Updated November 2026
A simple formula for getting useful answers from any AI chatbot, tested in five of them, with examples you can copy.
The short version: Most weak AI answers come from prompts that leave out what the AI needs to know. Give it five things: a role, a clear task, the context, the format you want back, and any constraints. Then treat the first answer as a draft and ask for changes. We tested it in ChatGPT, Claude, Gemini, Grok and Meta’s Muse, It should work in Perplexity too, though we didn’t test it there. Every prompt on this page has a Copy button.
Why do most beginner prompts fail?
Usually for one reason: the prompt doesn’t tell the AI enough. “Write me a blog post about productivity” leaves it to guess everything that matters: who the reader is, how long it should be, what tone to use, what to leave out. The AI fills those gaps with safe, middle-of-the-road choices, which is why vague prompts get bland answers.
The fix is to be specific. ChatGPT, Claude, Gemini, Grok, Muse and Perplexity all respond better to the same structure, laid out below.
What’s the 5-element prompt framework?
Anthropic, OpenAI and Google each publish their own prompting advice (Anthropic, OpenAI, Google). The wording differs, but the advice keeps coming back to the same five things. Use them as a checklist for anything more than a quick question:
- Role: who the AI should be. “You are a small-business accountant.” “You are a friendly career coach.” Anthropic says that even “a single sentence makes a difference” to Claude’s behavior and tone (Anthropic).
- Task: what to do, with one clear verb. “Draft a polite reminder email,” not “help me with this email.”
- Context: the background it needs. Who the reader is, what has already happened, what facts to assume. “This is the second reminder. The client is friendly. The invoice is 14 days overdue.”
- Format: the shape of the answer. Bullets, a numbered list, paragraphs, a table. “4 short paragraphs, plain text, no subject line.”
- Constraint: limits and things to avoid. Word count, reading level, what not to include. “Under 120 words. No legal threats. Don’t mention late fees.”
Here are all five in one prompt:
Role: You are a small-business accountant. Task: Draft a polite reminder email to a client whose invoice is 14 days overdue. Context: This is the second reminder; the relationship is friendly. Format: 4 short paragraphs, plain text, no subject line. Constraint: Under 120 words, no legal threats, do not mention late fees.
We tested this on October 30, 2026 in the free version of ChatGPT, against the vague version, “Write me an email about a late invoice.” The picture at the top of this page shows both answers side by side.
From our testBoth answers were usable. The difference was how much control the prompt gave us. The vague prompt got a formal template with a subject line and blanks for the due date, because ChatGPT had no idea how late the payment was or whether this was the first reminder. The five-part prompt got exactly what it asked for: no subject line, about 60 words, a friendly tone, and a mention of the earlier reminder and the 14 days. When we ran it again on November 7, signed in, ChatGPT’s email came to 97 words; answers vary from run to run.

- No subject line. The vague prompt got one; this prompt said not to include it, and ChatGPT left it out.
- It used the context. “My previous reminder” and “14 days overdue” came straight from the prompt. The vague version could only say the invoice “appears to still be outstanding.”
- Placeholders are still there. ChatGPT can’t know your client’s name or invoice number. Put them in the prompt and the email is ready to send.
Does the same prompt work in every chatbot? We tried five
On November 7, 2026 we sent both prompts, the vague one and the five-part one, to ChatGPT, Claude, Gemini, Grok and Meta’s Muse. Each prompt went into a fresh temporary or private chat in our own accounts, so earlier chats couldn’t shape the answers. Muse has no such mode, so it got a new side chat, which still shares what Muse remembers about the account. We used the model each app had selected: ChatGPT’s Instant setting, Claude Haiku 5.5, Gemini 3.1 Pro (the default on our Google Workspace account) and Grok’s Fast mode (Muse doesn’t name its model). Then we checked every answer against what the prompt asked for.
| Assistant | Vague prompt: added a subject line | Five-part prompt: words (limit 120) | Said 14 days | Said it was the second reminder | Left out the subject line |
|---|---|---|---|---|---|
| ChatGPT | Yes | 97 | Yes | Yes (“following up again”) | Yes |
| Claude | Yes | 141 | Yes | No | Yes |
| Gemini | Yes | 100 | Yes | Yes | Yes |
| Grok | Yes | 125 | Yes | Yes | Yes |
| Muse | Yes | 78 | Yes | Yes | Yes |
The vague prompt got much the same answer everywhere: a subject line and blanks to fill in for the due date, because no app knew how late the payment was. ChatGPT and Claude offered to tailor it, and Grok added a firmer version for a second reminder. Muse filled in the sign-off with the account holder’s first name, which it took from our account.
The five-part prompt fixed most of that in every app. All five left out the subject line, mentioned the 14 days and stayed away from late fees and threats. Two missed the word limit: Claude’s email ran to 141 words and Grok’s to 125. Claude also wrote five paragraphs instead of four and never said this was the second reminder. Muse stayed short at 78 words but used a client name, “Sarah,” that wasn’t in our prompt, and left no blank for the invoice number.

What this means for youA clear prompt gets you most of the way in any of these apps, but the AI may still break one of your rules. Check the answer against what you asked for: if it runs long, ask for it shorter (fix-it phrase 1 below), and if it adds a detail you never gave it, such as a name, fix that before you send anything.
What are the five most useful prompt patterns?
1. Few-shot (showing examples)
Give the AI two or three examples of what you want, then ask for more. Anthropic calls examples “one of the most reliable ways to steer Claude’s output format, tone, and structure” (Anthropic), and Google’s guide says “We recommend to always include few-shot examples in your prompts” (Google).
Here are 3 LinkedIn post openings I’ve written that did well:
1. [example]
2. [example]
3. [example]
Write 10 more in the same voice, for a product launch next week.
2. Chain-of-thought (think step by step)
For math, logic or a decision with several steps, ask the AI to work through it before answering. Many chatbots now have a built-in thinking or reasoning mode that does this for you; for everyday chat, one line still helps:
Work through this step by step before giving your final answer.
3. Role-play (specific expert persona)
“Act as a senior copy editor reviewing this for clarity” gets a different answer than “edit this.” The details matter: “senior,” “10 years of experience” and “writing for beginners” each change the vocabulary, depth and tone. A persona mostly changes the style. It won’t make the facts more reliable: a Wharton study of expert personas found they “generally did not improve accuracy” (Wharton).
Act as a senior copy editor with 10 years of experience editing for beginners. Review the text below for clarity and list your top 5 changes, most important first.
[paste your text]
4. Constraint stacking
Stack several concrete limits, such as length, reading level and words to avoid, to narrow the answer to what you need.
Explain how a credit score works. Bullet list. Under 200 words. No jargon. Write it for a 12-year-old. Don’t use the words “leverage” or “unlock.”
5. Iteration (treat first answer as a draft)
The first answer is rarely the best one. Push back: ask for it shorter, simpler, less formal, or ask the AI to critique its own answer. The next section has ten phrases for it.
What 10 phrases should you keep ready for iteration?
When the first answer isn’t quite right, type one of these next:
Make it shorter. Cut it to half the length.
Rewrite this for a complete beginner.
Show me, don’t tell me: give three concrete examples.
Critique your own answer, then revise it.
What did you miss or oversimplify?
Give me the opposite view too.
Rewrite in plain English. No jargon.
Add a one-sentence summary at the top.
Format this as a table.
What would change if the audience were [new audience]?
What mistakes do beginners make with prompts?
- Asking too vaguely. “Write something about AI” gets generic copy. Add the reader, the format and the length.
- Burying the question. When you paste a long document, put your question after it, not before. Google’s guide says to supply the context first and “place your specific instructions or questions at the very end of the prompt” (Google).
- Asking for several unrelated things at once. “Edit this email, brainstorm 10 product names and summarize this PDF” splits the AI’s attention. Do one task per prompt and feed the result of one into the next. Anthropic calls this “prompt chaining,” though it notes that Claude now “handles most multistep reasoning internally” (Anthropic).
- Trusting it on recent facts. Chatbots can state dates, numbers and sources confidently and still be wrong. For anything current, use a tool that searches the web and shows its sources, such as Perplexity or ChatGPT and Gemini with search turned on, and check the sources.
- Not saying who it’s for or what shape it should take. “Write a guide” gives the AI nothing to aim at. “Write a 600-word guide for a 50-year-old who has never used AI, as numbered steps” already tells the AI what the answer looks like.
What are the tool-specific tips for 2026?
The five-part formula works everywhere. Once you use AI daily, these small differences help:
- Claude (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5; Anthropic): structure helps. Wrap different parts of a long prompt in tags such as <instructions>, <context> and <example>; Anthropic says “XML tags help Claude parse complex prompts unambiguously” (Anthropic). Say what to do rather than what not to do, and explain why a rule matters. Anthropic says its newer models are less wordy, though it notes that Claude Opus 5, an earlier model, gives longer answers by default (Anthropic). Whichever model you use, say how long you want the answer. In our test, Claude Haiku 5.5 went over a 120-word limit.
- ChatGPT (GPT-6 family; the free plan uses GPT-6 Luna): Role + Task + Format works well, and headings (lines starting with #) or tags can separate sections of a long prompt (OpenAI). OpenAI says its top model, GPT-6 Astra (on the paid Plus and Pro plans; OpenAI pricing), leans toward lists, tables and markdown and asks clarifying questions more often, so answer its question or say “make reasonable assumptions,” and ask for plain paragraphs if you want them (OpenAI).
- Gemini: include examples whenever you can, and put your question after any long document you paste (Google). Google’s guide says Gemini 3 models answer directly by default, and its newest app model, Gemini 4 Argon, answers “more directly with less hedging” (Google). If you want more detail, ask for it.
- Perplexity: phrase prompts as research questions. If you want a list, say how long it should be; Perplexity’s own guide notes that without a cap it picks an arbitrary length (Perplexity). Our tip: add a time frame (“as of this month” or “since January”) when you need recent results.
- Grok (made by SpaceXSI, the company formerly called xAI): put the task, the limits and the format up front. The Grok developer guide says its reasoning models “reason before they answer, so state the task, constraints, and output format directly” (xAI). For a long PDF, tell Grok which page numbers to read (Grok FAQ).
- Muse (Meta’s personal AI agent): describe the end goal and let it plan the steps. Meta says Muse is “quite good at breaking down goals and giving you actionable plans to reach them,” and you can send several tasks without waiting for each answer (Meta). In our test it added a client name we never gave it, so read its drafts before anything goes out.
What advanced patterns are worth knowing about?
- Standing instructions. Rules you want in every chat (“Use plain English. Define any jargon. Keep answers short unless I ask for more.”) can be saved once instead of typed each time. ChatGPT calls this custom instructions and Claude has Instructions and Projects. Grok and Gemini have a related feature, Skills, which saves a repeat job such as “summarize this meeting” rather than a rule for every chat (xAI). On personal Google accounts, Gemini’s Skills take over from Gems on November 17, 2026; work and school accounts follow in 2027 (Google). Muse keeps memory files about you that you can read and edit.
- Several messages instead of one huge prompt. For a big job, build it up across a conversation: outline first, then each section, then a final pass.
- Self-critique. Fix-it phrase 4 works as a habit too: ask for a critique and revision before you start editing yourself.
- Show, don’t tell. Ask for examples, not descriptions. “Show me three sample subject lines” beats “what makes a good subject line?” because you can use the answer right away.
Use plain English. Define any jargon the first time you use it. Keep answers under 200 words unless I ask for more. If you’re not sure about a fact, say so.
How do you practice prompting without feeling silly?
- Use AI for things already on your list. The email you need to send, a question you’ve been meaning to ask, a decision you’re weighing. You know what a good answer looks like for your own task, so you learn faster.
- Ask for changes at least twice on anything important. The first answer is the draft; the second and third are usually much better.
- Keep a prompt file. Every time a prompt works well, save it in a note. Over time you build a set for meeting summaries, project plans, weekly reviews and emails that you reuse with small changes. Our newsletter’s prompt-writing guide has a shorter three-part version (who, what, how) that fits at the top of that file.
Frequently asked questions
How long should a prompt be?
Long enough to cover role, task, context, format and constraints, and no longer. Two lines is often enough for a simple job; a complex, multi-step task can need a couple of hundred words.
Are there magic words that make prompts work better?
Not really. Researchers at Wharton found that offering a chatbot a cash tip or threatening it has “no significant effect” (Wharton), and that telling a model that already has a thinking mode to go step by step gives “only marginal, if any, gains” (Wharton). A few phrases are still handy: “show me three examples” when you want something you can use, and “critique your own answer and revise it” for a better second draft. None of them replace saying clearly what you want.
Should I use the same prompt structure for every AI?
Mostly, yes. The five-part formula works on every major chatbot. The tool-specific tips are small adjustments on top. Start with the formula and add the quirks later.
Do I need to memorize prompt-engineering rules?
No. Save 10 to 20 prompts that worked as templates. When a new job is like an old one, copy the closest template and change two or three details.
What if the AI keeps misunderstanding what I want?
Try three things. First, show an example of a good answer (“Here’s one I wrote before: …”). Second, ask it to repeat the task back before answering (“Before you answer, tell me what you understand the task to be”). Third, if it still misses, start a new chat with a clearer first prompt; once a conversation drifts, it is easier to start over than to steer it back.
Made for regular people, by a real person
One AI prompt or workflow a day, in plain English, that you can use the same morning. Curated and edited by a human.
Free forever. Unsubscribe anytime.
You may also like
- How to Use AI: A Step-by-Step Guide
- ChatGPT vs Claude vs Gemini
- Best AI Tools for Beginners
- AI for Dummies
- AI Glossary
Sources
- Anthropic, prompt engineering overview and prompting best practices.
- Google, Prompt design strategies.
- Anthropic, models overview; OpenAI, Using GPT-6; Perplexity, prompt guide.
- OpenAI, prompt engineering guide.
- DAIR.AI, Prompt Engineering Guide.
- Grokipedia, Prompt engineering.
- Wharton Generative AI Labs, Prompting Science Reports 2, 3 and 4 (2025).
- xAI, best practices and Grok Skills; Meta, How we designed Muse; Google, Gems to Skills and Gemini 4 Argon.
- Our tests: ChatGPT, free and logged out, October 30, 2026; ChatGPT, Claude, Gemini, Grok and Muse in our own accounts (temporary or private chats; a new side chat for Muse), November 7, 2026. Screenshots by Beginners in AI.
Last reviewed: November 2026. Prompting advice changes with each major model release, so check the vendor guides above before relying on a specific tactic.

