Inside ChatGPT’s Retrieval Stack: the Index, Cache, and Pages It Actually Reads

When ChatGPT cites a web page, where did that page come from? In July, our team at RESONEO captured and dissected 1,200 ChatGPT answers, 88,000 search results, and 26,900 distinct pages to find out.

( Disclosure: I'm the co founder of RESONEO.) We found three layers in how ChatGPT grounds its answers: a discovery index that finds pages, a reading cache that keeps full copies of pages it has fetched, and a small set of pages it opens live. Each layer has its own rules, its own staleness, and its own blind spots. The same pattern also shows up in ChatGPT’s Search Index Serves Small Sites Too, where the practical question is how the signal becomes visible.

How we uncovered ChatGPT's retrieval system

Our Chrome extension records the raw data stream ChatGPT sends to the browser, including fields the interface never displays. Until July 21, that stream contained a field called result_source that named the internal pipeline behind every. The practical read is that brand signals need to be consistent enough for both people and AI systems to form a stable view of the company, its expertise, and its trust signals.

The risk is usually hidden in the execution layer. A page can look fine to a human and still fail for an automated visitor if the form, call to action, rendering path, or confirmation step is not accessible enough for the agent to complete the task.

The cast of pipelines

What we had established before this round, and what still holds: labrador is OpenAI's own retrieval hub. Besides web search, it returns news, academia (arXiv), Reddit, and YouTube results, so it acts as an orchestrator of in house indexes,. For search teams, the important part is not the headline movement by itself. It is whether the shift changes which communities, forums, video surfaces, or publisher pages now satisfy the query better than the old ranking pattern.

The operational question is whether the public business data is complete enough to support the query. Hours, categories, services, reviews, photos, and page content need to reinforce each other so Google can understand the business in a specific situation, not only as a generic listing.

August update: The Think button changes which web ChatGPT sees

Six weeks after the July study, we replayed the same prompts across every ChatGPT configuration: free and paid, Instant, Think, and Thinking, at different effort levels. Until this rollout, Thinking was a paid only product mode. Free users. The search implication is whether the section improves the evidence around the page, not simply whether it adds more wording. Clear entities, crawlable structure, internal links, and useful context are what make the topic easier to evaluate.

The useful check is whether this improves the system behind search performance, not only the words on the page. Internal links, crawlable content, clear entities, current evidence, and a sensible page structure all help the recommendation become easier to trust.

OpenAI's web index works differently from Bing

OpenAI has its own web index, and Bing doesn't supply it. We checked three ways. Only 1.5% of labrador URLs appear in Bing's top 20 for the same fan outs. No labrador snippet matches a Bing snippet. Bing hard caps titles at 75 characters,. The search implication is whether the section improves the evidence around the page, not simply whether it adds more wording. Clear entities, crawlable structure, internal links, and useful context are what make the topic easier to evaluate.

The routing is economic, not technical

Why does OpenAI maintain both? Cost is the answer. In instant mode, ChatGPT must answer in a few seconds, and the user typically pays nothing. So it queries only what OpenAI already owns: the labrador hub. No page gets opened (zero pages. The search implication is whether the section improves the evidence around the page, not simply whether it adds more wording. Clear entities, crawlable structure, internal links, and useful context are what make the topic easier to evaluate.

A reading cache shared

This is the part Jérôme Salomon of Oncrawl dug into with us, and it changed how we think about "ChatGPT visited my page." Alongside the index, ChatGPT maintains a cache of every page it has ever fetched. Not snippets: full. The search implication is whether the section improves the evidence around the page, not simply whether it adds more wording. Clear entities, crawlable structure, internal links, and useful context are what make the topic easier to evaluate.

The analytics blind spot

You've probably seen utm_source=chatgpt.com in your analytics reports. That parameter is appended to the clickable links displayed to users. It tracks outbound clicks. But not every citation carries that tracking parameter. We observed. The measurement question is whether this signal changes a decision, not whether it adds another number to a dashboard. Useful reporting connects visibility, engagement, and business outcomes without pretending every AI influenced journey will produce a clean click path.

The reporting question is whether this signal changes a decision. If it only creates another number in a dashboard, it adds noise. If it helps separate profile activity, website visits, calls, bookings, and direction requests, it can make local performance easier to understand.

The mystery: Citations that come from nowhere

One finding resists explanation, and we want the community's help on it. A share of the results reaching the model carry no snippet at all, just a title and a URL. We see it for arXiv or Reddit results, which come from databases OpenAI has. For search teams, the important part is not the headline movement by itself. It is whether the shift changes which communities, forums, video surfaces, or publisher pages now satisfy the query better than the old ranking pattern.

So, should you write 289 character titles?

The full title reaches the model untruncated, so it's part of your grounding budget. The 200 characters after your H1 are the only body text the model sees in instant mode. As such, moving down the category labels, dates, and widgets that. The strategic issue is whether automated visitors can understand, trust, and complete the same journey a human visitor can. Agent readiness is partly technical, but it is also about clear tasks, accessible flows, and reliable evidence.

The durable battle: Owning the answers to real questions

Optimizing snippet mechanics is a tactic. The strategy is upstream: knowing what people actually ask AI assistants and being the source that answers it. Nobody asks ChatGPT for "a stroller with a 49 cm frame width." They ask for "a. Local visibility depends on whether the details across pages, profiles, categories, reviews, photos, and service descriptions reinforce the same answer for a specific location based query. This connects with Not the Outputs) when the same signal needs a clearer operating decision. A useful companion note is Not the Answers), because it looks at a nearby part of the same system.

What the visibility signal actually changes

What the visibility signal actually changes: inside ChatGPT’s Retrieval Stack: the Index, Cache, and Pages It Actually Reads: the Practical Angle should be treated as a visibility signal, not a standalone headline. Introduction When ChatGPT cites a web page, where did that page come from? In July, our team at RESONEO captured and dissected 1,200 ChatGPT answers, 88,000 search results, and 26,900 distinct pages to find out. ( Disclosure: I'm the co founder of RESONEO.).

What the visibility signal actually changes: the practical question is whether the page, brand evidence, and surrounding content make the answer easier to trust. If that support is weak, search systems can still understand the topic but fail to connect it confidently to the brand.

What the visibility signal actually changes: that is why the response should begin with an audit of the evidence already on the site before creating a new asset. The fastest improvement is often a clearer page, a better internal link, or a stronger explanation of why the brand belongs in the answer.