Why Your Brand Is Mentioned in ChatGPT but Not Linked

ChatGPT knows your brand. Ask it a question in your niche and it will name you. Then watch it cite a competitor's URL for the actual answer. That gap, where the model recognizes you but links someone else, is the most common AI-visibility failure we measure, and it is not the same problem as being unknown. Being unknown is a content problem. Being mentioned-but-not-linked is a citability problem, and the fix is entirely different.

Measuring your AI citation rate and setting it beside your recognition rate is how you see the gap in numbers. The conceptual ground is in What Is AI Citability, which defines citability as a measurable property of your pages rather than a function of how famous your brand is. This post is the diagnosis: why the gap exists, and which of three causes is producing yours.

Recognition and Citation Come From Different Machines#

Recognition and citation are produced by two separate mechanisms, which is the entire reason they diverge. Recognition is recall from training data: the model saw your brand often enough to name it from memory. Citation is live retrieval: when answering a current question, the model fetches candidate pages, parses them, ranks them by trust, and links the winner.

Your brand can clear the first bar and fail the second every single time. Being remembered is a frozen snapshot from the model's last training cut. Being cited is a real-time judgment about whether your page, right now, is the best source it can fetch and quote for this exact question. Nothing about being remembered guarantees being retrieved.

This is why "get more famous" is the wrong instinct. Fame feeds recognition, which you may already have in abundance. Citation is won at the page and domain level, by being retrievable and trusted for the query. A brand can be a household name in its niche and still lose every citation to a page that is simply easier to quote.

The Recognition-Citation Gap Is Measurable#

When we ran the AVR (AI Visibility Readiness) baseline against citability.dev's own topic queries, the four signals split apart instead of moving together. Recognition was the highest, and citation was a fraction of it. The model knew the brand and still answered most queries with someone else's URL:

Signal	Rate	What it measures
Brand recognition	50%	The model knows citability.dev exists
Citation rate	15.6%	The model links a citability.dev URL in the answer
Concept attribution	0%	The model credits citability.dev for the concept
Recommendation rate	0%	The model names citability.dev as a tool

Read top to bottom, this is the recognition-citation gap as a shape, not an anecdote. Recognition at 50 percent says the brand registered. Citation at 15.6 percent says the pages did not win retrieval for most queries. Attribution and recommendation at zero say the model does not yet treat the domain as a source for the concept or as a tool for the job. That descending staircase, from high recognition down to low citation and zero attribution, is the starting shape for most sites that have a name but not yet a structured, authoritative content layer.

What the Gap Actually Costs You#

Recognition without citation feels like visibility and behaves like invisibility. The model names you in passing, but the searcher who asked the question is handed a competitor's link as the answer. You get the mention; the competitor gets the click, the trust signal, and the next step of the relationship with that buyer.

The cost compounds in two directions at once. Every uncited answer is referral traffic that routed to someone else, so the gap is a daily leak rather than a one-time miss. And citation is itself a trust signal that feeds future ranking: pages that get cited get treated as more authoritative, which makes them more likely to be cited again. Recognition does not compound that way. A brand stuck at recognition-without-citation is not holding steady. It is slowly losing ground to whichever competitor is banking the citations its content should have earned, and the authority gap widens the longer it stays open.

Cause 1: Your Page Is Not Retrievable for the Query#

The most common cause is mechanical. The model wanted a page on your topic, tried to fetch one, and could not use what came back. AI crawlers may be blocked in robots.txt. The page may render its content client-side, leaving the fetched HTML nearly empty. Or the answer may be buried under preamble instead of stated up front.

You can detect this one yourself in minutes. Fetch your page the way a crawler does, with curl and no JavaScript, and read what comes back. If the answer to the target question is not in that raw HTML, the model cannot quote it either. Then check robots.txt for explicit allow rules covering GPTBot, ClaudeBot, and PerplexityBot, and confirm the page is server-rendered rather than hydrated in the browser after load.

The fix is concrete and fast. Allow the AI crawlers. Serve the content server-side so the raw HTML carries the answer. State the answer in the first two sentences of the relevant section. What AI crawlers see walks through exactly what the fetched version of your page looks like, which is rarely what you see in a browser, and it is the first place a citation gap hides.

Cause 2: The Model Does Not Treat Your Domain as a Source#

If your page is retrievable and you are still not cited, the cause is authority rather than mechanics. The model can read your page but does not yet rank your domain as a trustworthy source for the topic, so it cites a domain it does trust instead. This is the entity-authority gap, and it is patient work to close.

You can recognize this cause by its signal pattern. Citation is nonzero but low, while attribution and recommendation sit at zero. The model retrieves you occasionally for narrow queries but does not think of you as a source when explaining the concept or naming a tool. That is not a page problem you can fix on-site; it is a reputation the model has not formed about your domain yet.

The fix is slower and lives off-page. Earn citations from domains the model already trusts in your category, because the model learns who is authoritative partly from who else cites them. Establish a consistent definitional presence: the same branded anchor text describing the same concept across the places the model crawls. Where it fits your category, a Wikidata or reference-context entry anchors your entity in the structured graph these systems lean on. The publishing-side mechanics of earning those citations are covered in chudi.dev's guide to answer engine optimization.

Cause 3: A Competitor's Page Is Simply More Citable#

Sometimes nothing is wrong with your page in isolation. A competitor's page is just a better citation candidate for that exact query. It is answer-first where yours is narrative. It carries FAQPage or HowTo schema where yours carries none. Its headings match the question wording, so retrieval finds a clean, quotable passage at once.

This cause shows up as citation that varies query by query: strong on the posts you structured well, absent on the ones you wrote as essays. That variance is the tell. Where a specific competitor keeps winning a specific query, pull up their page and yours side by side and compare the first sixty words under the relevant heading. The difference is usually that theirs answers and yours warms up.

The fix is to make your page the better candidate, one post at a time. Match the question in an H2, answer it in the next sentence in forty to sixty words, and add the structured data the engine expects. The schema markup AI actually uses is a short list, not the whole of schema.org, and shipping the right few types closes this gap per page without touching the rest of your site.

A Worked Example: Reading One Site's Split#

Picture a site with 60% recognition, 20% citation, 0% attribution, and 0% recommendation. The recognition number says the model knows the brand. The citation number, low but nonzero, says some pages are getting retrieved while most are not. The two zeros at the bottom say the domain is not yet a trusted source for the concept or the category.

That pattern rules Cause 1 mostly out: if pages were unretrievable across the board, citation would be near zero too, not 20 percent. It points instead at a mix of Cause 3 (the cited pages are the well-structured ones, the uncited pages are the essays) and Cause 2 (the domain has not earned source-level trust, which is why attribution stays at zero). The fix order follows the diagnosis: structure the essay pages first because that lever is fast, then start the slow authority work that moves attribution off zero.

How to Tell Which Cause Is Yours#

The three causes have different fixes, so guessing wastes a quarter. The diagnostic is the signal split, read as a pattern. Zero citation across every query points to Cause 1, a retrievability problem with a fast mechanical fix. Nonzero citation with zero attribution and recommendation points to Cause 2, an entity-authority gap with a slow off-page fix.

Citation that varies query by query, strong on some posts and absent on others, points to Cause 3, where the per-page citability of specific posts is the lever you pull. Most sites have a primary cause and a secondary one. The reason to measure all three signals separately, rather than tracking a single blended "AI visibility" score, is that the blended number tells you something is wrong without telling you which of the three machines to go fix first.

Where to Start#

Rule out the mechanical causes before you spend a quarter on authority work. The free citability.dev scan checks the ten infrastructure signals behind Cause 1 and Cause 3 in a single request, so you learn in seconds whether your pages are retrievable and well-structured. If the scan comes back clean and the gap is still open, the cause is authority.

Where the infrastructure is sound and the gap persists, the audit quantifies the citation, attribution, and recommendation split across your real query set, so you fix the one signal that is actually leaking instead of all three at once. Whichever cause you fix, re-run the measurement after four to six weeks: model updates and fresh crawls shift citation patterns on their own timeline, and the only proof a fix worked is the citation rate moving against the same query set you started with. Being mentioned is not the finish line. It is the model telling you it knows you exist and is waiting for a reason to link you. Give it one.