TL;DR Summary:
Hidden Pipelines Switch: ChatGPT routes searches through four invisible systems, causing sources to change and citation tracking to become unreliable.Readability Drops Consistency: Switching pipelines reduces URL overlap by roughly 45%, meaning readable plain HTML and clear facts are essential for visibility.Query Types Matter: Different pipelines favor specific content like publishers or shopping data, so your page may appear or vanish based on the question type.Your content might be invisible to ChatGPT even when it answers questions in your space. New research published in July 2026 shows that ChatGPT switches between hidden search pipelines, and those switches change which sources get cited.
ChatGPT citations depend on hidden retrieval pipelines
Two researchers found that ChatGPT uses internal labels to route searches through different systems. Chris Green and Suganthan Mohanadasan both identified four distinct pipelines: Labrador, Bright, Oxylabs, and SERP. These labels sit behind the answer. You never see them in the citation cards ChatGPT shows you.
Green tested 1,000 prompts up to 10 times each and captured 9,946 completed search runs. Labrador handled 88.1% of primary search sources in his dataset. Bright accounted for 9.9%, Oxylabs for 1.7%, and SERP for 0.3%.
But 11.6% of prompts changed their primary search source when repeated. That small percentage creates a big problem for anyone trying to track visibility.
Source switching drops citation consistency by 45%
When ChatGPT switched pipelines, URL overlap dropped from 0.273 to 0.149. Domain overlap fell from 0.265 to 0.155. Green calculated that as roughly 45% lower URL overlap and 42% lower domain overlap.
That means the same question can pull from completely different sources depending on which pipeline handles the search. Your page might appear in one answer and vanish in the next, even when the visible response looks similar.
Each pipeline favors different content types
Mohanadasan inspected two days of raw ChatGPT network traffic from one Pro account and logged about 1,240 source records. He found a result_source field attached to web results, with the same four values Green identified.
Labrador appears to include established publishers and reference sites. Bright connects to Bright Data. Oxylabs connects to Oxylabs. SERP functions as an open-web baseline and appears mostly in news-style results.
Mohanadasan saw Bright play a larger role in his sample, especially for commercial, shopping, finance, weather, and local queries. Green's dataset showed Labrador dominating most prompts. The difference suggests that query type determines which pipeline ChatGPT selects.
Some queries skip web search entirely
Mohanadasan found that ChatGPT classifies queries before searching, using a turn_use_case field. Some prompts get filed as text and skip web search, even when they sound current. In those cases, no page gets fetched, cited, or used as evidence.
More complex queries behave differently. Mohanadasan found that ChatGPT can fan out into many searches for thinking-style prompts. It might run site: probes, pricing checks, and searches for unnamed competitors. That changes which pages enter the answer process because ChatGPT searches rewritten queries, direct site probes, or follow-up checks instead of the exact phrase you typed.
Fetched pages often go uncited
Mohanadasan separated three outcomes: fetched, cited, and mentioned. A page can be fetched into ChatGPT's context without being shown to users. It can be cited as the source behind a specific sentence. Or a brand can be mentioned without being the source of the claim.
In his small commercial-query sample, Reddit and YouTube were both fetched often. Reddit was cited. YouTube was not. He attributed the gap to text availability. Reddit threads expose text. YouTube search results often provide metadata rather than full video transcripts.
Vendor pages were cited for their own facts, such as prices and specs. Third-party pages were more likely to support broader recommendation claims.
How ChatGPT citations change based on readability
There's no single ChatGPT visibility result to measure. Your page might never be considered if ChatGPT skips search, uses another retrieval source, or finds a clearer third-party page to support the claim.
Readable pages were easier to use. Both analyses showed that ChatGPT's source selection depends partly on what it retrieves and reads.
Mohanadasan found cases where ChatGPT appeared to prefer official pricing pages, then fell back to third-party sources when prices were hidden behind JavaScript or otherwise hard to parse. Green's results showed that source routing changed which URLs and domains entered the answer set.
Plain HTML matters more now. Crawlable facts, clear pricing and specs, strong third-party coverage, and text-heavy pages become more important when source selection depends on retrieval and readability.
Why pipeline switching complicates AI visibility tracking
You cannot see which pipeline ChatGPT used by looking at the answer. The citation cards show the final sources, not the routing logic that selected them. That means you need to test the same prompt multiple times to understand how often your content appears.
Tools like AI Mentions help track when and how your brand appears across these hidden pipelines by monitoring actual ChatGPT responses in real time, revealing which retrieval sources are surfacing your content and when citation patterns shift.
Because these pipeline switches happen invisibly to users, continuous monitoring across multiple queries becomes essential to understanding your true AI visibility footprint. A single test tells you what happened once. It does not tell you which pipeline ran or whether the next user will see the same sources.
What to optimize when citations depend on hidden sources
Start with text. Make sure your pages expose facts in plain HTML rather than hiding them behind JavaScript. ChatGPT appears to prefer pages it can parse quickly.
List prices and specs clearly. Mohanadasan's research suggests ChatGPT prefers official sources for factual claims when those sources are readable. If your pricing lives behind a login or requires interaction, ChatGPT will cite a third-party page instead.
Build third-party coverage. Vendor pages get cited for their own facts. Third-party pages support broader claims and recommendations. If you want to appear in ChatGPT answers about category comparisons or recommendations, you need strong external coverage that ChatGPT can fetch and read.
Monitoring tools can identify the gap between fetch and citation by tracking when your domain appears in ChatGPT's context but doesn't receive visible credit, helping you understand whether the issue is retrieval, readability, or relevance.
Tracking citation gaps across repeated queries
Green's data shows that most prompts stay on one retrieval source, but 11.6% switch. That percentage is small, but it means you cannot trust a single test to represent your visibility.
You need to run the same prompt multiple times to see how often your content appears and which pipeline variations include or exclude you. You also need to test variations of the same question because ChatGPT might rewrite queries before searching, and those rewrites can pull from different sources.
AI Mentions addresses the measurement problem these pipeline switches create. It tracks which specific queries trigger competitor citations instead of yours, revealing gaps in your content that prevent citation eligibility. Because pipeline routing changes which pages enter the answer set, you need tools that monitor real ChatGPT responses over time rather than relying on single snapshots. You can explore how AI Mentions tracks citation consistency across hidden pipeline changes.
