ChatGPT Ignores Your Page: How to Fix It
Most SEOs chasing AI visibility are solving the wrong problem. The conversation tends to center on ranking, domain authority, and whether your content is “optimized for AI” in some vague, hand-wavy sense. Get your page found, the logic goes, and citations will follow.
They won’t. Not automatically.
Ahrefs analyzed 1.4 million ChatGPT prompts and found that although ChatGPT retrieves around 33 URLs per query, it cites only roughly half of them. That means pages are being pulled into the retrieval pool, read (or at least considered), and then quietly discarded. No citation. No credit. Not because the content was bad, but because something earlier in the process filtered it out.
That earlier stage is what most people are missing.
Before ChatGPT opens your page and reads a single word, it screens a thin slice of metadata: your page title, a short snippet, and your URL. That screening process determines whether your content gets a fair hearing at all. Understanding how it works is the difference between showing up in AI responses and being perpetually retrieved but never cited.
This is not theoretical. The data is specific, and the fixes are practical.
The Two Stages Nobody Talks About
When ChatGPT runs a web search, it does not open every page it finds. That would be slow and expensive. Instead, it retrieves a batch of results, each one arriving with a package of lightweight metadata: the page title, a brief snippet or summary, the URL, and an internal ID number.
ChatGPT uses that metadata package to make a first-pass decision. Which of these results are worth actually opening? Which ones look relevant enough to read? Only the pages that pass this initial filter get opened and eventually considered for citation.
This means there are two separate moments where you can win or lose. The first is retrieval: does your page show up in the results pool at all? The second is selection: once you’re in the pool, does the metadata signal enough relevance to get you opened and cited?
SEO has trained most of us to obsess over the first stage. The Ahrefs study on ChatGPT citations makes a strong case that the second stage deserves equal, if not more, attention.
What the Gatekeeping Layer Actually Looks Like
The metadata screening is not a human editorial judgment. It is a form of semantic scoring in which ChatGPT estimates how closely your title and snippet align with the query it aims to answer. Pages that score well on that alignment get opened. Pages that do not get left in the pool, retrieved but uncited.
The study found that cited URLs consistently had higher semantic similarity between their title and the user’s original prompt. That gap widened further when the comparison shifted from the original prompt to what Ahrefs calls “fan-out queries,” the internal sub-questions ChatGPT generates to hunt down specific facts before composing its answer.
That second detail matters more than it might seem. When you ask ChatGPT something like “how do I get cited by AI,” it does not search for that phrase verbatim. It fragments the question into several more specific queries and searches for each one separately. Your page title needs to match those sub-questions, not just the surface-level prompt.
The practical implication is that writing your title to match obvious head terms is not enough. You need to think about the specific factual angles a model might search for when answering questions in your space. If you are working on your agency’s AI search strategy, the guidance on approaching AI-era SEO at Ethical Champ is worth reading alongside this data.
What is a Fan-Out Query?
When you type a question into ChatGPT, it rarely searches the web with exactly what you wrote. Instead, it breaks your prompt apart into a set of smaller, more specific sub-questions and runs separate searches for each one. Those sub-questions are fan-out queries.
Say you ask: “How do I get my website cited by ChatGPT?” ChatGPT does not search for that phrase verbatim. It might internally generate queries like “factors that influence ChatGPT citations,” “how ChatGPT selects sources from search results,” and “does URL structure affect AI citation rate.” Each one goes out as its own search. The results come back, get screened, and the pages that survive that screening get read and potentially cited in the final response.
The term comes from the way the process fans outward. One prompt becomes many searches, each targeting a specific angle of the original question.
This matters for your content because ChatGPT is not matching your page against the broad question a user typed. It is matching your page against those narrower, more specific sub-questions it generated behind the scenes. A page titled “How to Get Cited by AI” is competing on the broad prompt. A page titled “Does URL Structure Affect ChatGPT Citation Rate” is competing on a specific fan-out query, and that is where the real citation opportunity sits.
The Ahrefs study confirmed this directly. Cited pages had higher semantic similarity to fan-out queries than to the original user prompt. The gap was measurable and consistent across the data set.
You can actually see the fan-out queries ChatGPT generates for any prompt by using the browser devtools, opening the network tab, and searching the response data for the word “queries.” What comes back is the exact language ChatGPT searched with. That language belongs in your content.
How Semantic Similarity Scoring Works
ChatGPT is closed-source, so Ahrefs used cosine similarity computed from open-source embedding models to approximate how relevance scoring likely works. Cosine similarity is a standard way of measuring how close two pieces of text are in meaning, not just in exact wording. A score close to 1 means the texts are semantically near-identical. A score close to 0 means they are unrelated.
The study found a clear and consistent pattern: cited pages had higher cosine similarity scores between their titles and both the original prompt and the fan-out queries. Non-cited pages showed a significantly lower distribution, particularly in the search index results (which account for 88% of all citations).
What this means practically is that vague, broad, or keyword-stuffed titles hurt you twice. They are less likely to rank in organic search, and they are less likely to pass ChatGPT’s metadata filter. A title that precisely names the topic, the angle, and ideally the specific question being answered will outperform a generic one every time.
Moz’s writing on semantic SEO covers the underlying principles well if you want the theory. The Ahrefs data confirms that it applies directly to AI citation behavior.
The URL Gap: 89.78% vs 81.11%
One of the more concrete findings in the study is the difference in citation rates between pages with natural-language URL slugs and those without.
Pages with readable, descriptive URLs (think /why-chatgpt-ignores-your-page rather than /p=4821 or /article?id=9932b) were cited at a rate of 89.78%. Pages with opaque, non-descriptive URLs accounted for 81.11%. That is an 8.67 percentage point gap, and it is not trivial when you are trying to appear in AI responses at scale.
The reason is consistent with everything else in the study. ChatGPT’s metadata screening includes the URL as a signal. A URL that clearly describes the page’s content contributes to the overall semantic alignment score. An opaque URL contributes nothing, and may even slightly weaken the signal.
This is also consistent with longstanding SEO best practice, which makes it one of the easier fixes to justify internally. Clean URL structures have always helped with human readers and crawlability. Now there is direct data showing they help with AI citation rates too.
If your site is running on a CMS that generates messy URLs by default, changing them will require redirects and some care. But for any new content you publish going forward, there is no good reason to use anything other than a short, descriptive slug that reflects the actual topic of the page.
What to Actually Change on Your Pages
Start with your title tags. Look at your most important pages and ask whether the title precisely names the specific question or topic it answers. Not a broad category. Not a brand name with a vague descriptor. The actual thing a person (or an AI) would search for when they need what that page contains.
Then look at your H1 and your URL slug. These three elements, title tag, H1, and slug, form the core of what ChatGPT reads during metadata screening. They should all point at the same specific topic, ideally using the same language a searcher would use, not the language your internal team uses.
For existing content, particularly pages that rank but rarely get cited in AI responses, the intervention is often simple. A more specific title, a tighter H1, a cleaned-up URL. You do not need to rewrite the entire page. The body content is not even being read at the filtering stage.
For new content, build the topic specificity in from the start. If you want to understand what fan-out queries ChatGPT is generating for topics in your space, there is a practical method worth knowing: open ChatGPT, enter your target prompt, then right-click and inspect the page, go to the network tab, filter by the conversation ID from the URL, and look for the “queries” field in the response data. Those are the actual sub-questions ChatGPT searched with. Use that language in your H2s or build entirely separate pieces of content around each one.
The pages that get cited consistently are not necessarily the most comprehensive or the most authoritative. They are the ones whose titles and URLs made it obvious, at a glance, that they matched what was being asked. That is a much smaller lift than most people assume, and it is something you can start on before the end of the week.
