How hotels can use AI prompts? 20 Best Practices for Hotel AI Prompts
- David Lamelas
- 22 de set. de 2025
- 13 min de leitura
Atualizado: 24 de mai.
“An excellent model with a weak prompt can be outperformed by a weaker model with an excellent prompt.”
David Lamelas
Expanded Prompt Engineering Guide
A prompt is the instruction or question written to guide an AI model to give an answer, a result, an image, a video, that is, some output; it is the statement or instruction that tells the AI (NLP) what to do.
Think of the prompt as the request you make to a chef: if you say “I want food,” you get something generic; if you explain the dish, portions, dietary restrictions, and whether you like your steak well or rare, the result improves a lot. The same happens with AI.
For those technically curious: technically, the prompt is converted into tokens (text fragments) that the AI compares with patterns learned in its base training to predict the next sequence of words and build the response. Remember? LLMs (Large Language Models) or NLPs (Natural Language Processing), or more simply, the ChatGPT‑type models, are just AI models that build sentences very well, so well that they seem super intelligent and seem to know a lot, and they do (in the way I just described, building sentences, word by word, syllable by syllable).
What’s interesting: small changes in how a prompt is formulated, role, objective, limits, etc… (see more tips below), can drastically change the “reasoning” or style of the response. Like dialing a phone number: if you get one digit wrong, you call the wrong person. But in this case it’s even more frustrating, because often, with the same prompt you can get slightly different responses… As if dialing the same phone number called slightly different people. So it’s more complex than that, I’m sorry to say. That’s why it’s so important to be precise in prompt engineering, in the detail, in the fine tuning, and follow the best practices that I point out below, in order to minimize that natural randomness that NLPs (the ChatGPT‑type models) have, so that the answer is always more or less identical, and above all meets your request and the result is what you want.
So who is going to make the revolution of applied AI in hospitality?
In my opinion, it will be the smart hoteliers who master prompt engineering very well, because they are the ones who know which question is correct. Many just don’t yet know how to master it and take advantage of it with AI. Because part of the success of AI lies in the prompt, in the question you ask. Therefore, it is important to have hospitality‑industry specialists develop solutions for hospitality, because they are also those specialists who know how to ask the right question and give the right instructions without leaving any rabbit out of the hat.
Golden tip for beginners or the lazier:
By the way, golden tip: when you don’t know what quality prompt you should write, explain in simple words what you want to ChatGPT (or another LLM) and ask it to write the ideal prompt for you (also say which model you are going to use that prompt with).
Then use that prompt and that’s it. That is, take that prompt, copy to the model you want to use, make a few adjustments if you want, and you’re done, just wait for the result. It seems silly, but it works very well for those who are not yet very comfortable with prompt engineering or are at a more beginner level. But I don’t recommend this practice much, because that way you’ll never learn prompt engineering. But when hurry knocks at the door, it can also be a good solution, even for the more experienced.
✅ Best practices and what a good prompt should have and how it should be structured.
How can I create my prompt for the hospitality sector?
How to create my prompt for my hotel?
How hotels can use AI prompts?
Give a clear role
- Say who the AI should “be” or which specialty it should follow: “Act as a prudent financial analyst,” “You are a 9th‑grade teacher,” “You are a B2B copywriter,” “Act as a lawyer,” etc…
- If it makes sense, add seniority and sector.
- For example, if you ask “Act as a lawyer,” the AI will seem like a real legal adviser.
Expose the problem
- State the problem or situation, this helps give context.
- It doesn’t need to be a novel though. Don’t tell the story of your life, be concise.
Clear objective and clear tasks
- Explain what you want: a question, task, analysis, or ideas.
- Declare the desired result in one sentence.
- Example very simple: “I want a 150‑word summary for laypeople.”
Separate context from instructionsCreate sections. It works better like this:
- Context: 2‑3 useful lines.
- Task: what to do.
- Restrictions: limits and policies.
- Output: final format.
Give necessary details
- Audience, level of knowledge, language (PT‑PT), tools, deadlines, word limits.
- Reduce ambiguity
Use delimiters
- Isolate content with code blocks, <tags> or “Text: …”. Avoid confusion.
- Help the AI understand what is an instruction and what is a text you want to change.
- For example: Translate the following text into English: “Sem prejuízo do disposto ….”.
Show what you want (and what you don’t want)
- A good example is worth gold. It will help the AI understand your expectations about the expected result. Just like us humans, we always understand better when we are given an example.
- For example write: Here’s an example of the type of response to give to a customer in a complaint: “Dear Mr. Tiago Gonçalves, Thank you very much for bringing this point to our attention ….”
- Bring also a short anti‑example to exclude undesired results (for those who already know the quirks of some AI models, know how important it is to give anti‑examples).
Chain of Thought
- Structured / step-by-step reasoning
- This capability became much more present in our lives in 2024/2025, with models such as OpenAI o1 and o3 and others from Gemini, Claude or DeepSeek also.
- It is especially useful for tasks involving logic, mathematics, planning, analysis, or more careful decision-making.
- Instead of asking only for a final answer, we can ask AI to better structure its reasoning, analyse the problem in stages, and justify the approach being used.
- This helps avoid answers that are too quick, superficial, or poorly grounded.
- That is why, nowadays, it is not always necessary to literally write “think step by step”. Some models have already been designed to reason more effectively before answering.
- However, it is important to be careful: not all models, tools, or APIs have reached this level of evolution. So it is essential to know exactly which one you are using.
Restrictions and limits
- It is very important to say what you do not want!
- After some time we already know the tricks of a given model, and then we know we must say to avoid those tricks.
- For example: “Never use em‑dashes —”. By the way, another tip, if you don’t want it to be obvious you used AI, don’t use “—”, because that “—” gives away that you used AI, since that “—” was never used until ChatGPT appeared. Actually I challenge you to try to find on your keyboard that key to type that em‑dash “—”. Understood now?
- Declare costs, reading time you have, applicable policies and requirements for privacy (anonymization of sensitive data).
Format and length
- Say if you want a list, table, bar chart, steps, JSON… say how you want it.
- Indicate counts.
- For example: “Give 10 different suggestions”, “A list with max. 6 bullet points”.
- In translations/summaries, ask to keep formatting, emojis, line breaks, tags.
Ask for reliable references
- On factual topics, ask for sources and links.
- Ask for links to news so you can validate facts.
Reduce hallucinations
- Another golden tip, to lessen hallucinations use this tip below.
- If the AI does not have a basis to prove those facts, ask it to say “I don’t know / low confidence”. Include in all prompts something like: “If you are not 100% sure about an answer or a fact say you don’t know or that the confidence level is low; don’t worry about appearing less certain, it’s normal not to know everything; feel free with me, tell me when there is something you don’t know. The worst you can do is invent something just because you are not 100% sure.”AI models work via probabilities so they understand this part “If you are not 100% sure about an answer or fact say you don’t know or that the confidence level is low…”.
- Reduce ambiguity
Demand justification, but no essays
- Ask for criteria and determining factors in short points (avoid long reasoning when you only care about the result).
- Ask for criteria and factors in bullet points. Short and direct.
- Ask the AI to explain its criteria in short bullet points, focusing only on the key factors that justify the answer.
The more complete the prompt, the better the result
- A bit more precision now saves a lot of rework later.
Address gaps head‑on
- Allow the AI to ask for clarification if something critical is missing.
- If you proceed with assumptions, have them listed.
- Authorize a request for clarification when essential data is missing or, alternatively, explicit assumptions.
- For example include: “If my prompt is not clear, ask for the clarifications you need in order to give a 100% correct and assertive response.”
Ask for creative variety
- If you want a richer response with multiple angles, styles, or options, explicitly ask the AI for it.
- Ask for several responses and put them in competition.
- You can request the AI to generate 3 distinct options (e.g., low, medium, and high effort or conservative vs. creative tones) and then put them in competition or ask for a final synthesis.
- This prevents the AI from giving you just a single, predictable answer and opens room for fresh ideas.
Give enough context, not a novel
- Cut what is superfluous. Summarize long attachments in points.
- Sometimes if you include something not that essential, it can confuse the AI; it may focus too much on that part which you put and wasn’t essential.
- Here, as in everything, sometimes less is more.
- Detail makes all the difference, but neither extreme.
For code and data
- Specify language/version, environment, I/O, test cases and expectations of complexity.
- Reduce ambiguity. Strictly define each parameter, variable, the data, the calculations, the data excluded from the analysis, how to treat blanks, etc.
Reduce ambiguity
- Strictly define each parameter, variable, the data, the calculations, the data excluded from the analysis, how to treat blanks, etc. For example, if your prompt includes phrases like: '...Consider any review with enough text to identify at least one positive or negative theme as an "analyzable review"....'. The AI will decide what 'enough text' means, resulting in inconsistent data. You need to define what constitutes 'enough text' yourself, or just remove these ambiguous terms entirely.
Output evaluation
- Request a score (0–10) against criteria and suggestions for improvement of the output or prompt itself.
- Always leave space for the AI to be self‑critical of itself or whatever it is.
Test → measure → refine; if necessary restart from scratch
- Do A/B of prompts. Compare against acceptance criteria. Adjust and repeat.
- Sometimes you don’t hit the right prompt at first. In that case, sometimes the best is to understand what failed and restart, putting that learning about what failed in the next prompt and try again.
- Sometimes it drives you crazy, trying to correct a result via iteration (adjusting and tweaking the result given by an AI through conversation/prompts), because no matter how you explain, the AI may not understand. For example many times text you request to be in an image ends up off‑lines or misaligned, so sometimes it is harder to correct that mistake than to restart the conversation and include that detail from the start (“capable of fitting the text inside the image”, “if necessary reduce font size of the text so it fits in the image”).
🧪Very simple template ready to paste
I know it’s not always possible to follow all best practices in day‑to‑day. So here is the bare basics, just adapt.
Role: Act as an academic synthesizer.
Explain problem & objective: Expose {problem}. Explain {theme} for {audience}.
Context: {2‑3 sentences} or {a table}
Task: Produce a summary in 5 points.
Audience: For hoteliers
Restrictions: Max. 180 words; PT‑PT; without jargon.
Format: Define response length or format
Sources: Cite 2‑3 reliable sources. To confirm if it is true.
Output: Markdown with title, list and section “Limitations”.
✔️ Quick checklist before sending the prompt. How hotels can use AI prompts?
Assign a role to the GPT
Tone, language, align with audience
Define clear objectives (command, question, instruction)
Split your request (e.g., separate instruction from context)
Include essential details, be descriptive
Use delimiters (e.g., “text”)
Expectations, show examples or what you don’t want
Restrictions and limits, say what you don´t want
Define response length or format
Reduce ambiguity
Reference reliable texts
Eliminate hallucinations; if not ≥100% probable, say “I don’t know”
Request or show Reasoning “step by step”. If the model you are using does not yet include the "Chain of Thought" functionality.
The more complete the prompt, the better the result
Test, Test, Test (experiment > results > refine) and often it’s best to start over with the learning you got, if the first result wasn’t satisfactory.
🤠 Alternative Prompting Techniques for Hotels
Beyond the basic structure of a good prompt, there are a few practical prompting techniques that can help hoteliers get better results faster, especially when dealing with marketing, guest communication, reports, operations, or strategy.
1) The “Lazy” or “In a Hurry” Method: Ask AI to Create the Prompt for You
When you do not know exactly what technical instructions to use, ask the AI to build the prompt for you.
For example:
“I am a hotel professional and I need to use ChatGPT to analyse a P&L report. Act as a world-class prompt engineering expert and write the ideal detailed prompt that I should copy and paste into ChatGPT to get the best possible result.”
This is a very useful shortcut when time is limited. It allows non-technical hotel professionals to get a more structured and professional prompt without starting from scratch.
However, this does not mean you do not need to learn prompt engineering. To ask AI to create a good prompt, you still need to understand the basics: role, context, objective, constraints, format, examples, and expected output. In other words, AI can help you write the prompt, but you still need to know whether the prompt is good.
2) Interview-Style Prompting: Let the AI Ask You Questions First
Instead of trying to guess all the context the AI needs, ask it to interview you before generating the final answer.
Example:
“I need to create a 3-day activity programme for a VIP family staying at our hotel. Before writing the itinerary, interview me first. Ask one question at a time about the guests’ ages, preferences, dietary restrictions, budget, mobility, and special interests. When you have enough information, generate the final itinerary.”
This technique is especially powerful in hospitality because many tasks depend on variables that are easy to forget: guest profile, season, budget, hotel positioning, brand tone, destination, dietary needs, and operational limitations.
It also reduces assumptions. Instead of the AI inventing missing information, it asks for what it needs first. This leads to more accurate, more personalised, and more useful outputs.
3) Prompt Chaining: Divide Complex Tasks into Steps
For complex hotel tasks, do not ask the AI to do everything in one single prompt. Break the work into smaller steps and use the output of one step as the input for the next.
Example for a hotel marketing campaign:
Step 1 — Strategy:
“Analyse our target audience and propose 4 campaign angles for Easter stays.”
Step 2 — Expansion:
“Take campaign angle number 2 and write a landing page outline.”
Step 3 — Adaptation:
“Now transform the first two sections of the landing page into a LinkedIn post for corporate event organisers.”
This method works well because each step has one clear job. It also gives the hotelier a chance to validate the result before moving forward. For tasks such as website copy, email campaigns, SOPs, guest journeys, upselling flows, or revenue analysis, prompt chaining usually produces better results than one long, overloaded prompt.
4) Create Your Own Personal GPT or AI Agent for Prompts
A more advanced but very practical approach is to create your own GPT or AI agent trained with your preferred style, tone, hotel positioning, brand voice, and recurring use cases.
For example, you can create a personal AI assistant that already knows:
your hotel type, your audience, your writing style, your brand tone, your preferred structure, your common tasks, and the kind of prompts you usually need.
Then, whenever you need help, you can ask:
“Create a prompt in my usual style to generate a luxury hotel newsletter for returning guests.”
Or:
“Improve this prompt so it matches my tone and produces a more professional hotel marketing output.”
This is useful because it saves time and creates consistency. Instead of explaining your style every time, your personal GPT or AI agent becomes a prompt assistant adapted to the way you work.
The best use case is not replacing your thinking, but accelerating it. You still decide the objective, judge the quality, and validate the final output.
In short: when using AI in hotels, the best results often come not from one perfect prompt, but from choosing the right prompting method. Ask AI to help build the prompt when you are in a hurry, let it interview you when the context is incomplete, chain prompts when the task is complex, and create a personal GPT when you want consistency at scale. These techniques complement the core best practices of clear roles, context, objectives, constraints, format, and refinement already covered in this guide
❓ Frequently Asked Questions (FAQs)
1. What is a prompt in AI and why does it matter for hotels? A prompt is the instruction or question you give to an AI model to generate a response. In hotels, better prompts mean more accurate guest communication, improved marketing content, and smarter automation.
2. How can AI prompts improve guest experience in hotels? AI prompts can personalize guest communication, answer questions faster, provide tailored recommendations, and create a smoother booking or check-in experience.
3. Why is prompt engineering important for hospitality? Because asking the right question leads to better results. Hoteliers who master prompt engineering can make AI tools more efficient and aligned with real guest needs.
4. Can AI prompts help reduce hotel staff workload? Yes. With well-structured prompts, AI can handle repetitive tasks like FAQs, booking confirmations, upselling offers, and guest feedback, freeing staff to focus on human interaction.
5. What are some examples of AI use cases in hotels? Chatbots for guest support, upselling through personalized offers, automated review replies, multilingual translation for international guests, and AI-driven market analysis.
6. What makes a “good” AI prompt for hotels? A good prompt is clear, structured, includes context (e.g., guest profile, preferences), defines the expected format, and avoids ambiguity.
7. How can hotels avoid AI “hallucinations” or wrong answers? Hotels should instruct AI to say “I don’t know” when uncertain, set strict guidelines in prompts, and always ask for sources or low-confidence warnings.
8. Do hoteliers need technical skills to use AI prompts effectively? No. While technical knowledge helps, most hotel professionals can learn prompt best practices through examples and structured templates.
9. What are the main benefits of using AI prompts in hotel marketing? Faster content creation, consistent brand tone, personalized campaigns, and better SEO-friendly text for websites and social media.
10. What’s a quick template hotels can use to start with AI prompts? Define the role (e.g., “Act as a hotel concierge”), provide context (guest request, language, preferences), state the task (summarize, reply, suggest), add restrictions (word count, tone), and request clear output (list, email, script).