My AI Visibility
I spend my time measuring how AI engines recommend hotels. This page turns the method on me: a frozen panel of 31 prompts about my own niche — AI search for hotels — runs against five engines every Tuesday, and every result lands here in public, including the zeros.
Reference rate per engine, week by week
Share of each engine's 93 weekly captures that list a nicolassitter.com URL among the answer's sources. With 5 runs in, the lines are starting to mean something. Once the action log starts, vertical markers show what I shipped and when.
A reference means a nicolassitter.com URL appears in the answer’s source list — extracted from each engine’s structured sources, not parsed from the prose. For ChatGPT, Copilot and Gemini that source is one the engine actually cited; for Google AI Mode it can also be a result the engine surfaced without citing inline.
A mention is the name appearing in the answer text without a link — detected by an NER pass over every answer, never solicited by the prompts. The two charts split the old combined metric: “Nicolas Sitter”/“nicolassitter” above, “Hotelrank” here. Copilot tends to know the person; Google AI Mode and Gemini tend to know Hotelrank.
| Date | Type | Note |
|---|---|---|
| 2026-06-11 | dashboard_launch | Niche AI Visibility dashboard goes live (the measurement instrument itself) |
| 2026-06-12 | tool_launch | Schema tool upgraded to audit-first: fetch + score + pre-filled generator (NIC-21) |
| 2026-06-24 | content_refresh | Freshness + retrieval audition chunks shipped on 11 visibility-carrying articles + 3 guides. dateModified/sitemap bumped; FAQPage Quick-answers added per article. First optimization action after baseline closed. |
| 2026-06-24 | guide_launch | New guide: How AI Retrieval Works for Hotels (the mechanism under GEO). |
| 2026-06-30 | content_refresh | Made the hotel schema tool and the AI-visibility guide retrievable for their target prompts: front-loaded Quick answers + FAQPage schema for "free tool to check a hotel's schema markup" and "what is GEO for hotels". |
Reference rate by prompt tier × engine
The 31 prompts split into five intent tiers — each asks a different question about my visibility. Cells show the reference rate and the captures behind it; darker purple means more of that tier's answers list me as a source.
| Tier | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
Category Am I referenced when people learn the topic? | 9.0% 9/100 | 18.8% 12/64 · 13m | 9.4% 6/64 · 6m | 31.7% 26/82 · 28m | 14.6% 12/82 · 8m |
Engine Am I referenced when an AI explains how AIs work? | 11.4% 13/114 | 4.2% 3/72 · 7m | 25.4% 18/71 · 15m | 31.2% 29/93 · 40m | 12.9% 12/93 · 11m |
Resource Am I the canonical source for the data? | 57.9% 66/114 · 15m | 41.7% 30/72 · 30m | 47.2% 34/72 · 39m | 59.1% 55/93 · 57m | 54.8% 51/93 · 45m |
People Am I named as an expert? | 7.3% 6/82 · 29m | 17.3% 9/52 · 14m | 17.3% 9/52 · 15m | 17.9% 12/67 · 18m | 9.0% 6/67 · 6m |
Tool Do my free tools surface for task queries? | 0.0% 0/98 | 19.4% 12/62 · 6m | 0.0% 0/61 | 0.0% 0/80 | 0.0% 0/80 |
x/n = answers referencing me / captures for that tier this week; m = unlinked mentions. The People tier is expected to stay at zero for a long while — being named as an expert is the slowest signal to move, and reporting that honestly is part of the experiment.
Which of my pages get referenced
Every nicolassitter.com URL that appeared in an answer's sources this week. This list decides what I write next: the pages engines already trust are the ones worth deepening.
Engine self-awareness grid
Seven prompts each ask how one specific AI recommends hotels — including Claude and Mistral, which this panel doesn't capture as answerers. Rows are the engine being asked about; columns are the engine doing the answering. Does ChatGPT reference my anatomy-of-ChatGPT study when explaining itself?
| Asked about ↓ | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
ChatGPT “How does ChatGPT recommend hotels?” | — | — | mentioned | ref 3/15 | — |
Perplexity “How does Perplexity choose which hotels to suggest?” | ref 3/16 | — | ref 3/10 | mentioned | ref 3/13 |
Gemini “How does Google Gemini handle hotel recommendations?” | — | — | — | mentioned | — |
Copilot “How does Microsoft Copilot recommend hotels?” | — | — | — | ref 3/13 | — |
Google AI Mode “How does Google AI Mode handle hotel searches?” | — | ref 3/10 | ref 6/10 | ref 13/13 | ref 3/13 |
Claude “How does Claude search for hotels?” | ref 10/16 | — | ref 9/10 | ref 10/13 | ref 6/13 |
Mistral “How does Mistral's Le Chat find hotels?” | — | — | — | mentioned | — |
I have dedicated research for ChatGPT, Google AI Mode, Claude and Mistral, but no article yet on how Perplexity, Gemini or Copilot pick hotels. If those rows fill up with competitor references, that's my editorial calendar talking.
Every prompt, every engine
The full panel, no aggregation: which of the 31 prompts produce a reference, which only a name-mention, and which draw a blank — per engine. This is the raw layer everything above is computed from.
| Prompt | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
| c1How do hotels show up in ChatGPT recommendations? | 3/18 | 9/12 | 3/12 | 3/15 | 3/15 |
| c2What is GEO for hotels? | — | — | — | — | — |
| c3Do AI search engines actually read hotel websites? | 6/16 | — | — | 7/13 | — |
| c4How can a hotel improve its visibility in AI search results? | — | m 2/10 | — | 3/13 | — |
| c5Why does my hotel never appear in ChatGPT answers? | — | — | 3/10 | — | — |
| c6Does schema markup help hotels get recommended by AI? | — | 3/10 | m 1/10 | 13/13 | 9/13 |
| Prompt | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
| e1How does ChatGPT recommend hotels? | — | — | m 1/12 | 3/15 | — |
| e2How does Perplexity choose which hotels to suggest? | 3/16 | — | 3/10 | m 1/13 | 3/13 |
| e3How does Google Gemini handle hotel recommendations? | — | — | — | m 3/13 | — |
| e4How does Microsoft Copilot recommend hotels? | — | — | — | 3/13 | — |
| e5How does Google AI Mode handle hotel searches? | — | 3/10 | 6/10 | 13/13 | 3/13 |
| e6How does Claude search for hotels? | 10/16 | — | 9/10 | 10/13 | 6/13 |
| e7How does Mistral's Le Chat find hotels? | — | — | — | m 4/13 | — |
| Prompt | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
| r1What's the best research on how AI recommends hotels? | m 2/18 | — | — | m 2/15 | 6/15 |
| r2Are there studies on how often ChatGPT links directly to hotel websites? | 12/16 | 3/10 | 3/10 | 6/13 | 6/13 |
| r3Which websites does ChatGPT cite most for hotel recommendations? | 10/16 | 9/10 | 9/10 | 12/13 | 12/13 |
| r4How many hotels block AI crawlers in robots.txt? | 13/16 | m 7/10 | m 3/10 | 9/13 | 3/13 |
| r5What percentage of hotels use schema.org markup? | 12/16 | 9/10 | 9/10 | 9/13 | 9/13 |
| r6How much traffic do hotels actually get from ChatGPT and Perplexity? | 10/16 | 6/10 | 6/10 | 13/13 | 12/13 |
| r7hotel llms.txt adoption statistics | 9/16 | 3/10 | 7/10 | 6/13 | 3/13 |
| Prompt | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
| p1Who are the leading experts on AI search for hotels? | 3/18 | m 8/12 | 3/12 | m 2/15 | — |
| p2Who should I follow to understand how ChatGPT recommends hotels? | 3/16 | 9/10 | 6/10 | 9/13 | 6/13 |
| p3hotel AI visibility consultant | m 8/16 | — | m 5/10 | 3/13 | m 1/13 |
| p4Who does original research on generative engine optimization for travel? | — | — | — | — | — |
| p5Best newsletters or blogs about AI search in the hotel industry | m 4/16 | m 2/10 | — | — | — |
| Prompt | ChatGPT | Perplexity | Gemini | Copilot | Google AI Mode |
|---|---|---|---|---|---|
| t1Free tool to check a hotel's schema markup | — | 3/12 | — | — | — |
| t2How can I check if my hotel website is in Common Crawl? | — | — | — | — | — |
| t3How do I know if AI chatbots can crawl my hotel site? | — | m 2/10 | — | — | — |
| t4Is there an llms.txt generator for hotels? | — | 9/10 | — | — | — |
| t5Free AI visibility audit for hotels | — | — | — | — | — |
| t6Tool to analyze hotel guest reviews with AI | — | — | — | — | — |
n/N= referenced nicolassitter.com in n of N answers · m= name mentioned without a link · — = neither. Counts accumulate across all runs.
Do AI engines actually fetch my site?
The panel above measures what AI answers say across two weekly runs so far. This block is a different dataset: what AI systems do in my raw server logs (Vercel log drain, every week since launch on 2026-03-13 — 20,071 bot hits and 874 AI-referred visits so far). That's why these charts run back to March while the panel charts only have two points: crawlers don't wait for my prompts. Demand on three layers: search indexing, live answer-time fetches, and training crawls — plus the humans who arrive from AI answers.
Bot user-agents classified into search indexing (OAI-SearchBot, PerplexityBot, Claude-SearchBot…), live answer fetches (ChatGPT-User, Claude-User, agent fetchers…) and training crawls (GPTBot, ClaudeBot, Google-Extended, CCBot…). Hover any point for the count. 832 answer/search fetches inside my own run windows are excluded as self-induced (NIC-39); 19,239 external bot hits remain.
The W21 marker is the hotelrank.ai → Lighthouse announcement (/research/* redirects + AI-assistant buzz) — that spike is the event, not organic growth. The final point is dashed because the current week is still in progress (a partial count, not a drop).
The companion to the chart above, so the self-induced spikes are visible, not just netted out. These are answer- and search-time fetches that landed within 6h of one of my scheduled runs (niche Tue 03:00, landscape Mon 02:00 UTC): when a prompt names this site, the engine fetches it to answer — so this line is me, not the world. The niche panel only started in June (week of 2026-07-06 is its second run), so the earlier weeks here are my longer-running landscape panel, which has pinged weekly since March. 832 hits total, all subtracted from the external view above. Hover any point for the count.
The same thing as a ratio: what fraction of each week’s bot traffic was me, not the world. High weeks are mostly my own scheduled runs; the rest is genuine external demand.
Visits whose referrer is an AI product (zero-click answers don't show up here — this is a floor, the same caveat as the measurement guide).
The W21 marker is the hotelrank.ai → Lighthouse announcement — referrals jumped as the move was discussed inside AI assistants, not from organic discovery.
Bot hits on the files that govern AI access: /robots.txt (the crawl rules) vs /llms.txt + /llms-full.txt (the AI-reading guide). The gap is the point — crawlers fetch robots.txt constantly (1,739 hits) but almost never pull the llms files (35): the llms.txt convention isn’t adopted by the engines yet. Both llms lines sit so close to zero against robots.txt’s scale that they overlap on the axis — hover any point to read the real (tiny) counts.
Of the bots that read my crawl-rules file each week, what fraction also pull the AI-guide. A low line is the story: llms.txt isn’t part of the crawl yet.
| Bot | Purpose | Hits | Last seen |
|---|---|---|---|
ClaudeBot Claude | training | 5,777 | 2026-07-06 |
ChatGPT-User OpenAI | user | 4,479 | 2026-07-07 |
Bytespider ByteDance | training | 4,030 | 2026-07-07 |
GPTBot OpenAI | training | 2,735 | 2026-07-07 |
OAI-SearchBot OpenAI | search | 1,999 | 2026-07-07 |
PerplexityBot Perplexity | search | 618 | 2026-07-06 |
Claude-User Claude | user | 258 | 2026-07-05 |
CCBot CommonCrawl | training | 100 | 2026-06-28 |
CloudVertexBot Google | user | 30 | 2026-06-29 |
Gemini Deep Research Google | user | 24 | 2026-07-05 |
| Path | Hits | Bots |
|---|---|---|
| /projects/ai-hotel-landscape/[platform]/[[...path]] | 1,568 | 8 |
| / | 1,132 | 9 |
| /research | 1,103 | 10 |
| research/hotel-schema-adoption-study-2026 | 1,046 | 11 |
| research/ai-hotel-landscape-2026 | 849 | 11 |
| research/google-ai-mode-hotel-study-2026 | 689 | 12 |
| /projects/ai-hotel-landscape/hotel/[slug] | 312 | 3 |
| research/chatgpt-hotel-ads-live-2026 | 305 | 8 |
| index.segments/_tree.segment | 283 | 1 |
| index.segments/_head.segment | 273 | 1 |
Paths are Vercel route names, so dynamic pages appear as their template. Asset and feed requests are filtered out.
The frozen panel, the cadence, the caveats
Same standard as the research articles: the full prompt panel is published, the limitations are listed, and nothing about the measurement is hidden.
Panel. 31 prompts across five intent tiers, frozen on June 10, 2026. Prompts are never edited — additions only arrive as a versioned batch (panel v2) — so week-over-week movement stays comparable. No prompt contains my name or domain; name mentions are detected passively by an NER pass over every answer. Download the panel: prompts.csv.
Volatility control. Every prompt runs 3× per run, because single-shot AI answers flap — all rates are computed over the repeats. (Run 1, June 10, fired only six headline prompts in triplicate; full-panel triplicates start with run 2.)
Cadence and setup. Weekly, Tuesdays 03:00 UTC, via Bright Data. Logged-out sessions on a US proxy, English prompts only in v1 (French variants for 8 prompts are planned as v1.1). Real users with history and location will see different answers — the same caveat as every study on this site.
When I ping the engines (and why it matters for the server-side block). My own automated runs hit the AI engines on a fixed schedule: the niche panel fires Tuesdays 03:00 UTC and the hotel-landscape panel fires Mondays 02:00 UTC, plus the occasional ad-hoc research scrape. When a prompt names this site, an engine sometimes fetches it live to answer (ChatGPT-User, PerplexityBot…), so those hits land in my own logs and are self-induced, not organic demand. Read the crawler block with that in mind: a live-fetch bump in the hours after a scheduled run is probably me. Once the log drain is fully wired, the rule is to subtract hits inside those known run windows before reporting external AI-crawler traffic.
- n is small. Even at three shots per prompt, one week is a snapshot. Treat any individual week as noise; only the multi-week rates carry information.
- Reference positions aren't comparable across engines. Copilot and Google AI Mode expose sources as footnote pools rather than ranked lists, so this page never ranks by position.
- I'm measuring myself and optimizing for the metric. The mitigation is publishing everything: the panel, the action log, and the misses.
The full 31-prompt panel (v1, frozen 2026-06-10)
| c1 | How do hotels show up in ChatGPT recommendations? | headline |
| c2 | What is GEO for hotels? | headline |
| c3 | Do AI search engines actually read hotel websites? | |
| c4 | How can a hotel improve its visibility in AI search results? | |
| c5 | Why does my hotel never appear in ChatGPT answers? | |
| c6 | Does schema markup help hotels get recommended by AI? |
| e1 | How does ChatGPT recommend hotels? | headline |
| e2 | How does Perplexity choose which hotels to suggest? | |
| e3 | How does Google Gemini handle hotel recommendations? | |
| e4 | How does Microsoft Copilot recommend hotels? | |
| e5 | How does Google AI Mode handle hotel searches? | |
| e6 | How does Claude search for hotels? | |
| e7 | How does Mistral's Le Chat find hotels? |
| r1 | What's the best research on how AI recommends hotels? | headline |
| r2 | Are there studies on how often ChatGPT links directly to hotel websites? | |
| r3 | Which websites does ChatGPT cite most for hotel recommendations? | |
| r4 | How many hotels block AI crawlers in robots.txt? | |
| r5 | What percentage of hotels use schema.org markup? | |
| r6 | How much traffic do hotels actually get from ChatGPT and Perplexity? | |
| r7 | hotel llms.txt adoption statistics |
| p1 | Who are the leading experts on AI search for hotels? | headline |
| p2 | Who should I follow to understand how ChatGPT recommends hotels? | |
| p3 | hotel AI visibility consultant | |
| p4 | Who does original research on generative engine optimization for travel? | |
| p5 | Best newsletters or blogs about AI search in the hotel industry |
| t1 | Free tool to check a hotel's schema markup | headline |
| t2 | How can I check if my hotel website is in Common Crawl? | |
| t3 | How do I know if AI chatbots can crawl my hotel site? | |
| t4 | Is there an llms.txt generator for hotels? | |
| t5 | Free AI visibility audit for hotels | |
| t6 | Tool to analyze hotel guest reviews with AI |