Why ChatGPT Ignores Half the Pages It Fetches

When ChatGPT searches the web before answering a question, it doesn't read every page it finds. It retrieves roughly 16 pages per query, then filters down to the ones worth citing — and that filtering decision happens before it fully opens most of them.

Scout filed a detailed analysis of the Ahrefs 1.4 million ChatGPT prompts study on 2026-10-16, and one finding has direct implications for every local service business with a website: whether a page gets cited depends heavily on signals ChatGPT can evaluate without loading the full page — specifically, the title and the URL slug.

ChatGPT fetches 16 pages and cites about 8

The Ahrefs research (1.4 million ChatGPT 5.2 queries, February 2025) established a baseline that most businesses don't know: ChatGPT averages 16.57 retrieved URLs per query for pages that ultimately get cited, and 16.58 for pages that don't get cited. The retrieval count is nearly identical whether a page wins a citation or not.

That means getting into the retrieval pool — having a page that ChatGPT finds at all — is not the bottleneck most people assume. Roughly half of all retrieved pages are dropped before appearing in the final response. There's a second gate, and it runs on different signals than the first.

For a local service business, this distinction matters. Being indexed by Bing, having a Google Business Profile, and appearing in directory listings all help with retrieval. But they don't determine whether ChatGPT actually names your business in the answer it gives a customer asking for an HVAC contractor at 10pm.

The filter runs before ChatGPT opens the page

After running a web search, ChatGPT sees a standard results page: titles, URLs, and snippets. The Ahrefs data shows that citation selection correlates most strongly with the title-to-query semantic match — the degree to which a page's title directly answers the specific sub-query ChatGPT generated for that prompt.

"Sub-query" is the key concept here. ChatGPT doesn't search verbatim for what a user typed. A user asking "who's the best plumber near me in Burlington" triggers a fan-out process where ChatGPT generates multiple internal queries: "top-rated plumber Burlington Ontario," "Burlington plumbing contractors reviews," "emergency plumber Burlington." Each fan-out query has its own results page. The pages whose titles match those specific fan-out queries are the ones that survive post-retrieval filtering.

This is why "ABC Plumbing — Serving Burlington Since 1993" underperforms against a title like "Emergency Plumber Burlington — ABC Plumbing — Licensed, 24/7." Both pages might rank identically in Google. Only one tells ChatGPT, at the metadata evaluation stage, that it answers the query for "emergency plumber Burlington."

The 8.67-point URL readability gap

Scout's October 2026 analysis of the Ahrefs dataset identified a second structural finding that's more immediately fixable for most businesses: URL slug readability produces a measurable citation rate gap.

Pages with natural language URL slugs — `/hvac-contractor-burlington/`, `/emergency-plumber-toronto/` — were cited at 89.78% of the time when retrieved. Pages with opaque URLs — `/page?id=29183`, `/services/p/3892` — were cited at 81.11%. That's an 8.67 percentage point difference from the URL structure alone.

The mechanism is the same as the title filter: ChatGPT reads the URL at the metadata evaluation stage, before the page is opened. A readable slug confirms topical alignment. An opaque URL provides no signal, so ChatGPT is more likely to deprioritize it when other retrieved pages have URLs that explicitly name the service and location being asked about.

For the majority of small business websites, this is a one-time infrastructure fix. Most website platforms default to readable slugs when configured correctly. If your service pages are running on query-string URLs or auto-generated numeric IDs, you're paying a citation penalty on every query that retrieves your site — not because your content is wrong, but because the signal ChatGPT reads first is uninformative.

The title-overlap corroboration

Scout's 2026-10-10 analysis of the AirOps State of AI Search 2026 data filed this separately: pages with 50% or greater title-to-query overlap had a 20.1% citation rate against a 15% baseline for all retrieved pages. A 5.1 percentage point lift purely from title alignment.

The Ahrefs study is higher confidence (1.4 million prompts with disclosed methodology) and the primary vault anchor for this finding. The AirOps figure is directionally consistent from a different measurement framework. Both datasets converge on the same conclusion: title relevance is a citation lever that can be adjusted today, without waiting months for domain authority gains or Google index refreshes.

Why Reddit gets retrieved but not cited

One of the sharpest contrasts in the Ahrefs data: Reddit is retrieved heavily — ChatGPT pulls Reddit pages for context in a significant share of queries. But Reddit URLs were cited at only 1.93% when retrieved. The general web search channel cited pages at 88.46%. That gap produced a finding worth noting: 67.8% of all non-cited URLs in the Ahrefs dataset came from the Reddit source.

ChatGPT is systematically reading Reddit and then not citing it in the final response.

For businesses whose AEO strategy includes building Reddit presence hoping it converts to ChatGPT citations — the data says it won't reliably. Reddit matters for Perplexity, where our vault data documents roughly 46% Reddit citation volume for that platform. It matters for organic community visibility. For ChatGPT citations specifically, your own website, directory listings, and the metadata signals described above are the levers.

Three things to do with this

**Audit your service page titles against actual buyer query language.** If your titles describe your business rather than answer a query, ChatGPT's fan-out filter will pass over them when a more directly-titled page is in the same retrieval pool. "HVAC Services | ABC Company" is a description. "HVAC Contractor Burlington — Emergency AC Repair | ABC Company" is a title that survives metadata filtering.

**Check your URL slugs for opaque patterns.** This is Phase 1 infrastructure work — a one-time change with permanent citation benefit. Service pages and location pages get the most value. Fixing a URL from `/services/1249` to `/emergency-plumber-burlington/` is a small change that changes the signal ChatGPT evaluates first.

**Separate your Perplexity and ChatGPT strategies.** They use fundamentally different retrieval architectures. Perplexity follows fresh organic results and Reddit heavily. ChatGPT's fan-out queries reward semantic title precision and readable URL structure. An action that moves one platform may not move the other.

Signal Check tests your business against the specific queries buyers are using across ChatGPT, Perplexity, and Gemini. If you want to see where the metadata filter is costing you, that's the place to start.