{"@context":"https://schema.org","@type":"BlogPosting","headline":"ChatGPT's result_source for Flights (2026): A Different Retrieval Stack","description":"A flights replication of the hidden result_source study: 67 frozen prompts × US+GB × 15 iterations = 2,007 ChatGPT captures (July 2026), 23,308 fetched documents (82.8% tier-tagged), 4,003 cited documents. The licensed labrador tier drops from 99.85% (hotels) to 74.6%, bright carries 23.1% (skyscanner.net is its #1 domain), serp appears for the first time (0 of 30,002 hotel citations), only 8.6% of flight questions skip the web (hotels: 37.3%), no local/maps bucket and zero booking widgets, half of cited documents bypass the tagged retrieval layer via an untagged footnote path, Reddit is cited on 83% of its fetches, and airline official sites are cited almost every time they are fetched — but only when the user names the airline (68-83% of named captures vs 16-28% cited on anonymous route queries).","datePublished":"2026-07-21","dateModified":"2026-07-21","url":"https://nicolassitter.com/research/chatgpt-flights-retrieval-tiers-2026","category":"research","keywords":["ChatGPT flights","ChatGPT result_source","how ChatGPT finds flights","Skyscanner ChatGPT","Kayak ChatGPT","AI flight search","AI search retrieval tiers","GEO airlines"],"articleSection":"Research","wordCount":1800,"readTime":"8 min","articleBody":"July 2026AI Search · Retrieval · Flights\n\n# ChatGPT's `result_source`, part two:flights don't fly on the hotel stack\n\n**TL;DR:** We re-ran our hidden `result_source` study on flights: **2,007 fresh ChatGPT captures** from the US and the UK. The hotel-style tier monopoly is gone — the licensed `labrador` tier falls from 99.85% to **74.6%**, Bright Data's tier carries **23%**, and the open-web `serp` tier appears for the first time. Only **8.6% of flight questions skip the web** (hotels: 37.3%). And the most-trusted flight source isn't an airline, an OTA or a metasearch — it's **Reddit, cited on 83% of its fetches**.\n\nNS\n\nNicolas Sitter\n\nPublished July 21, 2026\n\n2,007\n\nChatGPT captures\n\n23,308\n\nFetched documents\n\n74.6%\n\nlabrador (was 99.85%)\n\n83%\n\nReddit cite rate\n\n[Read the Report](#tier-mix)\n\n[Summary](#executive-summary)[1\\. The tier mix](#tier-mix)[2\\. The search gate](#search-gate)[3\\. Two doors](#two-doors)[4\\. Retrieved ≠ cited](#retrieved-vs-cited)[5\\. Naming the airline](#named-airlines)[6\\. Where to book](#where-to-book)[7\\. US vs GB](#us-vs-gb)[Methodology](#methodology)[FAQ](#faq)\n\n## Executive Summary\n\nThe retrieval stack behind ChatGPT is vertical-specific — and flights use all of it.\n\nOur [hotel study](/research/chatgpt-result-source-retrieval-tiers-2026) found ChatGPT sourcing hotel answers almost entirely (99.85%) from one licensed retrieval tier. To test whether that generalises, we ran **67 frozen flight prompts × US + UK × 15 iterations = 2,007 captures** over four days in July 2026 and parsed the same hidden fields from the raw stream.\n\nIt doesn't generalise. Flights run on a genuinely mixed stack (`labrador` 74.6%, `bright` 23.1%, `oxylabs` 1.7%, `serp` 0.6%), almost every flight question triggers a live search, half of the documents ChatGPT actually cites bypass the tagged retrieval layer entirely — and citation trust looks nothing like hotels: flat across airlines, OTAs and metasearch, with one glaring exception. Reddit.\n\nSection 1\n\n## The tier monopoly is gone\n\nEvery page ChatGPT retrieves carries the undocumented `result_source` tag naming the pipeline that fetched it. For hotels, one licensed tier had a 99.85% monopoly. For flights, 82.8% of the 23,308 fetched documents carry a tag — and the mix is completely different.\n\nchatgpt-flights-tier-mix-2026\n\n![ChatGPT retrieval-tier mix, hotels vs flights](/_next/image?url=%2Fresearch%2Fchatgpt-flights-retrieval-tiers-2026%2Ftier-mix.png&w=3840&q=75&dpl=dpl_Ezg5UjTRsvGvriKW6YACftiwDdU2)\n\nresult\\_source distribution among tier-tagged documents (flights: 19,305 tagged fetches; hotels: 30,002 tagged citations).\n\nresult\\_source\n\nFlights (this study)\n\nHotels (Dec 2025–Jun 2026)\n\nWhat it carries for flights\n\nlabrador\n\n74.62%\n\n99.85%\n\nLicensed tier — Google, OTAs, metasearch, editorial\n\nbright\n\n23.05%\n\n0.14%\n\nBright Data datasets — skyscanner.net is its #1 domain\n\noxylabs\n\n1.70%\n\n0.01%\n\nScraped open web — the points/deals blogosphere\n\nserp\n\n0.63%\n\n0% (never appeared)\n\nLive SERP — fare-comparison and metasearch pages\n\nThe `serp` tier — **zero appearances in 30,002 hotel citations** — shows up for flights, and Bright Data's `bright` tier goes from a rounding error to nearly a quarter of tagged fetches. Notably, **skyscanner.net is the single biggest domain in the bright tier** (388 documents, ahead of google.com and expedia.com): metasearch inventory reaches ChatGPT through the scraper pipeline, not the licensed feed.\n\nSection 2\n\n## The search gate barely exists for flights\n\nBefore any retrieval, `turn_use_case` classifies the query and decides whether the web is touched at all. For hotels, 37.3% of questions were answered from training data. Flights are the opposite: prices move too fast to memorise.\n\nturn\\_use\\_case distribution across 2,007 flight captures vs the hotel study.\n\nturn\\_use\\_case\n\nShare of flight turns\n\nHotels\n\nHits the web?\n\ninstant search\n\n77.7%\n\n3.3%\n\nYes\n\ninstant answers\n\n8.7%\n\n1.3%\n\nPartial\n\ntext\n\n8.6%\n\n37.3%\n\nNo — training data only\n\nsearch\n\n5.0%\n\n31.8%\n\nYes\n\nlocal (maps)\n\n0%\n\n26.3%\n\n—\n\nTwo pipelines that dominate hotels are simply absent for flights: the `local`/maps bucket (26.3% of hotel turns) never fires, and in 2,007 captures we saw **zero shopping or booking widgets** — no Google-Flights-style panel exists inside ChatGPT. Flight answers are prose plus citations, which makes _who gets cited_ the entire visibility game.\n\nSection 3\n\n## Citations come through two doors — and half bypass the tagged layer\n\nA parsing detail with a real finding inside. The URLs ChatGPT actually cites are stamped `?utm_source=chatgpt.com`, so they never string-match the tagged fetch layer. Joining them back by canonical URL splits the 4,003 cited documents into three groups:\n\nProvenance of 4,003 cited documents relative to the tier-tagged fetch layer.\n\nCited-document provenance\n\nShare\n\nWhat it means\n\nFootnote-only\n\n50.2%\n\nNo tagged fetch counterpart at all — enters via a separate, untagged browse path\n\nDomain-only match\n\n35.1%\n\nSame domain was fetched with a tier tag, but a different page\n\nCanonical-URL join\n\n14.6%\n\nThe exact tagged page — overwhelmingly google.com via labrador\n\n**Half of what ChatGPT cites for flights never appears in the tagged retrieval layer.** The tiered system supplies the bulk fetch (Google, OTA and metasearch pages); a second, untagged path supplies the long tail that actually gets footnoted — route utilities like flightsfrom.com, Reddit threads, editorial, airline pages, Skyscanner and Kayak. The hotel data never showed this split.\n\nSection 4\n\n## Retrieved ≠ cited: Reddit is the only source ChatGPT treats like a brand site\n\nFor hotels, official brand sites were cited on 77–86% of their retrievals. For flights the citation curve is flat — almost everything sits between 3% and 36% — with one outlier at hotel-brand-site altitude.\n\nchatgpt-flights-retrieved-vs-cited-2026\n\n![Retrieved vs cited rate for top flight domains in ChatGPT](/_next/image?url=%2Fresearch%2Fchatgpt-flights-retrieval-tiers-2026%2Fretrieved-vs-cited.png&w=3840&q=75&dpl=dpl_Ezg5UjTRsvGvriKW6YACftiwDdU2)\n\nTop flight domains: times retrieved vs share actually cited, all 2,007 captures.\n\nDomain\n\nRetrieved\n\nCited %\n\nType\n\ngoogle.com\n\n2,023\n\n21%\n\nMetasearch\n\nflightsfrom.com\n\n1,125\n\n26%\n\nRoute utility\n\nexpedia.com\n\n1,040\n\n10%\n\nOTA\n\ncheapflights.co.uk\n\n686\n\n22%\n\nMetasearch\n\nflightconnections.com\n\n673\n\n28%\n\nRoute utility\n\nkayak.com\n\n669\n\n15%\n\nMetasearch\n\nskyscanner.net\n\n535\n\n10%\n\nMetasearch\n\nskyscanner.com\n\n421\n\n22%\n\nMetasearch\n\n**Reddit is cited on 83% of the fetches where it appears** — four times the rate of Google, five times Kayak, eight times Expedia. This mirrors our [hotel price study](/research/chatgpt-hotel-price-sources-2026), where Reddit hit a 100% cite rate: when ChatGPT wants a “real person” answer about travel, it reads the aggregators and quotes the forum.\n\nSection 5\n\n## Brand authority is query-scoped: name the airline and its site becomes the answer\n\nThe hotel study's headline — official brand sites get cited when retrieved — does replicate for airlines, but only when the traveller _names the brand_. On “What is {airline}'s checked baggage allowance?” the airline's own site is fetched in 68–83% of captures and cited in essentially every capture where it was fetched:\n\nNamed-airline prompts (baggage / economy experience): share of captures where the airline's own website was fetched and cited.\n\nAirline\n\nCaptures\n\nOwn site fetched\n\nOwn site cited\n\nRyanair\n\n60\n\n83%\n\n83%\n\nSingapore Airlines\n\n60\n\n78%\n\n78%\n\nEmirates\n\n60\n\n78%\n\n73%\n\nBritish Airways\n\n60\n\n77%\n\n77%\n\nAir France\n\n60\n\n75%\n\n75%\n\nDelta\n\n60\n\n75%\n\n75%\n\nThe same airline sites sit at **16–28% cited on anonymous route queries**. ChatGPT doesn't “trust airlines” globally — it trusts the official source _for questions about that brand_. For airlines, owning your baggage/fees/policy pages is the highest-certainty AI visibility asset; for route and price queries, the battle happens on metasearch, utilities and Reddit.\n\nSection 6\n\n## “Where should I book?” — the answer set ChatGPT actually cites\n\nSeven advice prompts (“best website for cheap flights”, “book direct or through a third party?”…) probe which booking brands ChatGPT puts in front of travellers. The cited set is deal newsletters, consumer editorial, Reddit — and the metasearch engines:\n\nTop cited domains on the booking-advice prompt family (420 captures).\n\nDomain\n\nCitations on advice prompts\n\nType\n\ngoing.com\n\n54\n\nDeal alerts\n\nkiplinger.com\n\n49\n\nFinance editorial\n\nreddit.com\n\n48\n\nForum / UGC\n\nkayak.com\n\n42\n\nMetasearch\n\nskyscanner.com\n\n42\n\nMetasearch\n\nmoneysavingexpert.com\n\n38\n\nConsumer editorial (UK)\n\nwhich.co.uk\n\n34\n\nConsumer editorial (UK)\n\nkayak.co.uk\n\n32\n\nMetasearch\n\nhelp.skyscanner.net\n\n29\n\nMetasearch help pages\n\ntravelandleisure.com\n\n25\n\nEditorial\n\nSkyscanner and Kayak are squarely inside ChatGPT's booking-advice answer set — including, in Skyscanner's case, via its **help-centre pages** (help.skyscanner.net, 29 citations), which ChatGPT uses to explain how the product works. Support content is citation inventory.\n\nSection 7\n\n## US vs UK: same engine, different storefronts\n\nEvery prompt ran from both a US and a UK vantage point. The sourcing strategy barely moves — the tier mix and class mix are near-identical — but the _domains_ localise.\n\nShare of captures where the brand appears among fetched documents, by capture country.\n\nBrand\n\nUS presence (share of captures)\n\nUK presence\n\nLocalisation\n\nGoogle\n\n49.2%\n\n45.8%\n\ngoogle.com both\n\nKayak\n\n33.0%\n\n31.9%\n\nkayak.com vs kayak.co.uk\n\nExpedia\n\n29.9%\n\n29.3%\n\nexpedia.com vs expedia.co.uk\n\nSkyscanner\n\n24.1%\n\n28.0%\n\nskyscanner.com vs skyscanner.net / .gg / .ie\n\nThe hotel-study null result — IP doesn't change sourcing — mostly holds for flights at the strategy level. But ChatGPT picks the **country storefront**: UK captures fetch skyscanner.net, kayak.co.uk and expedia.co.uk, and UK-specific editorial (moneysavingexpert.com, which.co.uk) enters the cited set. Multi-domain brands should treat every country TLD as its own AI surface.\n\nMethodology\n\n## Study Design\n\n### Data Collection\n\n-   67 frozen prompts × 2 countries (US, GB) × 15 iterations = 2,007 captures (3 lost to timeouts), 18–21 July 2026, via Bright Data.\n-   Four prompt families: 10 routes × 4 conditions (cheapest / round-trip price / best airline / nonstop, live August dates), 8 named airlines × 2, 7 booking-advice prompts, 4 no-search-bait questions.\n-   `result_source` and `turn_use_case` parsed from the raw SSE stream: 23,308 fetched documents (82.8% tier-tagged), 4,003 cited documents joined back by canonical URL (footnote URLs carry `?utm_source=chatgpt.com`).\n\n### Caveats\n\n-   One platform (ChatGPT), one collection vendor, a 4-day window — tier shares are a snapshot; the hotel study showed the tag is dialled in and out over weeks.\n-   10 routes and 8 airlines; a different panel shifts the domain table (not the structural findings).\n-   `serp` (n=122) and `oxylabs` (n=329) are small samples — directional.\n-   Fresh matrix vs the hotels' 6-month backfill: comparisons are fetched-docs vs tagged-citations where noted.\n\n**Open data.** All tables are published as CSV: [summary.csv](/data/chatgpt-flights-retrieval-tiers-2026/summary.csv), [tier\\_mix.csv](/data/chatgpt-flights-retrieval-tiers-2026/tier_mix.csv), [turn\\_use\\_case.csv](/data/chatgpt-flights-retrieval-tiers-2026/turn_use_case.csv), [top\\_domains.csv](/data/chatgpt-flights-retrieval-tiers-2026/top_domains.csv), [named\\_airlines.csv](/data/chatgpt-flights-retrieval-tiers-2026/named_airlines.csv), [advice\\_cited.csv](/data/chatgpt-flights-retrieval-tiers-2026/advice_cited.csv), [country\\_split.csv](/data/chatgpt-flights-retrieval-tiers-2026/country_split.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[ChatGPT's Hidden result\\_source — Hotels](/research/chatgpt-result-source-retrieval-tiers-2026)[How ChatGPT Pulls Hotel Prices](/research/chatgpt-hotel-price-sources-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-flights-retrieval-tiers-2026","mainEntityOfPage":{"@type":"WebPage","@id":"https://nicolassitter.com/research/chatgpt-flights-retrieval-tiers-2026"},"tags":["AI Search","ChatGPT","Flights","Citations","GEO","Retrieval"],"sameAs":["https://hotelrank.ai/research/chatgpt-flights-retrieval-tiers-2026"],"alternateFormat":{"html":"https://nicolassitter.com/research/chatgpt-flights-retrieval-tiers-2026","json":"https://nicolassitter.com/api/post/chatgpt-flights-retrieval-tiers-2026","rss":"https://nicolassitter.com/rss.xml"},"datasets":[{"name":"summary","contentUrl":"https://nicolassitter.com/data/chatgpt-flights-retrieval-tiers-2026/summary.csv","encodingFormat":"text/csv"}]}