# Stop Getting Generic Answers From AI: The 5-Part Prompt Recipe > Bland AI answers usually come from short requests. A 5-part prompt recipe for small business owners, with copy-ready templates and before/after examples. - Author: Tai Pham Phu - Published: 17 Oct 2026 - Tags: Applied AI, Small Business, Automation, Business - URL: https://phamphutai.com/2026/10/17/the-5-part-prompt-recipe/ You ask AI to write a post for your business. You type one line: *"write a post about my salon."* And you get this: > ✨ Welcome to our salon, where beauty meets passion! ✨ Our talented team is dedicated to making you look and feel your best! It could be any salon, in any city, in any year. Nobody you know talks like that. So you close the tab and decide AI just isn't very good. Here's the thing. If you gave the same tool five short pieces of information, you would get something you could actually post. Same AI. Different request. This post is the written version of my video on the same topic. It covers why AI answers come out bland, the five-part recipe that fixes it, four before-and-after examples, and copy-ready templates you can paste into whatever AI chat tool you already use. --- ## Why AI answers come out bland To fix the problem, it helps to know where it comes from. An AI chat tool writes by predicting the most likely next word, over and over. (I explain this in more detail in [What an "AI Agent" Actually Is](https://phamphutai.com/2026/10/14/what-an-ai-agent-actually-is/).) That's a powerful trick, but it has one big side effect. When you give it a short request, it has to fill in everything you didn't say: - **Who's the business?** It doesn't know, so it picks "a salon". - **Who's reading?** It doesn't know, so it writes for "everyone". - **What tone?** It doesn't know, so it goes with "✨ exciting ✨". - **How long?** It doesn't know, so it writes the usual length. Every gap gets filled with the **most typical answer**. The average salon. The average customer. The average post. Put enough averages together and you get something that sounds like nobody. So a bland answer usually isn't a bad AI. It's a short request. The AI didn't know anything about you, so it wrote about nobody in particular. The fix isn't a magic phrase or a secret trick. **It's telling the AI what you'd tell a new employee on their first day.** Who we are, what this job is, what the facts are, how we like things done, and "ask me if you're not sure." Throughout this post I'll use one example. Mai runs a small hair salon. She's fictional, and so are all her prompts and the AI's answers below. I wrote them to show the idea. But if you run a small business, you'll probably recognise every one of them. --- ## The 5-part recipe Here's the recipe. Five parts. You won't need all five every time, but the more you add, the less the AI has to guess.

Five short additions to your request: each one removes a gap the AI would otherwise fill with the average.

### Part 1: Who you are One or two sentences about your business. Not your life story. Just enough that the AI stops imagining a generic one. > "I run a small hair salon in a quiet neighbourhood. Most of my clients are regulars who've been coming for years." That single line already changes the tone of everything that follows. Regulars, quiet neighbourhood: the AI now knows not to write like a launch party. ### Part 2: The task, and who it's for Not "write a post". Say what the thing is, where it goes, and who will read it. > "Write an Instagram post for my regulars, announcing that we're now open on Sunday mornings." The reader matters as much as the task. A message to a regular of ten years and a message to a stranger who found you on Google should not sound the same. ### Part 3: The facts Anything the answer must get right. Times, prices, dates, names, addresses, policies. > "Sundays, 9 a.m. to 1 p.m., starting the first of next month. Same prices as weekdays." This part is not optional when facts matter. **If you don't give the AI the facts, it will make them up.** Not because it's lying, but because it's predicting what a typical post would say, and a typical post has a time and a price in it. If you don't supply yours, you'll get plausible-looking ones. ### Part 4: The shape, and ideally an example How long, what tone, what to avoid. > "Under 60 words. Warm, not salesy. No exclamation marks, and at most one emoji." The most powerful version of part four is **an example**. Paste one of your own past posts or messages and say: *"Match the style of this."* An example beats any description of a tone. "Friendly but professional" means a hundred different things. A real message you wrote last month means exactly one thing. ### Part 5: Let it ask End your request with: > "Before you write, ask me any questions you need." It sounds small. But now the AI fills the gaps with **your answers** instead of with the average. This is the part people skip, and it often does the most work. More on it in a moment. ### The recipe in one line **Who you are. The task and the reader. The facts. The shape. And let it ask.** --- ## The copy-paste template Here's the whole recipe as a template. Copy it into a note, fill in the brackets, and paste it into any AI chat tool. ``` I run [your business, in 1-2 sentences: what you do, where, who your customers are]. Write [the thing: post, reply, message, email] for [who will read it]. Facts it must get right: - [time / date] - [price] - [names, address, policy] Shape: - Length: [e.g. under 60 words / 3 lines / 2-3 sentences] - Tone: [e.g. warm and plain, not salesy] - Avoid: [e.g. exclamation marks, emojis, discounts, hype words] - Match the style of this example I wrote before: "[paste one of your own past messages]" Before you write, ask me any questions you need. ``` You won't fill every line every time. A quick message to a staff member doesn't need an example. But when the result will reach a customer, try to fill them all. --- ## "Ask me first" in action Part five deserves a closer look, because it changes how the whole conversation works. When Mai added "ask me any questions first" to her Sunday-opening request, the AI didn't write a post straight away. In this example, it came back with three questions: 1. Is it the full menu on Sundays, or a shorter list of services? 2. Do clients book online or by message? 3. Is there anything special about Sunday mornings you want to mention? Mai answered in about ten seconds: shorter menu, cuts and blow-dries only. Book by message. It's quieter, no rush. Every one of those answers ended up in the final post. And none of them were things she would have thought to include up front. That's the real benefit. **You don't have to think of everything in advance. You just have to let the AI ask.** It's very good at spotting what's missing from a request. It's just that, by default, it fills the gap itself instead of asking you. --- ## Four before-and-afters Here's the recipe on four tasks most small businesses do every month. All four are fictional examples written for the video, but the "before" answers are very typical of what a one-line request produces. ### 1. The Instagram post **Before.** Mai typed: ``` write a post about Sunday opening ``` She got something like: > Big news! ✨ We're thrilled to announce exciting new hours! 🎉 Come treat yourself! 💇‍♀️💅 **After.** With the recipe: ``` I run a small hair salon in a quiet neighbourhood. Most of my clients are regulars who've been coming for years. Write an Instagram post for my regulars, announcing that we're now open on Sunday mornings. Facts: Sundays 9 a.m. to 1 p.m., starting the 1st of next month. Same prices. Cuts and blow-dries only on Sundays. Book by message. Shape: under 60 words, warm, not salesy, no exclamation marks, one emoji at most. Match the style of this post I wrote before: "[paste a past post]" Before you write, ask me any questions you need. ``` The result, in this example: > Some of you asked, so we listened. From the 1st of next month we're open Sunday mornings, 9 to 1. Same team, same prices, slower pace. Cuts and blow-dries only. Message us to book a quiet Sunday trim. Same tool. The second one sounds like Mai, because Mai told it who she is and showed it how she writes. ### 2. The bad review reply Review replies are where a bland answer can actually hurt you, because they're public and they can sound like promises. **Before:** ``` reply to this review: [one-star review pasted] ``` In this example, the AI apologised three times, called Mai "the management team", and offered *"20% off your next visit"*, a discount Mai never agreed to. If she had posted that, she would have had to honour it. **After.** She added the facts and the shape: ``` I run a small hair salon. I reply to reviews myself and sign as Mai. Write a public reply to the one-star review below. Facts: the client arrived 20 minutes late, so we shortened the service. We explained this to her at the time. Shape: polite, 2-3 sentences. Acknowledge her frustration. Explain our late-arrival policy without blaming her. Invite her to talk to me directly. Do NOT offer a discount or any compensation. Review: "[paste the review]" ``` The reply came back calm, short and fair. It didn't promise anything she'd have to honour. Notice the line **"Do NOT offer a discount."** Telling the AI what *not* to do is half the recipe for anything a customer will read. AI tools tend to be eager to please, and "please the customer" often turns into "give the customer something." Rule it out in writing. ### 3. The voice note This is the one Mai uses most. Between clients, she records a messy voice note, gets it transcribed (many phones and chat apps can do this), and asks AI to turn it into a message. **Before:** ``` make this a message: [transcript pasted] ``` The AI kept every "um", every half-sentence, and her note to herself about buying more shampoo. Not something you'd send to a colleague. **After.** She added the task and the shape: ``` This is a voice note I recorded to myself. Turn it into a short WhatsApp message to Lan, my colourist, about next week's schedule. Keep ONLY the schedule changes. Leave out anything else (those are my own reminders). Shape: 3 lines at most, friendly, plain. Voice note: "[paste the transcript]" ``` Three clean lines came back, ready to send. The shampoo stayed on her own to-do list. The key line here is "keep only". When you paste in messy material, tell the AI what to keep and what to throw away. Otherwise it assumes everything you pasted is important. ### 4. The price increase The message nobody likes writing. **Before:** ``` write a message about our new prices ``` The AI wrote a corporate letter: *"Due to rising operational costs, we regret to inform you..."* Fine for a bank. Strange from someone who has cut your hair for eight years. **After.** Mai told it who reads it, gave it the exact numbers, and pasted a real message she'd sent her clients before as the example: ``` I run a small hair salon with two stylists. Most clients are regulars I've known for years, many over fifty. Write a short message to my regular clients about a price change. Facts: - Women's cut: [old price] -> [new price] - Men's cut: [old price] -> [new price] - Colour: unchanged - New prices start on [date] Shape: short, honest, warm. No corporate language like "operational costs" or "we regret to inform you". Sign it "Mai". Match the style of this message I sent my clients before: "[paste a past message]" Before you write, ask me any questions you need. ``` The result was short, honest and warm. It sounded like a text from Mai, not a letter from a bank. --- ## Three habits that make it stick Typing all of this every time sounds like work. These three habits mean you mostly do it once. ### Habit 1: Save your "who I am" paragraph Write it once, keep it in a note on your phone, and paste it at the top of every request. Better still, most AI chat tools now let you save standing instructions so they're included automatically. ChatGPT calls them **custom instructions**. Google Gemini lets you build a **Gem**, a saved set of instructions you can reuse. Other tools have similar features under different names. At the time of writing (October 2026), check your tool's settings for something along those lines. Here's Mai's, as an example: ``` I run a small hair salon with two stylists. Our clients are mostly regulars, many over fifty. We're friendly and plain-spoken; we never use hype words or lots of emojis. We sign messages "Mai". ``` Five lines, written once, used in every request. ### Habit 2: Fix, don't restart If the answer is close but not right, don't start a new chat. Say what to change: ``` Shorter. Less formal. Remove the second sentence. Say "book by message", not "book now". ``` The AI keeps everything it got right and only changes what you asked. Starting over throws all of that away. Mai's favourite fix is just three words: **"Sound more like me."** Because her "who I am" paragraph and an example message are already in the conversation, the AI knows what that means. ### Habit 3: Build a small prompt library When a prompt works, save it. A note called "AI prompts" with a few entries is enough: - **Review reply** - **Price change** - **Holiday hours** - **Voice note to staff message** - **Answer a guest/customer question** Next time, you copy the prompt, change the facts, and reuse the rest. The thinking is already done. If customer replies are your biggest time sink, I go deeper on that in [Automate Your Guest Messages Without Sounding Like a Robot](https://phamphutai.com/2026/10/14/automate-guest-messages-without-sounding-like-a-robot/). The same principle applies: your voice, your facts, written once and reused. --- ## When a short prompt is fine To be fair, you don't need the whole recipe for everything. **One line is fine** when there's no single right answer to miss: - Ten name ideas for a new service - A quick summary of a long email - A nicer word for "busy" - "Explain this term in plain English" **Use the recipe** when the answer: - **will go to a customer**, - **has facts in it** (prices, dates, times, policies), or - **needs to sound like you**. That's where the extra thirty seconds pays back. --- ## One warning A better prompt gets you a better draft. It does not get you a guaranteed-correct one. Check every fact before it goes to a customer, especially **prices, dates, and anything that sounds like a promise**. Even when you've given it the right facts, the AI can still mix them up, round a number, or add a nice-sounding line that commits you to something. The recipe makes the AI more useful. It doesn't make it responsible. You are still the one who presses send. That's also why, when we build [Santara](https://phamphutai.com/projects/santara/), the briefing links every item back to its source message or booking: so the host can check it in a few seconds instead of trusting it blindly. If you want to go further on judging what AI tools can and can't be trusted with, [How to Tell If an AI Product Is Real or Just a Demo](https://phamphutai.com/2026/10/10/how-to-tell-if-an-ai-product-is-real-or-just-a-demo/) is a good next read. --- ## The takeaway - Bland AI answers usually come from **short requests**, not bad AI. The tool fills every gap with the average. - Fix it with five parts: **who you are, the task and the reader, the facts, the shape (with an example), and "ask me first."** - Save your "who I am" paragraph, fix instead of restarting, and keep a small library of prompts that worked. - Use one-line prompts for ideas and summaries. Use the full recipe for anything customer-facing, factual, or in your voice. - **Always check the facts before it reaches a customer.** The AI isn't the one being lazy. It just doesn't know you yet. Tell it, and it writes like it does. *Sources: [OpenAI Help Center: ChatGPT Custom Instructions](https://help.openai.com/en/articles/8096356-chatgpt-custom-instructions); [Google Gemini Help: Get started with Gems](https://support.google.com/gemini/answer/15235603). Mai and her salon are a fictional example; all prompts and answers in this post were written to illustrate the recipe.* *Related: [AI Won't Replace Small Business Owners. It Will Replace Their Busywork.](https://phamphutai.com/2026/10/09/ai-will-replace-busywork-not-small-business-owners/), and [The Hidden Costs of AI Tools Nobody Shows in the Demo](https://phamphutai.com/2026/10/14/hidden-costs-of-ai-tools/).*