AI prompts for senior marketers (the ones worth keeping)

Most prompt advice for marketers is written for people who are new to AI and need somewhere to start. Lists of 50 prompts, none of them tested rigorously, all of them generic enough to be broadly applicable and therefore broadly mediocre.

This isn't that.

What follows is a set of prompts built for strategic marketing work — the kind of tasks where the quality of the output actually matters. Each one comes with context on why it's structured the way it is, because a prompt you understand is one you can adapt. A prompt you copy-paste is one you'll abandon when the context changes.

Why most marketing prompts produce mediocre output

Before the prompts: the structural problems worth fixing.

Vague instructions get vague results. "Write me an email campaign for a B2B software company" produces the exact kind of output that description deserves — technically correct, completely generic. The AI has no way to know which B2B software company, which audience, what problem is being solved, what stage of the funnel this is for, or what success looks like. It fills those gaps with averages. The output reflects that.

No role, no POV. Left to its own defaults, an AI writes like a careful, neutral professional trying not to offend anyone. That's the opposite of what good marketing copy sounds like. Assigning a specific role — including constraints on what that role wouldn't say — shifts the output toward a perspective.

Too much context is as bad as too little. A prompt stuffed with everything you could think of to include dilutes the model's attention and produces outputs that hedge across all of it. A focused 200-word prompt with a clear objective usually outperforms a 1,000-word prompt that covers every contingency.

No instructions on what to exclude. The things you don't want in the output are often more useful constraints than the things you do. "No clichés, no generic value propositions, no use of the word 'innovative'" forces a narrowing that produces sharper work.

The prompts below are structured around these principles. They're longer than most prompt templates you'll find, but that's because the specificity is doing the work.

The prompts

Audience analysis

The job: Rapid characterisation of a target audience — what they care about, what they're frustrated by, how they make decisions, and what language they use.

You are a senior qualitative researcher who specialises in B2B buyer psychology.

I'm building a marketing strategy for [company/product]. The primary audience is [describe role, seniority, industry, company size].

For this audience, give me:

1. The 3 biggest professional frustrations they're unlikely to say out loud in a sales conversation but that drive their buying behaviour

2. The internal objections they'd raise before approving a purchase like this

3. The language they'd use to describe their problem to a colleague — not the language a vendor would use

4. What success looks like to them 6 months after buying

Be specific to the audience described. Do not give me generic B2B buyer behaviour. If a point applies to every B2B buyer, it's not useful here.

What this does: the instruction to exclude generic points forces the AI toward the specific. The 4 questions are structured to surface the things a buyer wouldn't say explicitly — which is exactly what useful audience intelligence captures.

Positioning analysis

The job: Quick competitive positioning read before writing any client-facing strategy.

I'm going to paste the homepage copy and one key landing page from [competitor name]. Read it and tell me:

1. What positioning statement they're implicitly making — who they're for, what problem they solve, why they're different

2. The claims they make that are verifiable vs. claims that are vague or unprovable

3. What they're conspicuously not saying — gaps in their positioning that could be exploited

4. The audience they appear to be optimising for based on the language and specificity of their copy

[Paste competitor copy here]

Do not summarise the copy back to me. I can read it. Give me the analysis.

The final line matters. Without it, AI reliably opens with a summary of what you just gave it. You don't need the summary. The instruction to skip it saves you the edit.

Creative brief writing

The job: Turning a strategy direction into a brief a designer, copywriter, or agency can actually use.

You are a creative director who writes briefs that get used rather than ignored.

Write a creative brief for the following project:

- Objective: [what this piece needs to accomplish, as specifically as possible]

- Audience: [who it's for, what they know, what they're trying to do]

- Key message: [the single thing the audience should believe, feel, or do after experiencing this]

- Tone: [2-3 adjectives, then describe what those adjectives mean in practice for this audience]

- Constraints: [format, length, compliance considerations, brand restrictions]

- What success looks like: [how you'd evaluate whether this worked]

The brief should be one page. Do not include a background section, a market overview, or a company description. Start with the objective.

The instruction to omit background, market overview, and company description cuts the 30% of every brief that nobody reads. The "start with the objective" instruction forces the brief to lead with the work's reason for existing.

Campaign concept generation

The job: Fast ideation on campaign concepts before a brainstorm or pitch.

I'm developing campaign concepts for [client/product]. Background:

- What we're selling: [describe the product/service and its primary differentiator]

- Who we're talking to: [audience description]

- What we want them to do: [specific conversion action]

- What we're not doing: [constraints — no humour, no competitor comparisons, no reliance on brand recognition the audience doesn't have yet]

Generate 5 distinct campaign concepts. Each concept should include:

- A one-sentence campaign idea (not a tagline — what the campaign does)

- The central tension or insight it's built on

- Why this audience would respond to it

Make the 5 concepts genuinely different from each other in their strategic logic — not 5 executions of the same idea. At least one should feel uncomfortable to present.

"At least one should feel uncomfortable to present" is the most useful instruction in this prompt. AI defaults to concepts a committee would approve. This instruction forces at least one option that has an edge — which is usually the one worth developing.

Messaging hierarchy

The job: Building a structured message architecture before writing any copy.

You are a brand strategist. I need a messaging hierarchy for [company/product].

Context:

- Primary audience: [description]

- What we actually do: [honest description, not marketing language]

- What our best customers say about us in their own words: [paste quotes or describe if you don't have them]

- What we want to be known for: [the 1-2 things you want associated with the brand]

Build a messaging hierarchy with:

1. Positioning statement (internal use — clear and honest, not designed to impress)

2. Primary message (the one thing to say to the primary audience)

3. Three supporting messages (evidence or elaboration for the primary message)

4. Proof points for each supporting message

Avoid: superlatives, vague differentiators ("innovative," "trusted," "best-in-class"), claims that 10 competitors could make equally. If a message requires those words to work, the message isn't strong enough yet.

The final constraint — avoiding claims that 10 competitors could make — is the most productive editing instruction in brand work. It's easier to apply as a prompt constraint than as a revision note after the fact.

Post-campaign analysis

The job: Turning a data export into a clear read on what worked and what to do next.

I'm going to paste campaign performance data from [channel/campaign]. Analyse it and give me:

1. What worked — specific elements (audience, creative, format, timing) that outperformed the rest, with numbers

2. What didn't — specific underperformers and what you'd hypothesise as the cause

3. What to test next — 2-3 specific A/B tests that follow logically from the data, stated as testable hypotheses

4. The one thing to change immediately if we ran this campaign again tomorrow

[Paste data here]

Do not restate the numbers back to me in narrative form. Interpret them.

"Do not restate the numbers back to me in narrative form" is doing a lot of work. Without it, you get a paragraph that says "The Facebook campaign generated 1,247 clicks at a CTR of 2.3%, compared to the LinkedIn campaign which generated..." You have the spreadsheet. You don't need it read to you.

A note on adapting these

These prompts are frameworks, not scripts. The parts to customise are obvious: the audience description, the product context, the constraints. The parts to leave intact are the structural instructions — the things to exclude, the specificity requirements, the instruction to skip the summary.

The quality of what comes back is a function of the quality of the context you provide. A prompt that asks Claude to write a positioning statement without telling it who the product is for, what it actually does, and what competitors are saying will produce a positioning statement that could apply to anyone. Fill in the specifics. That's where the work is.

Save the prompts that produce good output. Iterate on them. The ones that keep working become templates. The ones that don't tell you something about what context was missing.

The HEM prompt library is updated as new frameworks get tested. Subscribe to get additions as they're built — practical prompts for the kind of strategy work senior marketers actually do.