{"@context":"https://schema.org","@type":"BlogPosting","headline":"ChatGPT's Hidden result_source (2026): How It Really Sources Hotel Answers","description":"ChatGPT's web-search stream carries two undocumented fields: result_source (the retrieval tier/vendor that fetched each page) and turn_use_case (a pre-search query classifier). Across 50,899 ChatGPT hotel captures and 30,002 tier-tagged citations: 99.85% of citations come from a licensed labrador tier, the open-web serp tier never appears, the field first showed up 26 May 2026 and rolled out intermittently (peaking at 81.8% before dropping to 7.2%), 37.3% of turns never search the web at all, and within the labrador tier official brand sites are cited 77-86% of the time while aggregator listicles are cited 1-11%.","datePublished":"2026-06-25","dateModified":"2026-06-25","url":"https://nicolassitter.com/research/chatgpt-result-source-retrieval-tiers-2026","category":"research","keywords":["ChatGPT result_source","turn_use_case","how ChatGPT picks sources","ChatGPT web search citations","labrador tier","AI search retrieval tiers","GEO ChatGPT"],"articleSection":"Research","wordCount":1700,"readTime":"7 min","articleBody":"June 2026AI Search · Retrieval\n\n# ChatGPT's hidden `result_source`:how it really sources hotel answers\n\n**TL;DR:** Every web page ChatGPT retrieves carries an undocumented tag, `result_source`, naming the pipeline that fetched it. Across **30,002 hotel citations**, **99.85% come from one licensed tier** (`labrador`) and the open-web `serp` tier never appears. A companion field decides whether the web is searched at all — and **37.3% of hotel questions never search**. Within the licensed tier, ChatGPT cites official brand sites 77–86% of the time while discarding the aggregator listicles it retrieves.\n\nNS\n\nNicolas Sitter\n\nPublished June 25, 2026\n\n50,899\n\nChatGPT captures\n\n30,002\n\nTagged citations\n\n99.85%\n\nFrom labrador tier\n\n37.3%\n\nTurns never search\n\n[Read the Report](#the-field)\n\n[Summary](#executive-summary)[1\\. The hidden field](#the-field)[2\\. The rollout](#rollout)[3\\. The search gate](#query-gate)[4\\. Trusted sources](#trusted-sources)[5\\. Per-tier signature](#per-tier)[Methodology](#methodology)[FAQ](#faq)\n\n## Executive Summary\n\nChatGPT runs hotel search on a licensed retrieval tier — and tells you so in a field users never see.\n\nChatGPT's web-search stream stamps every retrieved page with `result_source` — the pipeline or vendor that fetched it — and classifies each query with `turn_use_case` before it decides to search. Neither is documented. We backfilled both across our full ChatGPT hotel capture history: **50,899 captures and 30,002 tier-tagged citations**.\n\nThe picture is lopsided and actionable. Hotels are sourced almost entirely from the licensed `labrador` tier (99.85%); the open-web `serp` baseline never appears. The field is brand new (first seen 26 May 2026) and was rolled out intermittently. And inside the licensed tier, _retrieved is not cited_: official brand sites win, aggregator listicles get pulled and dropped.\n\nSection 1\n\n## A three-tier sourcing system, hidden in plain sight\n\nEvery `search_result` block in ChatGPT's raw response carries a `result_source` tag next to the publisher attribution and publish date. It takes one of a small set of values — and for hotels, one of them dominates completely.\n\nresult\\_source distribution across 30,002 tier-tagged ChatGPT hotel citations.\n\nresult\\_source\n\nShare of tagged citations\n\nWhat it is\n\nlabrador\n\n99.85%\n\nLicensed / quality-gated content tier\n\nbright\n\n0.14%\n\nBright Data structured web datasets\n\noxylabs\n\n0.01%\n\nScraped open web (Oxylabs is a scraping vendor)\n\nserp\n\n0%\n\nOpen-web baseline — never appears for hotels\n\n### What it looks like in the raw stream\n\n// labrador — licensed editorial / OTA / brand content\n{ ..., \"pub\\_date\":1777852800,\n       \"result\\_source\":\"labrador\",\n       \"attribution\":\"discoverzermatt.com\" }\n\n// bright — Bright Data structured dataset (an individual luxury hotel)\n{ ..., \"result\\_source\":\"bright\",\n       \"attribution\":\"Schweizerhof Zermatt\" }\n\n// oxylabs — scraped open web\n{ ..., \"result\\_source\":\"oxylabs\",\n       \"attribution\":\"Tripadvisor\" }\n\nFor hotel queries, ChatGPT answers almost entirely from a **licensed** content tier (`labrador`, 99.85%). The open-web `serp` tier that appears in other verticals **never shows up** — the model isn't reading the live SERP for hotels, it's reading a curated feed.\n\nSection 2\n\n## A brand-new field — switched on, off, then on again\n\nBecause we have citations from before the field existed, the share carrying a tag traces the rollout. It first appears on **26 May 2026**, peaks the first week of June, then drops sharply — the signature of a staged experiment, not a permanent flip.\n\nchatgpt-result-source-rollout-2026\n\n![ChatGPT result\\_source rollout by week](/_next/image?url=%2Fresearch%2Fchatgpt-result-source-retrieval-tiers-2026%2Frollout.png&w=3840&q=75&dpl=dpl_EiXTouAarQZoA85yXjguDUaCC1DY)\n\nThe tag jumped from **0% to 82%** in two weeks, then fell back to **7%**. If you sample ChatGPT's internals on a handful of queries over a few days, you can easily catch a tier “everywhere” one week and “gone” the next. Population-scale capture is what separates a real behaviour from a snapshot artefact.\n\nSection 3\n\n## Does your question even reach the web?\n\nBefore any retrieval, `turn_use_case` files the query into a bucket that decides which pipelines fire. For hotels, the single most common bucket is `text` — which means **no web search at all**.\n\nturn\\_use\\_case distribution across ChatGPT hotel captures.\n\nturn\\_use\\_case\n\nShare of turns\n\nHits the web?\n\ntext\n\n37.3%\n\nNo — answered from training data\n\nsearch\n\n31.8%\n\nYes\n\nlocal\n\n26.3%\n\nYes (maps / places)\n\ninstant search\n\n3.3%\n\nYes\n\ninstant answers\n\n1.3%\n\nPartial\n\nthinking / unknown\n\n0.1%\n\n—\n\n**More than a third of hotel questions never search the web.** For those, citability is irrelevant — the answer comes from what the model already memorised. The wording of the query, not just the topic, decides whether your content can be surfaced at all.\n\nSection 4\n\n## Retrieved ≠ cited: who ChatGPT actually trusts\n\nThese are the top domains ChatGPT _retrieves_ inside the `labrador` tier, with the share of those retrievals it _actually cites_. The gap is the whole story.\n\nchatgpt-labrador-retrieved-vs-cited-2026\n\n![Retrieved vs cited rate for top labrador hotel domains](/_next/image?url=%2Fresearch%2Fchatgpt-result-source-retrieval-tiers-2026%2Fretrieved-vs-cited.png&w=3840&q=75&dpl=dpl_EiXTouAarQZoA85yXjguDUaCC1DY)\n\nTop labrador-tier hotel domains: times retrieved vs share actually cited.\n\nDomain\n\nRetrieved\n\nCited %\n\nType\n\ntripadvisor.com\n\n2,209\n\n54%\n\nReview aggregator\n\nmarriott.com\n\n1,019\n\n68%\n\nBrand\n\nexpedia.com\n\n926\n\n29%\n\nOTA\n\nbooking.com\n\n632\n\n60%\n\nOTA\n\noyster.com\n\n411\n\n20%\n\nAggregator\n\nhyatt.com\n\n399\n\n84%\n\nBrand\n\nhilton.com\n\n360\n\n80%\n\nBrand\n\nfourseasons.com\n\n321\n\n82%\n\nBrand\n\n**Own your brand content.** Official brand `.com` sites are cited 77–86% of the time when retrieved; aggregator listicles 1–11%. ChatGPT pulls the listicles, then leans on the official source. For hotels, a clean first-party page beats a placement on a “best hotels” roundup.\n\nSection 5\n\n## Each tier has a different publisher fingerprint\n\nThe three tiers aren't redundant — they map to _how_ a page was acquired, and it shows in what each one carries.\n\nPer-tier publisher signature observed in the ChatGPT hotel data.\n\nTier\n\nCharacter\n\nRepresentative sources\n\nlabrador\n\nLicensed / quality-gated — OTAs, big-brand chains, established editorial\n\ntripadvisor.com, marriott.com, booking.com, expedia.com, hyatt.com, hilton.com, fourseasons.com, timeout.com\n\nbright\n\nBright Data structured datasets — premium directories + individual luxury properties\n\nFive Star Alliance, Tablet Hotels, Forbes Travel Guide, montcervinpalace.ch, tajhotels.com, agoda.com\n\noxylabs\n\nScraped open web — long tail incl. social + individual hotel sites\n\ninstagram.com, resortpass.com, thebarnett.com\n\nThe plumbing tell: an individual hotel's own website or its Instagram reaches ChatGPT through the **scraper** tiers (`bright` / `oxylabs`), not the licensed `labrador` feed. Big brands and OTAs are licensed in; everyone else is crawled ad-hoc.\n\nMethodology\n\n## Study Design\n\n### Data Collection\n\n-   50,899 ChatGPT hotel captures via Bright Data, 25 Dec 2025 – 22 Jun 2026.\n-   `result_source` and `turn_use_case` parsed out of the raw SSE response stream and joined to 30,002 flattened per-citation rows.\n-   ChatGPT-only — no equivalent field is emitted by Gemini, Perplexity, Copilot or Google AI Mode.\n\n### Caveats\n\n-   One vertical (hotels) and one collection method; tier mix is query-type dependent.\n-   `serp` never appears for hotels (0 of 30,002) but may surface in other verticals.\n-   The field is ~4 weeks old and intermittent; `bright` (n=42) and `oxylabs` (n=3) samples are tiny — directional, not definitive.\n-   _Cited %_ is share-of-retrievals-cited, not a ranking guarantee.\n\n**Open data.** Headline stats and the underlying tables are published as CSV: [summary.csv](/data/chatgpt-result-source-retrieval-tiers-2026/summary.csv), [weekly\\_rollout.csv](/data/chatgpt-result-source-retrieval-tiers-2026/weekly_rollout.csv), [labrador\\_top\\_sources.csv](/data/chatgpt-result-source-retrieval-tiers-2026/labrador_top_sources.csv), [turn\\_use\\_case.csv](/data/chatgpt-result-source-retrieval-tiers-2026/turn_use_case.csv).\n\n### Summarize with AI\n\nFAQ\n\n## Frequently Asked Questions\n\n## Continue Reading\n\nMore field tests of how AI engines find and cite sources.\n\n[AI Search for Yoga Studios in Paris](/research/yoga-studios-paris-ai-search-2026)\n\n[All Research](/research)","author":{"@type":"Person","name":"Nicolas Sitter","url":"https://nicolassitter.com/about","sameAs":["https://www.linkedin.com/in/nicolassitternolleau/","https://github.com/Nicositter88","https://hotelrank.ai"]},"publisher":{"@type":"Person","name":"Nicolas Sitter","url":"https://nicolassitter.com"},"image":"https://nicolassitter.com/api/og/chatgpt-result-source-retrieval-tiers-2026","mainEntityOfPage":{"@type":"WebPage","@id":"https://nicolassitter.com/research/chatgpt-result-source-retrieval-tiers-2026"},"tags":["AI Search","ChatGPT","Citations","GEO","Retrieval"],"sameAs":["https://hotelrank.ai/research/chatgpt-result-source-retrieval-tiers-2026"],"alternateFormat":{"html":"https://nicolassitter.com/research/chatgpt-result-source-retrieval-tiers-2026","json":"https://nicolassitter.com/api/post/chatgpt-result-source-retrieval-tiers-2026","rss":"https://nicolassitter.com/rss.xml"},"datasets":[{"name":"summary","contentUrl":"https://nicolassitter.com/data/chatgpt-result-source-retrieval-tiers-2026/summary.csv","encodingFormat":"text/csv"}]}