The direct answer: AI hotel rankings churn so hard that a single week’s position is close to meaningless. Across 11 consecutive weeks (May 11 – July 20, 2026), a research pipeline fired the same 616 hotel-discovery prompts at ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode every Monday — and found that, on average, 70.8% of each engine’s top 10 is replaced from one week to the next. The composite top-10 keeps only 29.2% of its members week-over-week; the top-100 keeps 49.7%. The median hotel that stays on a 500-deep list still moves 55–80 places between Mondays depending on the engine (Nicolas Sitter, AI Hotel Volatility 2026, July 22, 2026). If your marketing report says “we entered ChatGPT’s top 10” based on one week of checks, it will usually un-happen by itself. The honest way to measure AI visibility is a two-layer read: a stable pool membership (are the engines reaching for you at all?) tracked weekly, and a volatile position judged only on a smoothed 4–8 week rolling window.
Why this matters right now
Two numbers explain the timing. First: 91% of hotel chains already use AI somewhere in the business, and another 8% plan to within 12–24 months — but only 28% have a company-wide AI strategy led by senior leadership (h2c benchmark, reported by TravelDailyNews, October 5, 2026). Adoption is near-universal; discipline is rare. Second: a February 2026 study re-ran identical hotel queries minutes apart and found just 50.5% position-1 stability — ask twice, get a different winner half the time (Sitter rankings-consistency study, February 2026). The July study asked the harder question: does that shakiness average out once you aggregate hundreds of prompts into a weekly leaderboard? It does not.
That combination — near-universal AI adoption, almost no strategy layer, and rankings that rewrite themselves weekly — is exactly the environment where hotels overreact to noise: celebrating one-week wins, panicking over one-week losses, and paying vendors for weekly “AI rank” screenshots that are largely reselling churn. This article is about measuring the thing properly.
What the study actually measured
Every Monday for 11 weeks, the pipeline ran 616 identical prompts (56 destinations × 11 templates, English, mixed personas and budgets) at the five engines, then resolved every recommended hotel to a canonical hotel ID and ranked properties by weekly mention count into a 500-deep leaderboard per engine. With the prompt side held fixed, all movement comes from the engines’ answers themselves — meaning this is the churn a hotel tracking its own AI visibility would experience. Grok was excluded after scraper failures; ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode carried the full run (Sitter, July 22, 2026).
The headline numbers, engine by engine
| Engine | Top-10 kept week-over-week | Top-100 kept | Median rank move | Hotels in top-100 all 11 weeks | Never-seen newcomers per week |
|---|---|---|---|---|---|
| ChatGPT | 34% | 51.8% | 80 places | 5 | 2.8% |
| Gemini | 29% | 56.7% | 55 places | 18 | 2.5% |
| Perplexity | 17% | 28.4% | 78 places | 0 | 10.5% |
| Copilot | 33% | 53.5% | 58 places | 6 | 2.1% |
| Google AI Mode | 33% | 57.9% | 60 places | 12 | 2.7% |
(Week-over-week retention averaged over 10 week-pairs; “median rank move” covers hotels present in consecutive top-500s; “newcomers” = share of each week’s top-100 never seen before in any prior week’s top-500. Source: Sitter, AI Hotel Volatility 2026, July 22, 2026.)
Three things jump out. ChatGPT is the steadiest at the very top (34% of its top-10 kept) but its median survivor still slides 80 places a week. Google AI Mode is the most stable at depth (57.9% of its top-100 retained). And Perplexity is the outlier on every axis: in a typical week it keeps fewer than two of its top ten, not a single hotel anywhere held its top-100 for all eleven weeks, and 10.5% of its weekly top-100 are properties never previously seen — a near-lottery.
Three engines, three kinds of churn — and only one is actionable
The study’s most useful contribution is separating same-sized churn into three behaviors, because the right response depends on which one you’re looking at.
Gemini: jitter around a stable core. Track Gemini’s opening-week top-100 forward and the curve is flat — 53% still there after one week, 60% after ten. Hotels drop out and come back: 42 of its 60 surviving cohort members fell out of the top-100 at some point and returned; 18 never left at all, the most of any engine. A Gemini dip is usually noise that mean-reverts. Do not rebuild strategy over it.
ChatGPT: slow drift. Decent weekly retention (51.8% at top-100), but the cohort erodes almost monotonically — 59% → 40% → … → 24% over the season. Hotels that leave tend to stay gone. ChatGPT’s list isn’t jittering around a fixed set; the set itself is migrating. A sustained ChatGPT slide is real drift worth investigating — that’s the engine where a multi-week downtrend should trigger a root-cause check, not a shrug.
Perplexity: near-lottery. The cohort collapses to 31% in a single week and ends at 19%, with zero season-long anchors and the highest newcomer share. In this dataset, a Perplexity swing is close to weather. Judge it on the pool layer (below), not on position.
Model updates add a second volatility source on top: between GPT-5.1 and GPT-5.2, Reddit’s share of hotel-recommendation influence fell from 14.6% to 2.3% (Sitter AI Hotel Landscape 2026, June 24, 2026). A model refresh can re-weight the sources engines read overnight — reshuffling the hotel picture without anything changing at your property. If you want to understand which signals engines weigh when they settle, we broke that down in what moves AI hotel recommendations; the volatility layer above is what happens to those signals week to week.
The reframe that changes your KPI: pool membership vs. position
Here is the study’s reassuring counterweight, and the single most important idea for hotel marketers. Ask not “did the hotel keep its slot?” but “is it still anywhere on the 500-deep list?” — and the picture inverts. A hotel ranked in an engine’s top 10 this week stays somewhere in its top 500 next week 94–99% of the time on four of the five engines (Perplexity, again, at 86%). Even mid-list ranks hold strongly: hotels ranked #11–50 stay in the top 500 at 93%+ on Gemini, Copilot and AI Mode. The leak only opens deep in the tail — ranks 251–500, where hotels sit on one or two weekly mentions, survive at just 30–45% (Sitter, July 22, 2026).
AI hotel visibility behaves like a two-layer asset: a stable pool of hotels the engines keep reaching for, and a weekly shuffle deciding which of them get surfaced on top. Pool membership is durable enough to track weekly and to alarm on. Position is only meaningful smoothed over 4–8 weeks. This is also why one-off “AI visibility” scores tell you so little: they photograph the shuffle, not the pool. A structured travel AI audit builds the baseline — which engines know you at all, which never mention you — before you start arguing about rank.
Who actually survives: the 29 anchors
Across five engines and eleven weeks, only 41 engine-seats never turned over — held by just 29 distinct hotels worldwide. Eight held a permanent top-100 seat on two or more engines; only four managed three: Hotel Monteleone (New Orleans), the Cairo Marriott, Marina Bay Sands, and The Langham Gold Coast. The Cairo Marriott’s median position on ChatGPT across the whole run: #1 (Sitter, July 22, 2026).
Two implications for everyone who is not an anchor. First, the anchor tier is brand-heavy: 28 of the 41 permanent seats belong to chain flagships, and on four of five engines chain-branded hotels survive week-over-week more often than independents — the widest gap on Google AI Mode (61.6% vs. 53.8% retention). Perplexity is the lone inversion, where independents churn slightly less. Second, and more important: for the other ~99.9% of hotels, volatility is the normal condition of AI visibility in 2026. Anchor-tier dominance is not a reason for independents to give up on AI visibility — it is a reason to stop measuring yourself against flagship charts and start measuring the layer you can durably own: pool membership, the same way independents built review-volume and direct-booking moats against the same chains. We cover that build-out — entity consistency, citation depth, answer-shaped content — in our AI visibility program for hotels.
The measurement playbook: five rules
1. Never quote a single week’s rank. With median surviving hotels moving 55–80 places between Mondays, a 4–8 week rolling average is the shortest honest read. This applies double to vendor screenshots and “top 10” badges.
2. Track two KPIs, not one. Pool membership (mentions anywhere across your 25–50 priority prompts, per engine) is your stability metric — check it weekly, alarm on it. Smoothed position is your trend metric — review it monthly. A drop out of the pool is a real event; a slide from #12 to #41 is not.
3. Interpret each engine by its churn type. Gemini drop = likely jitter, wait it out. Multi-week ChatGPT slide = likely drift, investigate content, citations and entity data now. Perplexity swing = weather, judge the pool only. A single dashboard number across engines hides all three signals.
4. When drift is real, fix the layer that drifted. Genuine ChatGPT drift usually traces back to source mix — model updates re-weight what they read (the Reddit 14.6% → 2.3% drop is the cleanest example), and fresh signals (recent reviews, updated facts, new third-party coverage) are what pull a drifting hotel back. Our travel GEO agency runs this exact cadence — fixed prompt set, fixed cadence, rolling-window reporting — for hotels, tour operators and DMCs, and pricing is structured so the baseline audit stands alone.
5. Connect the pool to revenue, not rank. The commercial question is whether pool membership produces AI-referred visits and direct bookings — measurable, as we detailed in measuring AI-driven direct bookings. Rank screenshots convince nobody; referral trends converted against commission savings do.
What not to do
Do not celebrate or panic over weekly rank changes — the data says 71% of top-10 membership turns over every week regardless of anything you did. Do not buy weekly “AI position” reports at face value; ask any vendor what their rolling window is, and walk away if the answer is “this week.” Do not treat all engines identically — responding to Perplexity noise with a content overhaul is spending budget on weather. And do not conclude from chain dominance that AI visibility is only for flagships: the pool layer is exactly where independent hotels have always won — review depth, distinctive positioning, local citation strength — and 94–99% weekly pool retention means that layer, once earned, holds.
FAQ
Why does my hotel’s ChatGPT ranking change every week? Because AI rankings are structurally volatile: an 11-week study of 616 identical prompts found 70.8% of top-10 members replaced weekly on average, with the median surviving hotel moving 55–80 places between Mondays (Sitter, July 22, 2026). Some churn is noise (Gemini), some is slow drift (ChatGPT), and some is near-random (Perplexity). Only multi-week rolling averages distinguish your real position from the shuffle.
What is the difference between AI visibility and AI ranking? Visibility = pool membership: whether an engine mentions your hotel at all across a fixed set of buying-intent prompts. Ranking = where you sit among the mentioned, week to week. The study shows the pool layer is highly stable (94–99% of top-10 hotels stay in the top 500 the next week) while ranking churns constantly. Track visibility weekly; judge ranking monthly, smoothed.
How should hotels measure AI visibility correctly? Run a fixed prompt set (25–50 real guest queries) across ChatGPT, Gemini, Perplexity, Copilot and Google AI Mode on a fixed weekly cadence; track mention presence per engine (the pool) as your alarm metric; assess position only on a 4–8 week rolling average; and interpret drops by engine type — investigate sustained ChatGPT slides, wait out Gemini jitter, ignore Perplexity swings. This is the core method behind a travel AI audit.
Do independent hotels have any chance of stable AI visibility? Yes — at the pool layer. Only 29 hotels worldwide held permanent top-100 seats across the study, mostly chain flagships, so nobody owns stable position except anchors. But pool membership is durable and brand-agnostic: it is built from entity consistency, review recency and depth, third-party citations and answer-shaped content — the same assets independents used to win local search. Chains churn less (up to +7.8 points retention on Google AI Mode), but the pool itself is open.
Should I report AI rankings to my GM or board weekly? Report the pool weekly (presence/absence per engine across priority prompts) and the smoothed position monthly. Weekly rank reporting imports the engines’ own noise into your decision-making and will produce whiplash budget reactions the underlying data does not support.
Related reading
- AI hotel recommendations feel like a lottery — here’s why they’re not
- Measuring AI-driven direct bookings
- What moves AI hotel recommendations
Your AI ranking was never a scoreboard. It is a slot machine sitting on top of a much steadier machine — the pool of hotels the engines actually know. Build the pool, measure the pool, and let the shuffle be the shuffle.