Prompt engineering for marketers: a beginner’s guide with examples

The pattern: role, context, task, format, constraints

Most weak AI output comes from prompts that only state the task: “write an Instagram caption for my gym”. The tool has to guess the audience, the offer, the tone, the length and what to avoid, so it guesses the most generic version of each. The fix is to stop making it guess.

We teach a five-part structure. It is not the only one that works, but it is easy to remember and it covers what most marketing tasks need. You will not always need every part, and the order matters less than making sure nothing important is missing.

• Role — Who the AI should act as, such as a performance marketer, a reel scriptwriter or an SEO editor. This sets the vocabulary and priorities.
• Context — The business, the audience, the offer, the location and anything the AI cannot know on its own.
• Task — Exactly what to produce, in one clear sentence.
• Format — Length, structure, number of options, headings, tables or bullet points.
• Constraints — What to avoid: words, claims, tone, emojis, prices, anything off-brand or non-compliant.

Worked example: an ad caption

Here is a prompt for a Meta ad caption for a local business. Notice how much of it is context. That is deliberate: the more the tool knows about the business and the customer, the less it falls back on clichés.

Prompt: “You are a performance marketer who writes Meta ad copy for local service businesses in India. Context: the client is a physiotherapy clinic in Rohtak. Most patients are working adults with back and neck pain from desk jobs. The offer is a first assessment session where the physiotherapist explains the cause of the pain and a treatment plan. The ad will run as a lead form campaign. Task: write three ad caption options. Format: each option should have a first line under 12 words that names the problem, two or three short lines of body copy, and a clear call to action. Constraints: no medical promises or claims of cure, no prices, no emojis, no exclamation marks, plain English that a busy reader understands in five seconds.”

Why it works: the AI knows the audience, the offer, the campaign type and the compliance limits. You will still edit the output, but you will be choosing between three reasonable drafts instead of rewriting one generic one.

Worked example: a reel script

Reel scripts need timing and visual direction, not just words. If you only ask for a script, you usually get a paragraph that reads well and films badly.

Prompt: “You are a short-form video scriptwriter for small business Instagram accounts. Context: the business is a home bakery that takes custom cake orders. The audience is parents planning children’s birthday parties. The goal of this reel is to show how a custom order works from enquiry to delivery, so viewers feel confident ordering. Task: write a script for a reel of about 30 seconds. Format: a table with three columns: time range, what is shown on screen, and what is said or shown as text. Start with a hook in the first two seconds that a parent would stop scrolling for. End with one clear call to action to send a message. Constraints: the owner will film on a phone in her own kitchen, so no scenes that need special equipment or other locations. Keep spoken lines short enough to say naturally. No prices, no emojis.”

The constraint about filming conditions is the part beginners leave out, and it is the part that decides whether the script gets shot at all.

Worked example: a blog brief

For SEO content, we use AI to build the brief rather than the finished article. A good brief makes the writing better whether a person or a tool does it, and it keeps the human in charge of facts and experience.

Prompt: “You are an SEO content editor. Context: the website belongs to a dental clinic in a mid-sized Indian city. We want to write an article for people searching for information about getting braces as an adult. The clinic’s dentists can supply real experience and answers. Task: create a content brief for this article. Format: suggested title, a one-paragraph summary of search intent, an outline with H2 and H3 headings, a list of questions the article must answer, points where a dentist’s own experience should be added, and three internal pages the article could link to. Constraints: do not invent statistics, costs or treatment durations. Mark any point that needs a fact checked by the clinic. Keep the outline focused on what an adult patient actually wants to know.”

The line about marking facts to check is important. It turns the AI’s tendency to fill gaps into a visible to-do list instead of a hidden error.

Worked example: a monthly content plan

Content plans fail when they are a list of nice ideas with no link to the business goal. Give the AI the goal, the capacity and the formats, and ask for a plan you can actually execute.

Prompt: “You are a social media strategist for local businesses. Context: the client is a clothing boutique selling ethnic wear for women. The goal for next month is more in-store visits during the festive season. The owner can create three posts a week and two reels a week, filmed on a phone. Task: create a four-week Instagram content plan. Format: a table with columns for week, day, format (post, carousel or reel), topic, hook line and call to action. Include a mix of product showcases, styling ideas, behind-the-scenes and customer-question posts. Constraints: do not plan anything that needs a professional shoot or paid models. Do not include prices or discount percentages, because the owner will decide offers separately. Keep hook lines under ten words.”

Once you have the plan, ask follow-up questions in the same chat: “Rewrite week two to focus more on reels” or “Which three posts are most likely to bring people into the store, and why?” Iterating in the same conversation is often more useful than writing a perfect first prompt.

How to check and edit AI outputs

Never publish AI output unread. We review every AI draft on live client work against a short checklist, and students follow the same habit from their first week. Most problems are easy to fix once you look for them.

Read for facts first, then for fit. A caption can be grammatically perfect and still wrong for the brand, too long for the format or quietly promising something the business cannot deliver. If the same problem repeats across outputs, fix the prompt rather than editing each draft by hand.

• Facts — Are any numbers, claims, dates or features invented? Check every one against the client’s real information.
• Compliance — Does it make claims that ad platforms or the law would not allow, such as health cures or promised results?
• Voice — Does it sound like this business, or like every other AI caption? Cut generic phrases.
• Format — Does it fit the platform: length, hook position, call to action, character limits?
• Usefulness — Would a real customer understand the offer and know what to do next?

Build your own prompt templates

Once a prompt works, save it as a template with blanks for the parts that change: business, audience, offer, goal, constraints. Over time you build a small library for the tasks you repeat most, which is where the real time saving comes from.

Keep templates in a plain document, grouped by task, and note what each one is bad at. Update them when a tool changes or when you notice a repeated weakness in the output. A template is a starting point, not a script to follow blindly; the context section should be rewritten properly for every new client.