fetch the page and read the sentence? (Cause 2)
3. Does a page exist for that exact question? (Cause 3)
4. Is the answer in the first forty words, with a specific in it? (Cause 4)
5. Are you on the pages the engine reads for “best X”? (Cause 5)
6. Is the measurement itself sound? (Cause 6)

Causes 1 to 4 are entirely within your control and cost nothing but writing. Cause 5 is slower and involves other people. Cause 6 is free and should honestly be checked first. None of the first four needs a vendor. If you get to the end of them and want to hand the work over, tool, agency or in-house is the decision, and what is actually proven is the evidence you should ask any of them for.

What we found when we ran this on ourselves

Causes 1, 2 and 6 were clean: our questions were buyer questions, our site is static HTML that any crawler can read, and our measurement excludes failures by design. Cause 3 was the answer. On 19 September 2026 we had six pages, none of which was titled as a question a buyer asks, against competitors with dozens. We have been fixing that in public ever since, and the monthly re-run will say whether it worked.

Common questions