Seba BlancoI help independent hotels sell more effectively and operate smarter by combining hotel technology with sales and marketing.
Do not index
Do not index
Last week I argued that AI search doesn't read your website — it reads your guests. The practical consequence is uncomfortable: your review history is the largest qualitative study your hotel will ever run, and almost nobody actually analyzes it.
This post is the toolkit. Eight prompts, in the order you should run them — plus the part nobody explains, which is how to get your reviews out of the platforms in the first place.
Step Zero: Getting Your Reviews Out of the Platforms
Don't paste reviews into a chat box. Every AI tool worth using accepts file uploads, so point the prompt at an attached CSV instead. That also sidesteps the context limit you'll hit around month four of pasting.
The real work is extraction, and each platform is a different world.
Google is the only genuinely open one. The Google Business Profile API returns your full review history legitimately. One trap if you plan to automate it: Make's native Google Business Profile integration has no module to list reviews — only "Create/Update a Review Reply." To read them you need the generic "Make an API Call" module pointed at the reviews endpoint. Perfectly doable, just not sitting in the module list.
Booking.com won't let you bulk-export review text from the Extranet. A Guest Reviews API exists inside the Connectivity APIs, but that access runs through your connectivity partner — your channel manager or PMS, not you directly. Worth one email to them.
Tripadvisor is the most closed of the three for this purpose.
Don't just hand the AI your review page URL. Browsing models read the first page or two — your most recent reviews — and hand back exactly the recency bias this whole exercise exists to defeat. You get fresh anecdotes, not patterns.
If you already pay for a reputation tool (TrustYou, Revinate, MARA and similar), export from it and run these prompts on the export anyway. Their dashboards answer the questions their product team chose. Prompt 2 below is not one of them.
Chrome extensions that scrape Booking reviews exist and work. They also violate platform terms of service. Your call, eyes open.
For a small independent, here's what actually pays: build a Google Sheet once with five columns — date, platform, score, review text, guest type. Twelve months takes about two hours. Then ten minutes a month keeps it current, and every prompt below runs on that one file forever.
Prompts 1–3: Find the Patterns
Run these three in the same conversation — each builds on the last.
1. Pattern extraction. The base exercise.
You are a hospitality operations analyst. Attached are guest reviews for my hotel from the last 12 months, across multiple platforms.
Your task: identify REPEATED patterns, not individual opinions.
Return:
1. The 5 most frequently repeated complaints. For each: how many reviews mention it, verbatim quotes (max 3 per theme), which platforms it appears on, and which months.
2. The 5 most frequently repeated compliments, same format.
3. Any theme appearing in fewer than 3 reviews — list these separately under "Anecdotes, not signals."
Rules:
- Quote verbatim. Never paraphrase a guest.
- Do not invent, infer, or soften anything.
- If a theme is ambiguous, say so instead of forcing it into a category.
- State the total number of reviews analyzed at the top.2. Reputational pain. This is the one that separates a bad score from a structural problem.
Using the same review set, rank my reputational issues by PAIN, not by star rating.
Weight each recurring issue by:
- Frequency: how many reviews mention it
- Persistence: spread across all 12 months, or clustered in one period?
- Trajectory: getting better, worse, or flat?
- Emotional intensity: mild inconvenience or genuine anger?
- Deal-breaker status: does the guest say it would stop them returning or recommending?
Output a ranked table, most painful first, one line of justification per row.
Then answer explicitly: which issues would a reader notice from only the 10 most recent reviews, and which are visible only when reading all of them?3. Journey mapping. Patterns are useless until you know where they happen.
Map each recurring complaint to the exact stage of the guest journey where it occurs:
pre-booking research / booking process / pre-arrival communication / arrival and check-in / the room / breakfast and F&B / staff interaction during stay / problem resolution / check-out / post-stay follow-up
For each stage give: the issues located there, mention count, and whether the fix is a policy change, a training issue, a physical/capex issue, or a communication issue.
Flag any stage with zero complaints — that may be a real strength or a blind spot where guests simply don't comment.Prompts 4–6: From Patterns to Positioning and Operations
4. Hidden attributes. What you sell versus what they value.
From the positive reviews only, extract what guests actually value about my hotel — in their own words and their own priority order.
Compare that against what I currently claim to sell: [PASTE YOUR WEBSITE COPY].
Tell me:
1. Attributes guests praise that my marketing barely mentions.
2. Attributes I lead with that guests almost never mention.
3. The specific words and phrases guests repeat when describing what made the stay good — I want their vocabulary, not mine.
Be blunt about the mismatch. Do not reassure me.5. The promise-delivery gap.
Here is the promise my marketing makes: [PASTE WEBSITE COPY, OTA DESCRIPTION, TAGLINE].
Here is what guests report experiencing: [PASTE THE OUTPUT FROM PROMPT 1].
Identify every point where promise and delivered experience diverge. For each gap, tell me whether the honest fix is (a) change the operation to meet the promise, or (b) change the promise to match the operation.
Give a recommendation per gap with reasoning. Assume limited budget and a small team, and tell me which single gap, if closed, would remove the most guest disappointment.6. The operations brief. This is how the analysis leaves the marketing folder.
Turn the top 3 reputational issues into an operations brief for a small hotel team.
For each issue:
- The problem in one sentence, in operational terms, not marketing terms
- Two verbatim guest quotes as evidence
- The department or role that owns it
- One specific change implementable within 30 days on minimal budget
- One measurable signal that would show in 90 days whether it worked
Under one page. Written for a head of housekeeping and a front desk supervisor, not a marketing team. No jargon.Prompts 7–8: Test How AI Recommends You
Run these in a fresh session, on ChatGPT, Gemini and Perplexity separately. Fresh matters — if you've already mentioned your hotel in that conversation, you've contaminated the test.
7. The decision-intent test.
I'm traveling to [CITY] for [3 nights, business]. I want somewhere [quiet / well located / good breakfast — use your real guest profile] with no recurring service complaints.
Which properties would you recommend, and why? For each one, tell me what evidence you're basing it on and where that evidence comes from.Then, in the same conversation:
Now tell me honestly what you know about [YOUR HOTEL] in [CITY]. What are its recurring strengths and weaknesses according to available guest feedback? What would make you hesitate to recommend it?8. Competitive evidence.
Compare the guest-review evidence for my hotel against these 3 competitors: [NAMES].
For each property summarize: what guests consistently praise, what they consistently criticize, and the single attribute that most distinguishes it.
Then answer: for a traveler who wants [YOUR TARGET GUEST PROFILE], which property does the available evidence most support recommending, and why? Answer as an outside analyst with no loyalty to any of them — if my hotel isn't the answer, say so, and explain what evidence would have to change.How to Actually Run This
Prompts 1 through 6 belong in one conversation, because each one eats the previous output. Prompts 7 and 8 stand alone and are worth repeating quarterly — model answers shift as new evidence accumulates around you.
One deliberate detail: the "quote verbatim, don't soften" instruction appears in several of these. Without it, models paraphrase complaints into something gentler. That's precisely where the signal dies.
Block ninety minutes. Build the sheet, run prompts 1 and 2, and see whether your worst recurring complaint is one you already knew about. Most hoteliers find at least one they didn't.
AI Doesn't Read Your Website. It Reads Your Guests.
AI Hotel Reputation: Your Guests Write Your Brand's Evidence
AI Hotel Reputation: Your Guests Write Your Brand's Evidence
AI search doesn't read your website — it reads your reviews. What one experiment with Santiago hotels reveals about reputation, patterns, and being recommendable.
Subscribe to Hotel Marketing Point for practical AI workflows built for independent hotels — no tech team required.
Written by
Seba Blanco
I help independent hotels sell more effectively and operate smarter by combining hotel technology with sales and marketing.
