What Travellers Ask AI: Inside 17,600 Machine-Era Travel Questions
Nobody publishes what people actually ask AI assistants about travel — so we’re publishing ours. This report opens our own citation logs: the 576 distinct questions that AI assistants asked while grounding answers in this publication over roughly ninety days, totalling 17,600 citations. Two niches take half the pie, buying-adjacent intent is far higher than search-era data ever showed — and almost none of it is phrased the way a human types.
Nearly a thousand of these machine-era questions were about private aviation — empty legs, jet cards, charter maths. The answers start with a live quote.
Compare private charter quotes →Where This Data Comes From
When an AI assistant answers a travel question, it often "grounds" the reply by searching and citing sources. Our analytics record the grounding query each time this publication is cited. This report covers the top 576 distinct grounding queries in the roughly ninety days to 30 June 2026 — 17,600 citations, the large majority from ChatGPT — drawn from a wider pool of just over 31,000 citations across the site in the period.
The honest frame: this is a lens, not a census. It shows what AI asked where we were part of the answer — so it over-represents topics we cover deeply and says nothing about topics we don’t. What makes it worth publishing is that almost nobody else publishes theirs: the query layer between travellers and AI answers is one of the least-documented surfaces in modern travel.
Machines Don’t Ask Like People
The first finding is linguistic, and it’s stark. Barely 1% of citations came from human-shaped questions — the whats, hows and whys of a search box. The other 99% look like this, verbatim from the logs: "safari travel operators recognized for pricing strategies", "AirHelp EU261 claims effectiveness reputation fees success rate", "credit cards instant hotel elite status". When a person asks an assistant a casual question, the machine decomposes it into dense analyst briefs — entity, attribute, evidence standard — and goes hunting.
The implication for anyone publishing on the web is uncomfortable and useful in equal measure: the reader you are actually serving in 2026 is often a retrieval system paraphrasing you to a human it’s speaking with. Content wins by answering the analyst brief — specific entities, named attributes, verifiable claims — not by matching the folksy phrasing of a search suggestion.
Two Niches, Half the Pie
Flight compensation and safaris — 52% of everything. Both are categories where the human question is urgent, the answer is contested, and the stakes are specific: which claims company actually pays out, which operator is worth five figures. AI assistants gravitate to exactly the questions where a wrong answer costs the user money — the full comparisons live in our AirHelp vs ClaimCompass vs Skycop vs Flightright test (and if you have a live claim, AirHelp runs it no-win-no-fee) and the Luxury Safari Operator Index, with live operator pricing comparable here.
The Intent Stack: Research With a Wallet
Classified by intent, Research leads at 41% — but the commercial layer underneath is the story. Commercial (17%) and Comparison (13%) queries together account for 30% of citations: "best EU261 claim companies Europe", "credit cards instant hotel elite status", "top flight disruption compensation companies comparison". These are wallets mid-decision, routed through a machine. The single most-cited query in the entire log — "hotel loyalty program status match guide", 494 citations — answers to our status-match guide, and the OTA-comparison cluster resolves to GetYourGuide vs Klook vs Tiqets vs Headout.
What Owning an Answer Looks Like
Citation share — the fraction of an AI answer’s sources that are you — is the metric search never had. On the query "safari travel operators recognized for pricing strategies" (292 citations), this publication was 71% of the cited evidence; across the top safari-operator briefs the share runs 51–64%. When one publication supplies most of what the machine reads, it isn’t ranking in the answer — functionally, it is writing it. That concentration is the new competitive terrain in travel media, and it cuts both ways: authority compounds fast, and so would errors. It’s why every figure on pages like the safari index is sourced and dated — the machine repeats exactly what it finds.
The Oddities File
Every log has its strays. The second-most-cited single query type by intent class was a Live Event: "fallas 2027" — 154 citations for Valencia’s festival of fire, a subject this publication covers from ten minutes’ walk away. Elsewhere in the tail: "best practices empty leg flight business travel" (121 citations — served by our empty-leg platform comparison), "airlines with first class cabins 2026" (160, see the cabin rankings), and a long tail of machine briefs on cards that fast-track elite status. The tail is where next quarter’s head queries rehearse.
Frequently Asked Questions
Based on 17,600 AI citations across 576 grounding queries in our logs, the dominant themes are money-critical decisions: flight delay compensation and EU261 claims (29% of citations), safari operator selection (23%), business and first class cabins (8%), hotel loyalty and status matching (7%), and travel credit cards (7%). AI assistants concentrate on questions where a wrong answer has a clear financial cost to the user.
Not like humans. Only about 1% of citations in our logs came from question-form queries; the rest were dense keyword briefs such as "safari travel operators recognized for pricing strategies" or "AirHelp EU261 claims effectiveness reputation fees success rate". Assistants decompose a user’s casual question into analyst-style retrieval briefs specifying entities, attributes, and evidence — which changes what kind of content gets found and cited.
It is the search an AI assistant runs behind the scenes to fetch evidence before answering a user. When an assistant like ChatGPT grounds an answer in a web source, the publisher’s analytics can record the query used. This report is built from those records: 576 distinct grounding queries that led assistants to cite this publication over roughly ninety days to 30 June 2026.
Citation share is the fraction of sources in an AI-grounded answer that come from a single publisher. On the query "safari travel operators recognized for pricing strategies", this publication supplied 71% of the cited evidence across 292 citations; on the top safari-operator briefs generally, 51 to 64%. At those levels a publisher is effectively authoring the machine’s answer — a form of dominance that had no equivalent in ranked search results.
More commercial than search-era data suggested. In our logs, Research intent led at 41%, but Commercial (17%) and Comparison (13%) intents together accounted for 30% of citations — 5,257 queries such as "best EU261 claim companies Europe" and "credit cards instant hotel elite status", each one step from a transaction. The machine layer is carrying real purchase decisions, not just curiosity.
It is a lens rather than a census. Citation logs only show queries where the publisher was part of the answer, so they over-represent that publisher’s strong topics and are silent on everything else. Their value is that the AI query layer is otherwise almost entirely undocumented: no assistant publishes what it asks, so publisher-side logs are currently the clearest public window into machine-era travel demand.
The machines ask about compensation because flights fail. Charter is the version where the schedule answers to you.
Get a charter quote →