How to Write AI Prompts That Work — Beginner Guide

6 principles with before/after examples. Works on ChatGPT, Claude, and Gemini — no jargon, practical from the first read.

Getting poor output from AI is almost always a prompt problem, not an AI capability problem. The tools are capable of producing genuinely useful results — but they need specific input to do so. This guide covers the six principles that separate effective AI users from people who tried it once and concluded “AI is not that useful.”

Every principle includes a before/after example so you can see exactly what changes and why it works. If you are not yet sure which AI tool to use, our ChatGPT vs Claude vs Gemini guide compares all three free tiers so you can pick the right one before you start.

The core insight: AI is not a search engine that reads your mind. It is a collaborative tool that produces output proportional to the quality of your input. Treat every prompt like a brief you are giving to a smart but uninformed assistant.

Principle 1 — Be specific about what you want

The most common mistake — and the easiest fix

The biggest reason AI gives generic, unhelpful output is generic, unhelpful prompts. “Write me an email” produces a mediocre email. “Write a 3-paragraph follow-up email to a potential client I met at a conference last week. The client runs a mid-sized architecture firm. I want to remind them of our conversation about interior project photography, express genuine interest in working with them, and suggest a 15-minute call next week. Tone: professional but warm, not overly formal.” produces something you might actually send.

Every detail you add about the situation, audience, tone, format, and length makes the output more usable.

Key takeaway: Before sending a prompt, ask yourself: “Would a freelancer who just started working for me understand what I need from this?” If the answer is no, add more detail.

---

Principle 2 — Give context — who you are, who this is for

AI needs to understand the situation, not just the task

AI has no background on you, your business, your audience, or your situation — unless you provide it. Adding context (who you are, what this is for, who will read it) changes the output dramatically.

“Write a social media bio” gives you something generic. “I am a freelance video editor who focuses on wedding films. My target clients are engaged couples in their late 20s–early 30s in the UK. I want a bio for my Instagram profile that sounds like a real person, not a corporate listing. Emphasize emotion and storytelling, not technical skills.” gets you something you can actually use.

Key takeaway: Create a 3-sentence “context block” about yourself that you paste into any AI tool before asking business-related questions. Update it once and reuse it forever.

---

Principle 3 — Tell AI what format you want

Output format matters as much as content

AI defaults to prose paragraphs unless you specify otherwise. If you want bullet points, numbered steps, a table, a script, a list of options, or a specific structure — say so explicitly.

“Give me 5 ideas for a newsletter subject line” produces 5 options. “Give me 10 newsletter subject line options, one per line, no explanations” produces exactly what you can copy directly into your email tool.

“Write a product description” produces a paragraph. “Write a product description in this format: first a one-line tagline, then 3 bullet point features, then a call-to-action sentence” produces something ready to paste into your product page.

Key takeaway: If you are going to use AI output in a specific tool or context (a spreadsheet, a Shopify product page, an email template), describe that context so AI structures the output to fit.

---

Principle 4 — Assign AI a role or expertise

Telling AI who to be changes how it responds

Starting a prompt with “Act as a [role]” shifts how the AI approaches the task. “Act as an experienced copywriter specializing in email marketing” produces different output than a generic request. “Act as a critical editor who finds weak arguments” makes the AI more aggressive in pointing out problems. “Act as a potential customer who is skeptical about this product” gets you objections to address.

The role does not have to be a job title — it can be a perspective or expertise level. “Explain this as if I have no technical background” and “Explain this assuming I have 10 years of marketing experience” produce very different explanations.

Key takeaway: The most useful roles for everyday AI use: experienced editor, skeptical customer, expert in [your field], beginner who needs simple language, and devil’s advocate.

---

Principle 5 — Treat the first response as a draft, not a final answer

One prompt is rarely enough — iteration is the workflow

The most effective AI users treat every first response as a starting point. After getting an initial output, follow up with specific instructions:

• “This is good but the tone is too formal. Rewrite it to sound more conversational.”
• “The first two points are what I want, but the third is off topic — replace it with something about [X].”
• “This is too long. Cut it to half the length without losing the key message.”
• “Give me 3 alternative versions of paragraph 2 — more aggressive, more empathetic, and more data-focused.”

Each follow-up gets you closer to exactly what you need without starting over.

Key takeaway: Save effective AI conversation threads. When a conversation produces exactly what you need, save the prompt sequence as a template for similar future tasks.

---

Principle 6 — Always verify facts and check for hallucinations

AI is confident even when it is wrong

AI tools state incorrect information with the same confident tone as correct information. Statistics, dates, prices, names, URLs, study citations — all of these can be made up. Before using any specific factual claim from AI output in something public-facing, verify it independently.

This does not mean distrusting everything — AI is excellent at tasks where facts are not the point (rewriting, brainstorming, summarizing your own documents, structuring ideas). It means applying appropriate skepticism to any claim that could be wrong.

Use Perplexity AI if you want web-sourced answers with citations that you can verify.

Key takeaway: Red flags for hallucination: very specific numbers (84.3% of users…), citations of research studies, URLs, and any claim that sounds too perfectly convenient for your argument. These are the most common areas where AI fabricates.

---

The 6-principle checklist

1. Be specific about what you want
2. Give context — who you are, who this is for
3. Tell AI what format you want
4. Assign AI a role or expertise
5. Treat the first response as a draft, not a final answer
6. Always verify facts and check for hallucinations

---

Frequently asked questions

Do these prompting principles work the same on ChatGPT, Claude, and Gemini?
Yes — all six principles apply equally to all three tools. Each model has different strengths (Claude tends to follow complex formatting instructions more precisely; ChatGPT is more flexible with creative tasks; Gemini integrates Google Search natively), but the fundamentals of a good prompt are the same across all of them.

How long should an AI prompt be?
As long as it needs to be. There is no word limit that makes a prompt “too long.” A detailed 150-word prompt almost always produces better output than a 10-word one. The only thing to avoid is including irrelevant information — extra context about your situation is helpful; off-topic background detail wastes tokens and can confuse the output.

What should I do when AI keeps giving bad output even after multiple follow-ups?
Start a new conversation. Long threads can confuse some models as context accumulates. In the new thread, describe both what you want and what you do NOT want — “don’t make it sound formal, don’t use bullet points, don’t include a summary at the end.” Negative instructions are often the fastest fix when positive instructions have not worked.

Good prompts work across all tasks. For a practical starting point, see the 10 ChatGPT prompts that save freelancers hours — each one follows the principles in this guide, so you can see them applied to real-world scenarios.