A prompt is a brief. The same principles that make a good creative brief, such as clear goals, audience and constraints, make a good prompt. The difference is that AI tools take instructions literally and cannot ask follow-up questions unless you invite them to. The Academy's glossary explains hook and brand voice.
The parts of a good prompt
| Part | Example |
|---|---|
| Role | You are a social media copywriter for a UK independent bakery |
| Goal | Write an Instagram caption that encourages pre-orders |
| Audience | Local families and office workers within five miles |
| Voice | Warm, playful, never pushy, British spelling |
| Content | New seasonal pastry, available Friday and Saturday only |
| Format | Under 80 words, one emoji at most, end with a question |
| Examples | Here are three captions we liked |
| Constraints | Avoid the words delicious and mouth-watering |
Templates
Caption: You are writing for [brand], whose audience is [audience]. Our voice is [voice]. Write three Instagram captions about [topic] that achieve [goal]. Keep each under [length]. Here are examples of our past captions: [examples].
Hooks: Give me ten opening lines for a short video about [topic] aimed at [audience]. Each should create curiosity or name a problem in under ten words.
Content ideas: Suggest twenty post ideas for [platform] within these pillars: [pillars]. For each, give the format and a one-line angle.
Repurposing: Turn this blog post into a LinkedIn post of about 150 words, a five-slide carousel outline and three short video hooks: [text].
Reply drafts: Draft a friendly, helpful reply to this customer comment in our voice. Do not promise anything we have not stated. Comment: [comment].
Iterate rather than restart
Treat the first draft as a starting point. Ask for changes: make it shorter, more conversational, add a question, remove jargon, try a different opening. Iterating in the same conversation keeps context.
Ask the tool to ask you
Add: Before writing, ask me any questions you need answered. This surfaces missing information and improves drafts.
Give examples
Examples are the fastest way to teach voice. Paste three posts that performed well and explain what you liked. Avoid examples you do not want copied.
Set constraints
Constraints improve output: word limits, banned phrases, reading level, British spelling, no hashtags or a specific number of emojis. Platform constraints matter too. LinkedIn posts allow 3,000 characters (first ~210 show before 'see more')✓, so ask for the key point in the opening line.
Avoiding bland output
Bland output comes from bland prompts. Add specifics: real customer quotes, numbers, local details, opinions and stories. Ask for a contrarian angle or a surprising fact as the opening. Then edit to add your own perspective. The Academy's guide to keeping AI content on brand has more techniques.
Prompts for analysis
AI can summarise comments, group themes and suggest insights. Example: Here are 200 comments from our last campaign. Group them into themes, estimate how common each is and give three example comments per theme. Check results against the data.
What not to put in prompts
Do not paste personal data about customers, confidential business information or anything covered by agreements into public AI tools. Use approved enterprise tools for sensitive work. The Academy's guide to an AI policy for marketing teams explains how to set rules.
Building a prompt library
Save prompts that work in a shared document with notes on when to use them. Over time, the library becomes a team asset. The Academy's guide to the prompt library for marketers shows how to organise one.
Checking outputs
Review every output for accuracy, tone, claims and originality. Verify any facts, statistics or platform specifications. AI can invent numbers and features that sound plausible.
Chaining prompts
Complex tasks work better in steps. First ask for ten angles on a topic, then pick one and ask for an outline, then ask for the caption, then ask for three hook variations. Each step builds on the last, and you can steer the direction at every stage rather than hoping one long prompt gets everything right.
Using your own data
Results improve when prompts include your real data: top-performing posts from analytics, common customer questions, review quotes and product details. Paste the relevant extract into the prompt and ask the tool to use it. Keep personal data out unless you are using an approved tool.
Prompting for different platforms
The same idea needs different treatment on each platform, and prompts should say so. Ask for a short, punchy hook and on-screen text for TikTok, a scannable caption with a question for Instagram, a professional insight with a clear first line for LinkedIn and a keyword-rich title and description for Pinterest. Naming the platform and its norms in the prompt saves heavy editing later.
A worked example
A gym's first prompt, write a post about our new spin class, produces a generic caption. The revised prompt adds the audience (beginners nervous about spin), voice (encouraging and down to earth), goal (book a free taster), length (under 70 words) and two example posts. The new draft needs only light edits and leads to more taster bookings than previous captions.
Common mistakes
- One-line prompts with no context.
- No examples of brand voice.
- Accepting the first draft.
- Pasting personal data into public tools.
- Not checking facts in outputs.
Learning to prompt well
The Academy's AI tools module in the second level covers prompting for drafting, editing and analysis, and the copywriting craft module helps you judge and improve drafts. The free assessment shows where you would start.
