There's a specific flavor of bad AI marketing copy that you've definitely read. The relentless enthusiasm. The stacked adjectives. The vague benefit claims. The call to action that's somehow both urgent and meaningless.
"Unlock your potential with our revolutionary solution and take your business to the next level today!"
Nobody talks like that. Nobody convinces anyone with that. And yet that's what you get if you walk up to an AI and say "write me marketing copy for my product."
The failure here isn't the AI. It's that "write me marketing copy" is one of the least constrained requests you can give a language model. The model defaults to the average of everything it's seen, and average marketing is terrible.
Here's how to break out of that.
Step One: Before You Write Anything, Brief It Properly
In a real agency or in-house team, someone writes a creative brief before copy gets written. The brief tells the copywriter who the audience is, what they care about, what the product does that's relevant to those cares, the tone, and the specific job the copy needs to do.
AI needs a brief too. Except most people skip it.
Here's a template:
I need help writing [ad copy / email / landing page / social post / etc.] for [product/service].
Product: [What it is in one sentence]
Audience: [Specific description — not "businesses" but "B2B sales reps at companies with 20–200 employees who spend 3+ hours per day in their CRM"]
Core benefit: [The one thing the audience cares about most — what does this actually do for them?]
Proof: [What makes this believable — a stat, a testimonial, a mechanism, a credential]
Tone: [3 words that describe the brand voice. e.g., "Direct, irreverent, confident" or "Warm, expert, reassuring"]
Objective: [What do you want someone to do or feel after reading this?]
Constraints: [Word count, platform requirements, things to avoid]
One thing NOT to do: don't lead with product features. Lead with the audience's problem or desire.
That last line does a lot of work. Feature-first copy is the default, and it's almost never the most persuasive approach.
Describe the Audience, Not the Customer
This is a distinction most marketers understand in theory but frequently collapse in practice when prompting.
"Our customer is a busy professional" is not a useful audience description. Every product on earth claims this.
Useful audience descriptions are specific about:
- What they're trying to accomplish (the job-to-be-done)
- What's frustrating them about current solutions
- What they're skeptical of (especially relevant for your category)
- How they see themselves
Here's what that looks like in practice:
Generic: "Marketing managers at small businesses."
Useful: "Marketing managers at software startups who are stretched across too many channels and are tired of tools that promise insights but just generate dashboards nobody reads. They're skeptical of AI-powered anything because they've bought three tools that didn't deliver."
When you brief AI with that level of specificity, it can write to the actual person instead of to an abstraction. The copy it produces will be more specific, more credible, and more likely to land.
The Problem With "Benefit-Focused" Copy
Prompting guides often tell you to focus on benefits, not features. This is correct. It's also a trap.
Benefits are still generic if they're not grounded in something specific. "Save time" is a benefit. "Reduce the time your sales team spends on manual data entry from 4 hours per week to 20 minutes" is a benefit that someone will actually believe.
When you prompt for copy, push for specificity in the benefits:
Write three variations of a hero headline for a product that [feature]. The headline should lead with the benefit, not the feature.
For each headline, also write a one-sentence sub-headline that makes the benefit concrete and believable — a specific number, timeframe, or comparison where possible.
This forces the model to find the tangible version of the benefit rather than staying at the vague "save time and grow your business" level.
Teaching AI Your Brand Voice
Brand voice is one of those things that's easy to recognize and hard to describe. Most brand voice guides use adjectives like "confident," "human," and "approachable" — which means nothing, because no brand describes itself as unconfident, robotic, and off-putting.
The better approach: give the model examples and let it infer.
Here are five examples of copy we've written that feels right for our brand:
[Example 1]
[Example 2]
[Example 3]
[Example 4]
[Example 5]
And here are two examples of copy that feels wrong — too formal, too salesy, or just not us:
[Bad example 1]
[Bad example 2]
Based on these, describe what our brand voice is in your own words. Then write [the thing you need] in that voice.
The positive examples show what you want. The negative examples are equally important — they help the model understand the boundaries.
Prompts for Specific Marketing Use Cases
Ad copy (multiple variants for testing):
Write 5 variations of a Facebook ad for [product].
Target: [audience description]
Core message: [the one thing you want them to take away]
Offer: [if there's a specific promotion or CTA]
For each variation, use a different lead — a question, a bold claim, a pain point, a social proof statement, a "what if" scenario. Keep each under 90 words.
Email subject lines:
Generate 10 subject lines for an email announcing [what you're announcing].
The audience is [description]. They receive a lot of marketing email and have a medium level of familiarity with our brand.
Vary the approach: some should create curiosity, some should be direct and informational, some should lean on FOMO or urgency, one or two should be unconventional.
Do NOT use exclamation points. Do NOT use all caps. Do NOT use emoji unless it genuinely improves the line.
Landing page above the fold:
Write the above-the-fold section of a landing page for [product].
Above the fold = the headline, subheadline, and CTA button text that's visible before a visitor scrolls.
The visitor just clicked an ad that promised [what the ad promised]. This landing page must immediately confirm they're in the right place and give them one clear reason to keep reading.
Don't try to explain everything — just nail the handoff from the ad and give them a reason to scroll.
Case study narrative:
Turn the following information into a 400-word case study narrative. Write it like a story, not a listicle. Lead with the customer's problem. End with the outcome.
Customer: [name and brief description]
Their situation before: [the problem or challenge]
What they tried: [solutions they had tried]
What they did with us: [how they used your product/service]
The outcome: [specific results]
Quote (if available): [direct quote from customer]
Tone: [professional but not stiff — this is a real story, not a press release]
The A/B Testing Mindset
One thing AI is genuinely great for in marketing: generating variations to test.
A good copywriter can write 3 versions of a headline. An AI can write 15 in the time it takes to make coffee. That changes what's possible in a testing roadmap.
Use this structure:
Write 10 versions of [this headline / this CTA / this email subject / this hook] for [product/audience].
Each version should take a meaningfully different strategic approach — not just different words for the same idea. Label each with its underlying strategy (e.g., "social proof," "pain point," "curiosity gap," "bold claim," etc.)
You're not planning to run all 10. You're looking for 3 genuinely different approaches worth testing. Generating 10 gives you better candidates than generating 3.
What AI Is Bad at in Marketing
To be honest about the limits:
Strategy — AI can execute strategies well. It's not good at originating them. "What should our brand stand for?" needs human thinking. "Write copy based on our brand positioning" is where AI shines.
Cultural nuance — Humor, slang, and cultural references are risky territory. AI has read a lot but it doesn't have lived cultural context. Review carefully anything where you're trying to be current or culturally aware.
Truly original creative concepts — AI remixes and synthesizes what it's seen. For a campaign concept that's genuinely new, you still need human creative direction. AI can help you execute it.
Tone-of-voice perfection on the first try — Voice always needs editing. AI gives you material to work from, not a finished product.
The deeper you get into structured prompting for output consistency — making sure campaigns have a coherent voice across touchpoints — the more useful the techniques in the Intermediate Track become. Constrained generation in particular is worth understanding for anyone building brand-consistent content at scale.