It’s 11:43pm. You’ve got 15 tabs open, the price on the hotel room you liked just jumped 12% since this afternoon, and the dynamic pricing keeps moving the goalpost every time you refresh.
After spinning your wheels for hours, you finally pick a room.
You get to the hotel, and the “5-minute walk to the beach” is a 15-minute hike down a dusty side path in August heat.
What if there was an easier way to find the best hotel just for you?
There are prompts that swear they put travel agents out of a job floating around Twitter. Find you the perfect place, save you so much money it feels like you’re getting paid to take the trip.
What you actually get is a fairy tale.
The issue is that picking where to stay is 3 decisions, and a single prompt smashes them into one.
The fix is a prompt chaining technique. You run a series of prompts in order, where each one does a single job and feeds the next. When done right, “find me a hotel” turns into a shortlist of the best hotels tailored just for you.
Works for any trip, any destination, any group. The example below is a beach town, but the steps don’t care where you’re going.
Why one prompt can’t find you the best hotel
Ask a model to “find the best hotel” and you’ve handed it 3 jobs at once: figure out what you care about, which part of town to stay in, and which specific places fit.
It does all 3 terribly because it doesn’t have context. We all know AI won’t say “I don’t know” or, god forbid, “I need more information.”
Then there’s the data. A bare model will do a quick web search and gloss the top few hotels with the best SEO, but it can’t pull a real bookable price for your dates, and it won’t rank anything against what you actually care about. You get a shallow, generic answer in a confident voice.
A chain of prompts fixes these problems. It splits the decision into steps and pulls real data at each one.
The first step is to give the model something real to work with.
What to do before you run the hotel prompts
Most AI apps have connectors. Think of a connector as a bridge that helps the model grab real-time data instead of a guess. For hotels, 3 good ones are built into Claude for free: TripAdvisor, Booking.com, and Trivago.
Step 1. Connect your hotel tools. Open your AI’s connector directory and turn on at least one. They’re free and one-click install. Turn on 2 or 3 if you want the model to cross-check prices and reviews against each other, which is where this method really shines.
Step 2. US trips, add Uber. With the Uber connector on, the model can pull live ride-cost estimates between areas. This one only works in the US. On any trip outside the US, the model skips fares and reasons by which area things are in. Bonus: the model will understand distances a bit more accurately.
Step 3. Ask your model what it has. On ChatGPT or Gemini, ask which hotel connectors it’s got. This one line is what makes the method work on any AI. You don’t need to know what’s available, the model tells you.
Connect at least one tool, then run the 3 prompts below in order. The order matters. Each one builds on the last, and if you don’t connect the AI to anything, it’ll hallucinate.
Prompt 1. Figure out what matters most on your trip
Before any recommendations, the model needs to know which priorities matter most, so it understands what “best hotel” means for you.
When you rank those priorities, you’re handing it a rubric to judge every option against. This is the step nobody does, and it’s the one that changes the output of everything after it.
Run this prompt as is. The fields are there to show you what to cover, not to fill in. The model reads them, then asks. You answer in plain language, the way you’d tell a friend about the trip.
I'm planning a trip and want help deciding where to stay. We'll do this in 3 steps. Don't recommend places yet. Work only from what I give you.
Read the checklist below, then ask me about my trip. I'll answer in my own words, you don't need me to fill in blanks.
- Destination, and the areas I'm torn between (name the city if you don't have specific neighborhoods)
- Who's going (solo / couple / family with kids / friends)
- Nights and dates
- Nightly budget
- What I want to do there (be honest. beach, a boat day, food, pool time, kid stuff, exploring)
- How I'll get around (taxi or rideshare / transit / driving)
Ask me for anything I leave out, all of it at once, in one list. Don't ask in dribs and drabs.
Once you have what you need, list the factors that should drive this decision, in priority order, most important first. Show me the order and ask me to adjust it. These are my priorities, so I decide.
Once I confirm, hold the final ordered list. We'll use it for the rest of this chat.Don’t tell the model that having a big pool matters more than kid activities, and it’ll invent a ranking you never agreed to. It does this quietly. You’ll never see it happen, and you’ll never know the shortlist came out skewed.
The output is a ranked list of what matters, confirmed by you. Beach access first, then budget, then kid activities, or whatever your real order is.
This list significantly changes the trajectory of the next two prompts.
Prompt 2. Choose the area before you pick the hotel
Here’s a mistake that makes people regret a booking. They pick a specific hotel before they pick the part of town. We’re all guilty of this. We fall for a gorgeous room at a bargain price, and only later figure out why it was discounted.
Area first. Hotel second. Always.
Paste this straight after you’ve agreed on the priority list. The model already has your list from the same chat, so there’s nothing to re-paste.
Good, that order works. Now help me choose an area to base in. No specific places yet.
1. Map my activities to the areas. Where's most of what I want to do?
2. Check recent traveler sentiment from reviews and Reddit on each area, from roughly the last 3 years. Honest vibe of each, and which fits my group.
3. For each area, tell me which activities are right there versus in a different area. Same-area stuff is a non-issue. Different-area stuff is what costs me in taxis and time, so flag those as the trips that add up, roughly how far and how often. (US trips: if Uber's connected, pull a real fare for the longest regular trip.)
4. Budget check. Tell me what my nightly budget realistically buys in each area. Stay within 20% of my number. If I say 10,000, don't show me 6,000 rooms. I picked that number for a reason. If price is my top priority, find a discount on a hotel at my level, don't drop me to a cheaper class of place.
5. Recommend the 1 or 2 best areas, ranked by my priority order, with the tradeoff of each in a line.
Don't push me somewhere isolated to satisfy a comfort factor like quiet. Being close to what I came to do still counts. Weigh it.
Lead with a one-line answer, then short bullets. No long paragraphs.The activity-mapping in points 1 and 3 is what makes the trip yours. Instead of “this neighborhood is nice,” you get a real picture of what’s a short walk versus what’s a daily taxi. That’s money and time you’d otherwise find out about standing on a curb trying to get a ride.
Point 4 is the guardrail against the race to the bottom. Tell a model your budget is $200 and some of them go hunting for the cheapest thing that technically fits, which is how you end up looking at a roach motel because it was a good deal. The line keeps it shopping at your level, and if cheap is your actual priority, it finds a discount on a good place instead of a worse one.
The last guardrail matters more than it looks. Tell a model you want quiet and it overcorrects, steering you somewhere isolated and calling “far from everything” a feature. Quiet and a 30-minute ride to dinner every night isn’t worth the trade.
Pick one area. Then you shortlist the hotels in it.
Prompt 3. Shortlist real hotels with the tradeoffs showing
Now the model picks specific hotels from live data, ranked by your priority list. Not one winner. A shortlist, because the best hotel depends on tradeoffs only you can weigh.
Paste this once you’ve locked the area. Same chat, so it still has everything from the first two prompts.
Area's locked. Give me a shortlist for it, not a single winner.
What I want: [hotel or resort], [nights and dates], [guests and rooms].
Use the connected hotel tools for real prices, ratings, and reviews. If something isn't from a tool, say "estimate" or "reviews don't mention this." Never present a guess as real. If tools disagree on price by more than 10%, say so and resolve it before ranking. Don't rank on an unverified number.
1. Pull the real options that fit my area and budget. Prices and ratings straight from the tools. Stay within 20% of my budget, same rule as before.
2. Check reviews for how close each place really is to what I want to do. Cross-check across tools. Where they agree, that's a real signal. Where they're silent, say so.
3. Red-flag scan. Only flag a problem when a real cluster of guests (roughly 5 or more) report the same concrete thing, not 1-off gripes. Things like a noise source next door, a loud crowd, hidden fees, a recurring maintenance issue, anything that keeps coming up. Skip complaints that show up once or twice.
4. Give me the top 3, ranked by my priority order. For each one, in a few bullets: the real price, why it could be the pick, what I'd give up to choose it, and any red flag from the cluster scan. Show the tradeoffs, don't hide them behind one winner. Don't let one comfort factor outweigh being near what I came to do. Keep it succinct and lead with the main point.
5. Then tell me what to check myself before booking: the exact spot on a map against my activities, and current prices or discounts on the hotel's own site and the booking apps, since those move. Stop at the shortlist. Don't book anything.
Lead with a one-line read on the shortlist, then a short block per place. No long paragraphs.The data rules up top are the anti-hallucination troops. “Say estimate” forces the model to label its guesses instead of selling you a dream that turns into sticker shock at checkout. The price-disagreement rule catches the case where one tool says $140 and another says $185, which usually means a resort fee or a date mismatch hiding in the gap.
The red-flag scan in point 3 is the part most people get wrong.
A single bad review is noise. Somebody argued with the front desk because they thought they could get the Ritz for $60.
What tells you something is the same complaint, from different people. 5 guests all mentioning the call to prayer at 5am or a pattern of “the AC was broken.” Set the threshold at 5-plus on the same thing, ignore the one-offs, and you get warned about what’ll actually affect your stay.
Point 4 refuses to hide the tradeoffs. The cheaper place is louder. The one with the great pool has a hidden resort fee. Ranked options with the tradeoff spelled out beat a confident winner, because you’re the one who knows whether you’d trade quiet for location.
Point 5 is in your hands. The model gives you the hotels, then tells you to check its homework. Verify the exact map spot against your plans, and the live price on the hotel’s own site, since that’s where the discounts hide and that’s the number that moves before checkout.
Why a chain of prompts beats one
The whole method rests on one idea. A good hotel decision can’t be one-shotted.
Each prompt does one job.
Prompt 1 makes you say what matters before the model can bias the answer.
Prompt 2 settles the area before a pretty photo hijacks the choice.
Prompt 3 surfaces real options with the tradeoffs showing, then hands the final check back to you.
You could cram all of that into one giant prompt. I tried. But what you miss out on is the best hotel, because the model ignores your priorities and answers its own question instead.
Run it on your next trip. Connect a tool, work the 3 prompts in order, and see what the shortlist looks like when the model reasons from real data against your priorities instead of guessing in a friendly voice.