A Locked Museum, an Unreadable Menu, a Red Departures Board
WEEK 98 :: POST 3 :: CLAUDE
Directions Given To The A.I. This Week+
Instructions Given to each A.I. — Please provide 3 prompt variations that share this objective:
Each A.I. also received two static attachments: the blog post template (structure) and the authoring instructions (voice and standards). The text below is the week-specific assignment as sent — reflowed for the web; wording unchanged.
I'd like you to write this week's Ketelsen.ai post. Two files are attached: the blog post template (the structure to follow) and the authoring instructions (context, voice, and standards). Please read both before you begin, then produce the complete post in a single response.
This week's theme: "Your AI Travel Companion" — In-Trip Prompts and Real-Time Problem Solving.
This is Week 7 of an eight-week series on planning a vacation with AI. Every week until now has been preparation. This week the reader is on the trip, and the prompts change character completely: they are run in the moment, mostly on a phone, often when something has just gone wrong. The plan met reality, and this is the week about handling the gap.
The situations are the ones that actually happen on the road: weather or a sudden closure kills the day's plan and it has to be rebuilt on the spot; a menu or a sign needs translating and decoding, not just word-for-word but for what is actually worth ordering; an unexpected free afternoon opens up near a particular spot and the reader wants to use it well; a flight is delayed or cancelled and the reader needs to know what to say at the counter; and the running total needs checking against the budget ceiling from Week 1 before the last two days blow it.
The deliverable this week is different from the earlier weeks' plans and matrices. It is a small kit of fast, reusable prompts the reader can fire from their phone when they need them — each one short, each one solving a specific in-trip problem, each one written so it works with whatever the reader can see and type in the moment.
The three prompts should help a reader:
- Replan on the spot. Rebuild a half-broken day from what is still open and reachable, given the reader's current location, the time left, and what they had planned — the AI reorganising around the constraint rather than starting from scratch.
- Decode and communicate. Translate and interpret menus, signs, and short exchanges in a way that helps the reader act — what a dish actually is, what a notice is telling them, how to ask a specific question — rather than a bare literal translation.
- Know the script when travel breaks. Turn a delay, cancellation, or overbooking into a short, calm, word-for-word set of lines to use with the airline or hotel — what to ask for, in what order, and what the reader is generally entitled to raise, while pointing them to confirm the specifics with the carrier.
At the advanced tier, the strongest version of this week is a compact, labelled prompt kit — a handful of ready-to-paste prompts, each named for the moment it solves, plus a daily budget-check prompt the reader runs against the Week 1 ceiling. Brevity is the craft here; a bloated prompt the reader cannot use one-handed on a phone has missed the point.
A hard constraint, and it takes a particular shape this week. AI models cannot see live flight status, current weather, real-time transit, today's prices, or the reader's actual entitlements under a specific carrier's policy or a specific jurisdiction's rules. No prompt may ask the AI to state a flight's current status, quote a live price, promise a specific compensation amount, or assert the reader's legal rights as settled fact. The rebooking and delay-rights prompts should give the reader strong, calm scripts and the right questions to ask — and explicitly tell them to confirm entitlements with the airline or a posted policy — not a guarantee of what they will get. A wrong promise about compensation made at a counter is worse than useless.
Design the prompts so the AI does what it is genuinely good at on the move: reorganising a plan under a live constraint, interpreting language in context, and giving the reader composed, well-ordered words to use under stress. The reader supplies what is happening right now; the AI supplies structure, script, and clarity. Posts whose prompts depend on the AI knowing live facts should expect to be marked down on Practical Utility.
Series dependency chain, for the Metadata block: Week 7 consumes the budget ceiling from Week 1 (for daily tracking) and the full trip plan built across Weeks 2 to 6 — the itinerary it repairs, the bookings it works around, the protection protocol it falls back on. Week 7 produces the in-trip record — what actually happened, what was spent, what went wrong — which Week 8's post-trip reconciliation and reusable-template build both draw on directly.
Because readers may arrive at this post mid-trip without having read the earlier weeks, the prompts should work for someone who simply has a problem in front of them right now, while making clear they connect back to the budget and plan the series built.
Three difficulty tiers as always — Beginner, Intermediate, Advanced — each a genuinely different approach to the same problem, not the same prompt at three lengths.
On examples: this is a consumer travel topic. The template lists tech startup / retail / freelance as suggested industry examples — those are marked MAY, and this week you should almost certainly adapt them. A parent salvaging a rained-out day with restless kids, a couple stranded by a cancelled connection, a solo traveller decoding a menu in an unfamiliar script, and a group splitting a running tally against a shared budget are the right contexts here. Choosing them over the suggested business examples is correct behaviour and will not be scored against you.
A note on supplied figures. Anything marked `[SUPPLIED — use as given]` above came from Ketelsen.ai's own research brief. Use it freely — you are not fabricating by repeating it, and you will not be marked down for leaving it uncited. Do not attach an invented source to it. (No supplied figures this week. Given the live-facts constraint above, this is a bad week to invent any — if you find yourself reaching for a compensation amount or a live price, that is the signal to restructure the prompt so the reader gets a script and confirms the number at the source.)
## BEFORE YOU SUBMIT — STRUCTURAL CHECK
(This block is identical every week. It exists because these specific items are the ones posts drop, and a dropped structural item costs compliance points for something that takes one minute to add.)
Your post is parsed by a script before any human reads it. Confirm all seven:
1. ☐ Response begins with `PLATFORM: <your name>` and `WEEK: 7` 2. ☐ `## Lead` present once, at the very top, before Variation 1 3. ☐ `## In one line` present in all three variations 4. ☐ `## What this prompt gives you` present in all three variations 5. ☐ `## The Prompt` present in all three variations, with the prompt in double quotes beneath it 6. ☐ `## Introductory Hook` and `## Current Use` present in all three variations (three of each — not one) 7. ☐ Every template heading written as `##`, none bolded instead; prompt breakdown is running text split on ` : `, with no `###` headings inside it
A complete post has 57 `##` headings. If your count is well short, a section is missing or was bolded instead of hashed.
One extra check this week: confirm no prompt asks the AI to state a live flight status, quote a current price, promise a specific compensation amount, or assert the reader's legal rights as settled fact. Those must be scripts and questions the reader confirms with the carrier or a posted policy.
Six weeks of planning meet a locked museum door, a menu in an alphabet you can't read, and a departures board that just turned red. This week's three prompts are built for that moment: a beginner day-rescue prompt that rebuilds a broken afternoon in about a minute, an intermediate decoder that turns any menu, sign, or notice into something you can act on, and an advanced kit-builder that generates a whole set of labelled, one-handed prompts — including a counter script and a daily budget check against your Week 1 ceiling. Every one of them is short enough to use standing in the rain.
The Day Rescue
Rebuild a broken day from where you're standing, in one message.
The plan was good. The plan is now irrelevant. It's 11:40, the site you built the day around is closed for a private event, and you're standing outside it with a phone at 34% and two people looking at you expectantly. This is the exact moment most travellers do the worst possible thing: they open a map app, scroll aimlessly for fifteen minutes, and end up at the nearest chain café feeling defeated. The problem isn't a shortage of options. It's that rebuilding a day requires holding six variables in your head — time left, distance, energy, weather, what's already booked, what everyone can tolerate — while stressed and standing up. That's a structuring job, and structuring under pressure is exactly what an AI is good at.
Why this matters now
Weeks 2 through 6 of this series built you an itinerary. An itinerary is a prediction, and predictions break — a closure, a strike, a downpour, a kid who is done. What you need mid-trip isn't a new plan from scratch; it's a repair to the one you already have, made in under two minutes on a phone. This prompt exists because the alternative is losing an afternoon to indecision, and afternoons on a trip are expensive. It also feeds Week 8: the days you salvaged are part of the record you'll reconcile at the end.
I'm travelling in [city or area] and today's plan just broke: [what happened — rain, closure, missed train, someone's not feeling well]. It's now [time] and I need to be [where you need to be] by [time]. I'm at or near [where you are right now], getting around by [walking / metro / bus / car], and I'm with [who you're with — solo, partner, kids aged X, a group of Y]. Here's what I had planned for the rest of today: [list the two or three things]. Rebuild the rest of my day. Keep whatever still works, and replace what doesn't with kinds of places that are usually open, nearby, and a good fit for this constraint. Give me a simple hour-by-hour list with one line on why each stop is there, plus one backup version if the weather gets worse. Don't tell me whether any specific place is open right now — you can't know that. End with anything I should call or check before I set out.
How the AI reads this prompt
Practical examples from different industries
A parent with a rained-out day and two restless kids.
The family had budgeted the whole Wednesday for a hilltop park and a walk down through the old town; it has been raining sideways since eight. The parent types the prompt from a doorway: rain, 10:15 now, dinner reservation at 19:00, near the cathedral, walking, kids aged 5 and 8. The output keeps the dinner, drops the hill, and builds a covered morning — a market hall, a small transport museum, a long lunch — with a note that the 5-year-old will need a break by 13:00 and a backup that collapses the afternoon into one indoor stop if the rain doesn't ease. The value isn't the ideas; it's that the decision took ninety seconds instead of an argument.
A couple whose morning site is unexpectedly closed.
They arrive at a monastery to find it shut for a ceremony, with four hours before a train they cannot miss. Feeding the prompt their location, the 15:40 departure, and the fact that they're on foot with luggage in a locker at the station, they get a plan that stays inside the walkable ring around the station, front-loads the one thing they actually cared about, and builds in a 40-minute buffer before boarding. The buffer is the part they wouldn't have added themselves, and it's the part that saves the train.
A solo traveller who lost an afternoon to a cancelled tour.
The booked walking tour didn't run and the refund conversation ate an hour. The prompt is fed the remaining three hours, a stated preference for "somewhere I can sit and not talk to anyone," and a walking radius. What comes back is a low-social-energy afternoon — a park, a quiet café district, one small collection — plus a line noting that late afternoon is when many smaller sites stop admitting visitors, so go to the ticketed one first. That sequencing note is the difference between three good hours and two.
Creative use case ideas
- A rained-out wedding weekend. Not your trip at all: you're a guest, the outdoor drinks reception is drowned, and forty people need somewhere to be for two hours. Same prompt, different constraint set.
- The airport layover you didn't plan for. Feed it a seven-hour connection, the terminal you're stuck in, whether you can clear immigration, and how much you're willing to risk. It will separate "leave the airport" from "don't" and build accordingly.
- A hospital or care-home visiting day. A relative's appointment ran long and you have three unstructured hours in a town you don't know. The prompt works identically and asks nothing about tourism.
- Salvaging a work trip evening. Client dinner cancelled at 17:00 in a city you land in twice a year. Instead of hotel room service by default, you get a plan that fits the ninety minutes before you're too tired to enjoy it.
- Testing a plan before you leave. Run it at home with an invented failure — "the main museum is closed on Mondays" — and you'll find the weak joints in your itinerary while you can still fix them cheaply.
Adaptability tips
Swap the constraint and the prompt covers an entirely different failure: replace "it's raining" with "everything costs more than I expected" and you get a low-spend day; replace it with "I'm exhausted" and you get a day with one stop and a long lunch. Tighten the radius by adding "nothing more than fifteen minutes from where I am" when you're low on energy or battery. If you're travelling with someone whose preferences differ sharply from yours, add a line naming both — "one of us wants quiet, one wants to see things" — and ask for a plan that alternates. For longer horizons, change "the rest of my day" to "the rest of my trip" and hand it your remaining itinerary; the same repair logic scales, though the output gets long enough that you'll want to be sitting down. And if you're in a place where you genuinely don't know what's normal, add "tell me what a local would do with this weather" before the output shape line.
Pro tips
- Screenshot the reply immediately. Signal dies, apps crash, batteries drain. A screenshot survives all three and doesn't need a connection to read.
- Add "assume I'm on foot in bad shoes." Absurd-sounding constraints work well because they compress several real limits — pace, distance, tolerance — into one instruction the model handles sensibly.
- Ask for the plan in reverse. "Start from the thing I can't miss and work backwards" produces different, often better sequencing than forward planning, because it protects the fixed point first.
- Keep the filled-in prompt, not just the answer. Once you've typed your details in once, the next break in the plan costs you one edit instead of a paragraph.
Prerequisites
You need to know roughly where you are — a landmark, a neighbourhood, or a metro stop is plenty — and what time you have to be somewhere else. Have your existing plan for the day to hand, even if it's just in your head; the prompt is much stronger when you can paste in what was supposed to happen. Know your mode of transport and whether you're carrying luggage. If you're travelling with others, have a rough sense of what they can tolerate before you type. No account tier, plugin, or setup is required beyond a phone with an AI app or browser tab open.
Required tools
Any general-purpose AI assistant on a phone — a browser tab is enough. No paid tier required; free tiers handle this prompt comfortably because the output is short. A maps app is useful alongside it for confirming that the categories the AI suggests actually exist nearby, but it isn't required to run the prompt. Offline note-taking of any kind helps, since the most common real-world failure here is losing signal after you've read the answer once.
Frequently asked questions
What if the AI suggests a place that's closed or doesn't exist?
Expect this occasionally, which is exactly why the prompt tells the model not to claim live opening hours and asks for kinds of places rather than a list of names. Treat any specific venue it does name as a lead, not a fact — a thirty-second check in a maps app confirms or kills it. The plan's structure is the durable part; the individual stops are swappable. If more than one suggestion turns out to be wrong, add "only suggest categories, no specific business names" and rerun.
I don't want to type all that on a phone. Can I shorten it?
Yes, and the shortest version that still works is: where you are, what broke, how long you've got, and who's with you. Those four facts carry most of the weight. What you lose by cutting the rest is the contingency branch and the verification list, so if you shorten it, at least keep the final line asking what to check. Better still, save the full version in your notes app before you travel and fill in the blanks — that takes about twenty seconds.
The answer came back too long to read while walking. What now?
Reply with "cut it to five lines, no explanations." Models honour compression instructions well, and asking for the short version after seeing the long one is often better than asking for short up front, because the model has already done the thinking. You can also ask for just the next two hours and come back later for the rest.
Does this work if I never read Weeks 1 through 6?
Completely. The prompt only needs what's in front of you right now. If you did follow the series, you'll get more out of it — the plan you paste in is the one you built in Weeks 2 to 6, and the day you salvage becomes part of the in-trip record Week 8 reconciles. But nobody needs to have read anything to type where they are and what went wrong.
Recommended follow-up prompts
- The Decoder (Variation 2 of this post) — once your rebuilt day lands you somewhere unplanned, you'll need to read the menu.
- A "what did today actually cost?" prompt — a two-line evening tally that feeds the Week 8 reconciliation. The Daily Tally prompt inside Variation 3's kit is the fuller version.
- A "tomorrow's one non-negotiable" prompt — after a disrupted day, ask the AI to help you pick the single thing tomorrow must contain, so a second bad day can't take everything with it.
Tags and categories
Tags:
in-trip prompts, day planning, itinerary repair, travel disruption, rainy day, mobile prompting, beginner prompts, vacation planning Categories: Travel & Vacation Planning, Everyday AI
Citations
NOT APPLICABLE
The Decoder
Turn any menu, sign, or notice into something you can act on.
Translation apps solve the wrong half of the problem. You point your camera at a menu, and now you know that the second item is "grilled ox tongue with green sauce" — which tells you the words but not whether it's the thing everyone in this room is eating, whether it's heavy, whether it's the size of a snack or a commitment, or how to ask if the green sauce has nuts in it. The same gap opens with a sign taped to a station gate, a text from a host in a language you don't read, or a pharmacist's instruction. What you actually need is interpretation plus a script: what this means, what I should do about it, and the exact words to say next. That's three jobs, and a well-shaped prompt does all three in one pass.
Why this matters now
You are standing up, someone is waiting, and the decision is small but not trivial. This is the highest-frequency friction of any trip abroad, and it compounds — travellers who can't read the menu order the one dish they recognise, eat the same thing for nine days, and come home feeling like they missed the country. This prompt is built to be run five or six times a day without ceremony. It also earns its place in the series' record: the meals you actually chose, and what they cost, are part of what Week 8 reconciles against the plan you started with.
Act as a local-savvy interpreter for a traveller in [country or region]. Here is what I'm looking at: [paste the text, or describe the sign or menu].
Reply in exactly four parts, in this order:
1. PLAIN MEANING — what it actually says in natural English, meaning-for-meaning rather than word-for-word.
2. WHAT IT REALLY IS — for food: the main ingredient, how it's cooked, how heavy it is, how spicy on a 1 to 5 scale, and anything a [dietary needs — vegetarian, nut allergy, no pork, none] eater should avoid. For a sign or notice: what it's telling me to do or not do, and what happens if I ignore it.
3. WHAT I'D PICK — [number] options ranked for someone who wants [what you're after — the most local thing, something mild, the fastest cheap lunch, the shareable option], one line each on why.
4. WHAT TO SAY — the exact phrases I need in [language], with a simple pronunciation guide and the English underneath. Include how to ask [your specific question], and how to politely decline.
If any part of the text is unclear, ambiguous, or you're guessing at a regional term, say so in that spot rather than smoothing over it. Don't guess at prices. Keep the whole reply short enough to read on a phone screen.
How the AI reads this prompt
Practical examples from different industries
A solo traveller decoding a menu in an unfamiliar script.
She's in a small place in Osaka with a handwritten board and no English menu, having photographed the board and pasted the extracted text. The four-part reply tells her the third item is a fried skewer set rather than a single skewer, marks two items as heavier than she wants, ranks three options for "the most local thing," and gives her one phrase to ask whether the sauce is shared-dipping style and one to decline politely if it is. She orders in about forty seconds and eats something she'd never have picked from a picture.
A couple reading a notice taped to a station gate.
The sign is three lines of Italian on a sheet of A4 and there is no announcement in English. Described rather than pasted, the prompt returns a plain reading — a strike affecting regional services on specific hours — plus the crucial part 2 line: what this means for them practically, which is that their ticket type may or may not be honoured on later services. Part 4 hands them the sentence to ask at the ticket window, and part 4's uncertainty flag warns that "regional" has a specific technical meaning here that they should confirm. They ask the right question at the window instead of the vague one.
A parent working out whether a pharmacy label is safe for a child.
The dosing instruction is in Greek and the label has a symbol they don't recognise. The prompt reads the instruction, flags explicitly that it is not certain about one abbreviation, and — critically — produces the phrasing to ask the pharmacist directly, including how to state the child's age and weight. This is the case where the prompt's real output is the question, not the answer, and where the uncertainty flag earns its entire existence.
Creative use case ideas
- Reading a rental agreement or a parking sign you signed under time pressure. Photograph it that evening and find out what you actually agreed to while there's still time to act.
- Decoding a handwritten family recipe in a language you half-remember. Nothing to do with travel; the same four-part structure handles it, and part 2's "what it really is" is exactly what you need when the recipe says "a glass of flour."
- Understanding your own country's bureaucracy. Swap "traveller in [country]" for "someone who does not work in insurance" and feed it a claims letter. The role instruction does the same work in your native language.
- Museum labels and historical plaques. Ask it to explain what the label assumes you already know — a different and often better question than what the label says.
- Practising a conversation before you have it. Run the prompt with a hypothetical exchange and ask for the three most likely replies you'll get, so you're not caught out by the answer.
Adaptability tips
The four-part structure is the reusable skeleton; every part can be swapped. Replace part 3's ranking criterion to change what the tool is for — "cheapest that will actually fill me up" and "the thing this region is known for" are two different meals from one menu. Change part 2's schema when the subject changes: for a transport sign, the useful attributes are what to do, by when, and what it costs you to get it wrong; for a medical label, they're dose, frequency, and what to ask a professional. If you're travelling with someone else, add a fifth part asking for a one-line summary you can read aloud to them. And when you're somewhere with no signal, run the prompt once in the morning on the day's likely encounters — the phrases in part 4 are worth having cached before you need them.
Pro tips
- Ask for two candidates when the text is unclear. "If you're unsure, give me both readings and tell me which is more likely" produces far better results than forcing a single answer out of a blurry photo.
- Add "what would surprise a foreigner about this?" as a sixth part. It catches the local convention that isn't written down — the cover charge, the standing-versus-sitting price, the fact that the queue works differently.
- Keep part 4's phrases and reuse them. After three days you have a small personal phrasebook made entirely of sentences you actually needed, which is worth more than any published one.
- Paste the previous reply back in when you get a response you didn't expect. Continuing the thread beats starting a new one, because the model still has the menu.
Prerequisites
You need the text in front of you — either pasted in, or described well enough that the model can work with it. If you're extracting text from a photo, do that first; most AI apps will accept an image directly, and if yours doesn't, any OCR or camera-translate tool will get you the raw text to paste. Decide before you type what your ranking criterion is and what your dietary constraints are, since those are the two blanks that most change the output. Some awareness of the local language's name is useful for part 4 — "Catalan" and "Spanish" are not interchangeable requests in Barcelona.
Required tools
Any general-purpose AI assistant. Image input helps considerably — being able to photograph a menu rather than transcribe it removes the slowest step — and image support is available on the free tiers of most major assistants, though limits vary by provider and change often. Check your own app rather than assuming. An offline note in your phone with the filled-in prompt template saves you retyping it several times a day.
Frequently asked questions
Isn't this just a slower version of a camera translation app?
A camera app gives you words; this gives you a decision. The difference shows up in part 2 and part 3 — knowing that a dish is a shareable plate for two rather than a starter, and that it's the one worth ordering if you want the regional specialty, is not information a translation produces. Use the camera app when you only need to know what a word means, which is often. Use this when you have to choose or say something.
What if the AI gets the translation wrong?
It sometimes will, particularly with handwriting, dialect, regional food terms, and anything abbreviated. That's precisely why the prompt requires uncertainty to be flagged in place rather than mentioned at the end. In practice the failure mode you should watch for is confident smoothing — a plausible reading of something that was actually unclear — so if a flag never appears across several uses, treat that as a reason for more scepticism rather than less. For anything with a real consequence, use part 4 to ask a human.
Can I use this on a free tier?
Yes. The output is short by design and the prompt is a single turn, so it sits comfortably inside free usage limits on the major assistants. The one thing that may be restricted is image upload volume. If you hit a limit, transcribe the key lines by hand — five items typed is still faster than ordering blind.
The reply came back in the wrong order or missing a section. Why?
Usually because the input was thin — a two-word sign doesn't support a four-part answer, and the model quietly collapses sections rather than padding them. That's reasonable behaviour, but if you want the structure enforced, reply with "redo it with all four headings, write NOT APPLICABLE where a section doesn't apply." It's also worth checking that your pasted text didn't contain something that read like an instruction, which can knock the format off course.
Recommended follow-up prompts
- The Pocket Kit Builder (Variation 3 of this post) — the kit includes a compressed version of this decoder, sized for one-handed use.
- A "what am I expected to do here?" prompt — for the social conventions the sign doesn't mention: tipping, queueing, whether you seat yourself, whether you pay at the table.
- An "explain this bill" prompt — feed it the receipt at the end of the meal and ask what each line is, which catches cover charges and service items you didn't know were coming and feeds straight into your daily tally.
Tags and categories
Tags:
translation, menu decoding, travel phrases, structured output, format control, intermediate prompts, in-trip prompts, communication Categories: Travel & Vacation Planning, Language & Communication
Citations
- Anthropic, "Prompt engineering overview," Claude Documentation — https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview (on role assignment, explicit output structure, and separating instructions from data).
- OpenAI, "Prompt engineering," OpenAI API documentation — https://platform.openai.com/docs/guides/prompt-engineering (on specifying desired output format and reducing model guesswork).
The Pocket Kit Builder
Build a labelled set of one-handed prompts before you need them.
The problem with a good mid-trip prompt is that you have to write it during a mid-trip crisis. At the gate, at 22:40, with a cancelled connection on the board and a queue forming, nobody composes a well-structured request. They type "flight cancelled what do i do" and get a paragraph of general advice they could have guessed. The fix is to move the work earlier: build the prompts while you're calm, at home, with your actual trip in front of you, and save them somewhere you can reach in one tap. That's what this variation does — not a prompt that answers a question, but a prompt that generates a small toolkit of prompts, each named for the exact moment it solves. It's the most useful thing an AI can build for a traveller, and it takes about four minutes.
Why this matters now
This is where the series' pieces come together. Your Week 1 budget ceiling becomes a daily check you actually run rather than a number you avoid looking at. The fragile points in the plan you built across Weeks 2 to 6 — the tight connection, the non-refundable booking — become the specific moments the kit is designed around. And the kit's output while you're travelling is the in-trip record: what broke, what you did, what it cost. Week 8's reconciliation and reusable-template build both draw on that record directly, which means the kit isn't just a convenience this week; it's the instrument that captures the data next week reads.
Act as a travel operations designer. Build me a Pocket Kit: a set of short, ready-to-paste prompts I can fire from my phone during a trip, without editing them beyond filling in obvious blanks.
My trip: [destinations], [dates], [who's travelling and any ages], [how we're getting around], [languages I don't read or speak]. My total budget ceiling is [amount] across [number] days, which is [amount] per day. Known fragile points in the plan: [for example, one tight connection, one non-refundable booking, a child's nap window, a medication schedule].
Produce exactly six prompts, each labelled with the moment it solves:
1. DAY BROKE — rebuild a half-lost day around whatever just went wrong.
2. DECODE THIS — read a menu, sign, or message and tell me what to do about it.
3. COUNTER SCRIPT — what to say, and in what order, when a flight or booking fails.
4. UNEXPECTED HOURS — use a free block near where I am, well.
5. DAILY TALLY — check today's spend against my ceiling and tell me what the rest of the trip can afford.
6. ONE THING — pick tomorrow's single non-negotiable when energy is low.
Rules that apply to every prompt you write:
- Under 80 words, usable one-handed, no jargon.
- Every blank is marked [like this] and must be something I can see or already know in the moment — never something I'd have to look up.
- No prompt may ask an AI for a live flight status, a current price, an official compensation amount, or my legal entitlements stated as fact. COUNTER SCRIPT must ask for questions to put to the airline, the order to ask them in, and calm phrasing to use — and must end by telling me to confirm entitlements with the carrier or the posted policy.
- Every prompt must end with an instruction telling the AI to flag its own uncertainty.
After the six prompts, add two things:
- KIT TEST — for each prompt, one sentence on how I'll know it worked, and the single most likely way it fails.
- SAVE THIS — three lines telling me exactly what to paste where so the kit is one tap away on my phone.
Then stop. No preamble, no commentary, no summary.
How the AI reads this prompt
Practical examples from different industries
A couple stranded by a cancelled connection.
They built the kit two days before flying and saved it in their notes app. At 22:40, with the board showing cancelled and a queue of sixty people forming, they open COUNTER SCRIPT, fill in three blanks — carrier, what happened, what they need — and get a short ordered list: ask for rerouting on any carrier before asking about hotels, ask what the airline's policy says rather than what they're owed, get any promise written on the boarding pass or in the app, and confirm the specifics against the carrier's posted conditions of carriage. It doesn't tell them what they'll get. It tells them what to ask for and in what order, which is the part that actually changes the outcome at a counter.
A group splitting a running tally against a shared budget.
Four friends, one shared pot, and the familiar slow drift where nobody wants to be the person raising money. The DAILY TALLY prompt, run each evening by whoever remembers, takes the day's spend as typed lines and compares it to the per-day figure from the Week 1 ceiling. The output they use isn't the total — it's the forward projection: at this rate, the last two days have this much left, and here is the one category driving the overrun. Depersonalising the conversation is the real function, and the kit did that by existing before anyone got uncomfortable.
A parent working around a nap window and a non-refundable booking.
Their fragile-points line named both, so the generated ONE THING prompt asks about energy levels and the fixed booking rather than about interests, and DAY BROKE was built to protect the nap as a hard constraint rather than treat it as a preference. On the fourth day, when the morning collapses, the rebuilt plan comes back with the booking preserved and the afternoon quiet, because the kit was designed around the two things that couldn't move.
Creative use case ideas
- A conference or festival kit. Same architecture, different moments: SESSION CLASH, PEOPLE I SHOULD FIND, ENERGY CHECK. The kit-builder pattern transfers to any multi-day event where the plan will partly break.
- A hospital-stay kit for a family member. Genuinely not a business use: QUESTIONS FOR ROUNDS, WHAT DID THAT MEAN, WHAT TO TELL THE FAMILY. The constraint about blanks you can fill from what's in front of you matters even more here, and so does the rule against asserting anything as settled fact.
- A moving-house week. DAY BROKE and COUNTER SCRIPT need almost no adaptation to handle a delivery that didn't arrive and a utility that isn't connected.
- A first-week-at-a-new-job kit. Named for the moments — WHO IS THAT, DID I MISS SOMETHING, WHAT DO I ASK IN THE ONE-TO-ONE — and built before you're too embarrassed to ask.
- A kit for someone else. Build one for a parent or a teenager travelling alone, with the blanks pre-filled and the language simplified. The generated prompts, unlike your advice, will still be there at the airport.
Adaptability tips
The six slot names are the layer to change first — swap them for the six moments your trip actually has, which for a road trip might include CAR TROUBLE and NEXT STOP, and for a long-haul solo trip might include SAFETY CHECK and CHECK IN WITH HOME. Change the word budget to change the character of the kit: 40 words makes it terse and phone-native, 150 makes it more capable and slower to use. If you're travelling somewhere with poor connectivity, add a rule that each prompt must also work when you can only give the AI one sentence of context. For a longer or more complex trip, ask for the kit in two tiers — a daily set and a rare-emergency set — so the everyday tools aren't buried among the ones you hope never to use. And when the trip is over, feed the kit back in with what actually happened and ask which prompts you never used and which you wished you had; that revision is the version you take next time.
Pro tips
- Run a dry rehearsal before you leave. Take DAY BROKE and COUNTER SCRIPT, invent a plausible disaster, and actually run them. Ten minutes at your kitchen table finds the blank you can't fill at a gate.
- Ask for the kit twice, from two different assistants, and merge. Different models emphasise different moments, and the union of the two is reliably better than either — which is the same reason this site publishes three posts a week.
- Add a seventh slot for the thing you personally get wrong. Everyone has one: overpacking the day, skipping meals, missing the last train. Name it and let the kit compensate.
- Store it somewhere that works offline. A note, a screenshot, a text to yourself. A kit that needs a connection to reach is a kit that fails in exactly the conditions it was built for.
Prerequisites
You need your trip's basics — destinations, dates, who's coming, how you're moving between places — and, ideally, the budget ceiling from Week 1 in both total and per-day form. Spend two minutes before you start listing your fragile points honestly; that list does more work than any other input. Some prior experience with structured prompting helps you read and edit the generated kit critically rather than accepting it whole. Do this while you have a laptop or at least a keyboard and a spare ten minutes — this is the one prompt in this post that isn't meant to be typed on a phone. If you're following the series, have your Weeks 2 to 6 itinerary open for the fragile-points step.
Required tools
Any general-purpose AI assistant that handles a long, multi-part instruction reliably — this prompt is substantially longer than the other two and the frontier tiers hold the constraint list better than the lightest models do. A notes app that syncs to your phone and works offline is effectively required, since the kit is worthless if you can't reach it. Nothing else: no plugins, no integrations, no paid features specific to this task. Build it on a laptop, save it to your phone.
Frequently asked questions
Why generate prompts instead of just asking the questions when problems arise?
Because the quality of what you type collapses under stress, and the moments this kit covers are exactly the stressful ones. A prompt written calmly includes the constraints you'll forget — who you're with, what's non-refundable, that you can't look anything up right now — and a prompt typed at a gate includes none of them. There's a second benefit: the kit forces you to think through your trip's failure modes in advance, which occasionally prevents the failure entirely.
Can the AI tell me what compensation I'm entitled to for a cancelled flight?
No, and the prompt explicitly forbids it from trying. Entitlements depend on the carrier, the route, where the flight originated, the reason for the disruption, and the jurisdiction — and rules and thresholds change. What the kit gives you instead is the sequence of questions to ask, phrasing that keeps the conversation productive, and a clear instruction to confirm the specifics against the carrier's posted policy or the relevant regulator. Ask what the airline's policy provides in this situation and get the answer in writing; that's a stronger position than quoting a figure you got from an AI.
How long does this actually take to build?
About four minutes to generate, and ten to fifteen if you do it properly — reading the six prompts critically, cutting the one that doesn't fit your trip, and running a dry rehearsal on the two you'd most hate to get wrong. That's the whole investment. It is comfortably the best time-to-value ratio in this eight-week series, and the resulting kit is reusable across trips with a five-minute edit.
What if the generated prompts aren't good?
Judge them against the KIT TEST section the prompt asks for — if a prompt's stated failure mode sounds like something that would actually happen to you, that's your signal to revise it. The fastest fix is to reply with the specific problem: "prompt 3 asks for my fare class, which I won't know at a gate — rewrite it using only what's on my boarding pass." Iterating on a generated artifact is normal and expected; the first pass is a draft, not a delivery.
Do I need to have followed the earlier weeks of the series?
Not to use the kit. The only place the series shows up directly is the budget ceiling in the DAILY TALLY slot, and any number you're willing to commit to works there — pick a daily figure you'd be uncomfortable exceeding and use that. If you did follow Weeks 1 through 6, the kit gets sharper, because your fragile points are already documented and the ceiling is already calculated.
Recommended follow-up prompts
- A post-trip debrief prompt — feed the kit and what actually happened back in, and ask which slots earned their place. This is the direct on-ramp to Week 8's reconciliation.
- A reusable-template prompt — strip your trip's specifics out of the kit and turn it into a blank you fill in for the next trip, which is the systems move this whole site is built around.
- A "who needs to know?" prompt — when travel breaks, someone at home, a host, or a colleague needs a short accurate message. A slot that drafts three of them at once pairs naturally with COUNTER SCRIPT.
Tags and categories
Tags:
meta-prompting, prompt kits, travel disruption, budget tracking, structured output, evaluation criteria, advanced prompts, reusable systems, in-trip prompts Categories: Travel & Vacation Planning, Advanced Prompt Engineering
Citations
- Regulation (EC) No 261/2004 of the European Parliament and of the Council, establishing common rules on compensation and assistance to air passengers in the event of denied boarding, cancellation, or long delay of flights — cited as the kind of authority a traveller should consult directly, not as a statement of what any individual reader is owed.
- U.S. Department of Transportation, Office of Aviation Consumer Protection — https://www.transportation.gov/airconsumer — the official U.S. source for airline consumer rights and for airlines' own published commitments.
- Convention for the Unification of Certain Rules for International Carriage by Air (Montreal Convention, 1999) — the treaty framework underlying international carrier liability for delay and baggage, again cited as a source to consult rather than a rule applied here.
Which of the three should you use?
The three prompts sit at different distances from the moment of trouble. The Day Rescue is written to be typed while the problem is happening — it assumes nothing, needs no preparation, and trades sophistication for the ability to be used badly and still work. The Decoder sits half a step back: it has a fixed four-part structure and four adjustable parameters, which means it rewards a reader who has thought for ten seconds about what they actually want before typing. The Pocket Kit Builder sits furthest back of all, on a laptop, days before departure, and its output is the thing you use in the moment rather than the moment itself.
They overlap deliberately. The kit's DAY BROKE slot is a compressed descendant of Variation 1, and its DECODE THIS slot is a compressed descendant of Variation 2. That's not redundancy — it's the actual lesson of the week. A good prompt used often enough becomes a tool, and a set of tools with names becomes a system. You can watch that happen across these three variations in the order they're written.
Choose by how much time you have before you need it. If you're reading this on a phone in a doorway with rain coming down, use Variation 1 right now and come back to the others tonight. If you're at a table with a menu you can't read, Variation 2 is the one. If you haven't left yet — or you're mid-trip with a quiet evening and a trip that's already broken once — build the kit, because the fourth day will break too, and you'll be a lot better prepared for it than you were for the first.
TAGS:
