4 rounds of edits with an open-ended prompt. One small tweak with a hook-first prompt. Same AI, same campaign, completely different experience. And it comes down entirely to how you write the prompt.
As head of email and retention at FABO, a holistic growth partner for ecommerce brands in lifestyle and apparel, I see this gap quite a bit. Prompting well has become a marketing skill.
These are the 10 practices that separate AI outputs you send from AI outputs you spend an afternoon fixing. They’re tested and proven across lifecycle flows, campaign copy, and retention programs for real clients.
1. Start with the hook, not the prompt
Write the hook before you open a prompt, not after. I write the hook, paste it in, and ask the AI to build the email around it. I write something like: "Here's the hook: '[hook]'. Write a 250-word email that delivers on this exact promise. Don't introduce a new angle."
Why the hook has to come first
The hook is the promise you're making to your reader. It sets the scene for the marketing email. When you put it in the prompt upfront, the AI has a destination, so everything it writes is working toward it. Without it, you're essentially asking AI to write a letter to someone it's never met. You'll spend more time editing than if you'd just written it yourself.
I tested this back to back with a client in lifestyle ecommerce. With my open-ended prompt, I needed 4 rounds of edits. But with my hook-first prompt, we had just one small tweak and we were done. The subject line and body already matched because the AI knew where it was heading from the first sentence.
2. Articulate what your product does + how it should make customers feel
Always tell the AI both what the product does and how it should make the customer feel. Not just the specs, but the meaning behind them: "Here are the product features: [list]. For each one, write the emotional benefit it delivers to [persona]. Then, write a launch email that leads with feeling, not fabric."
Why you need to tell AI, not assume they’ll get it
Left to its own devices, AI writes product descriptions that read like ingredient lists most of the time. That might work for a grocery store, but it doesn't work for a fashion brand. The moment you add an emotional layer to your prompt, the output stops being something people skim and becomes something they respond to.
I work with several fashion and lifestyle brands, and this is consistently where I see the biggest jump in engagement: not from a better subject line or a different send time, but from a brief that forces the copy to lead with the customer's end benefit instead of the product's attributes.
3. Pinpoint where your customer is in the journey
Context is everything, especially when working with lifecycle marketing. Don’t stop at the flow trigger. Tell the AI exactly where your customer is in their journey, and what they're probably feeling right now.
Here’s what it might look like: "Write email two of a win-back series for a customer who bought 3x in 5 months and then went quiet for 90 days. They probably didn't leave angry. Life just happened. Don't apologize. Lead with value. And don’t frame your copy with ‘We miss you.’"
Why your prompts need to be lifecycle-aware
If you just say "Write a win-back email," the AI has no idea who it's talking to. Is it a loyal customer who drifted? A one-time buyer who forgot you existed? The output will be vague because the brief was vague.
When you give AI the full picture (purchase history, time lapsed, likely headspace), the copy actually sounds like it was written for that specific person.
I used this approach to rebuild a win-back flow for a fashion/apparel brand. We sent the first email with no discount, just a sharp, personal re-introduction of value in a plain text format. It converted at 2x their previous win-back rate.
The only thing that changed was how specifically we briefed the AI.
4. Treat AI output as a first draft. Then make it more human-sounding.
I ask at least two follow-up questions before I use anything that came from AI. My questions range from "Is this actually true for this brand, or did you just make it sound plausible?" to "Does this opening sound like a real person or like AI trying to sound like a real person?"
Then I push back, request a rewrite, and add the details only I can add: the specific anecdote, the brand-specific word choice, the line that would never come from a model trained on everyone's content.
Why you need to keep your hands on the wheel
AI is confident by default but often misses the context. And you would be surprised to learn what it hallucinates that has no authenticity behind it. It produces output that looks finished even when it's generic, slightly off-brand, or just plain wrong.
I often see people giving up midway through their prompting because they’re not getting what they ask for. While AI does the heavy lifting, you need to bring judgment and authenticity.
You don’t always need a perfect prompt, but you do need to stay critical throughout the process. Validate the facts. Question the tone. Then make it sound like a human is at the helm, because ultimately a human is.
5. Include a “do not” list
Every prompt should have a list of things the AI is not allowed to do. I keep a running "no” list for each client and paste it in every time. For example: "Do NOT: start with 'just a quick reminder', use the word 'amazing', or lead with urgency. DO: keep it short, be direct, write like we respect the reader's intelligence."
Why constraints do more than instructions
AI defaults to the most common version of whatever you're asking for. In email marketing, that means clichés. We aim to create authentic experiences with the reader and create a relationship, so generic content or angling just doesn’t hit the spot.
Once you name what you don't want, the output quality jumps. It sounds obvious, but the constraints, not the instructions, are where your brand voice actually lives.
After I started including a “no” list in prompts, my editing time dropped by roughly 60%. It was all because I stopped letting AI default to patterns I'd then have to undo.
6. Make sure every team is prompting from the same brand voice
Your email copy might be perfectly on-brand. But if your ad team is prompting AI without the same guardrails, your paid creative could sound like it came from a different company entirely. When multiple teams are using AI in parallel (each writing their own prompts, each getting slightly different outputs), brand voice can fragment quickly.
The fix isn't a style guide PDF that lives in a shared drive nobody opens. It's embedding voice directly into the prompt itself.
Nanxi Fan, director of paid media and CRM at ASSOULINE, a brand that sells remarkable coffee table books and handcrafted library accessories, explains what that looks like for her team: "Giving the model concrete anchors prevents abstract interpretation and keeps outputs consistent, especially when multiple teams are prompting in parallel. In practice, this means embedding clear voice guidelines directly into prompts: specific adjectives, do/don't language, example phrases, and even 'anti-examples' to prevent drift."
Giving the model concrete anchors prevents abstract interpretation and keeps outputs consistent.
The anti-example piece is worth dwelling on. Similar to a “no” list, telling AI what your brand doesn't sound like is just as useful as telling it what it does. Abstract direction like "write in a luxury tone" leaves too much room for interpretation. Concrete anchors close that gap.
Why inconsistent prompts can erode brand equity
When prompts are grounded in a shared voice, outputs become more usable from the start and require less editing. Fan has seen the downstream effect of skipping this step: "Messaging tends to fragment (luxury in one channel, casual in another), which can erode both conversion and brand equity."
The upside of getting it right is equally concrete. Centralizing brand voice within prompts can reduce copy editing time by 20–30%, according to Fan. More importantly, it reduces back and forth across teams because everyone is working from the same baseline.
7. Build reusable prompt frameworks so your team stays consistent at scale
Liz Oh, associate director of digital at FRANKIE4, a podiatrist-founded Australian women’s footwear brand, emphasizes the importance of consistency across marketing assets for larger teams. “It matters more than one-off brilliance,” she says. For Oh and her team, the most effective way to use AI to achieve that consistency is to mirror how their go-to-market process already works and then turn that into structured, repeatable AI processes.
The team at FRANKIE4 builds reusable prompt frameworks by embedding fixed elements like brand tone of voice alongside variable inputs such as campaign goals, audience, product priorities, and channel requirements.
Why reusable prompts produce better output than brilliant one-offs
AI outputs stay both consistently on-brand and aligned to commercial objectives like inventory targets and revenue goals. “By standardizing inputs,” Oh says, “teams can quickly generate cohesive, multi-channel copy from a single brief, reducing fragmentation and speeding up production.”
Oh’s team at FRANKIE4 has seen some impressive results from this approach. “We’ve been able to produce briefs and copy 4–6 weeks in advance rather than manually building out these briefs and sub-items on a week-to-week basis,” Oh says. “This has given the go-to-market team time back to do more strategic thinking and forward planning over the grunt work of briefing.”
This prompt has given our go-to-market team time back to do more strategic thinking and forward planning instead of the grunt work of briefing.
8. Give your AI tool a shared memory system
Most marketers assume the problem is the prompt. Often, it’s the memory.
Emily Roberts, VP of growth marketing at Roswell NYC, an award-winning, full-service digital commerce agency, describes the pain that marketers experimenting with AI know all too well: most AI tools have no memory between sessions, so every conversation restarts from zero unless you give them somewhere to store information.
That’s why it’s so important to give your AI tool a shared memory system, Roberts says.
Roberts shares the two paths that work well for her marketing team:
- Markdown files in shared cloud storage (Google Drive, Dropbox, iCloud) that your AI reads via file access
- Notion workspace your AI reads via the Notion connector
Roberts uses the first path. The second one works best for teams already in Notion.
Either way, Roberts recommends maintaining a few standing references:
1. A top-level instructions page or file with team rules and structure
2. A memory log for corrections and standing preferences
3. A task list the AI updates each session
4. A per-project sub-page or subfolder with project-specific context
“End every session by asking the AI to log what changed,” Roberts suggests. “The next session, it reads those references first and picks up exactly where you left off, so there’s no re-briefing required.”
End every session by asking the AI to log what's changed. The next session, you'll pick up where you left off, no re-briefing required.
How Roswell NYC stopped re-briefing AI across 20 client folders
Roberts runs this across 20+ client brand folders. “When I'm building Klaviyo flows, drafting campaign copy, or running a retention audit, the brand's instruction file has the tone rules, banned words, audience segments, and design preferences already locked in,” she says.
Claude produces on-brand work from the first prompt, without her re-pasting the brand voice guide every session. And the set-up compounds, she says. For example: “a loyalty program built on top of last month's customer survey findings, a campaign calendar that remembers which offers we've already tested, an audit that picks up exactly where the previous session left off.”
AI prompting pro tip: This works with most context-aware AI tools, but pairs especially well with Claude in Cowork mode, which is purpose-built for this pattern.
9. Ground every prompt in real data before you ask AI to generate anything
Before you ask AI, give it something real to work with. That means actual campaign results, segment definitions, product reviews, or performance benchmarks, not just a rough idea of what you're going for.
Amy Last, marketing director at Must Have Ideas, a family-owned UK-based homeware company, puts it plainly: "Without data, AI is guessing what 'good' looks like for your business. When you anchor prompts in reality and ask the model to respond only within that context, the output becomes commercially credible rather than theoretically correct."
Without data, AI is guessing what 'good' looks like for your business.
The difference shows up immediately. Grounded prompts reduce speculative responses, shorten review cycles, and produce actionable outcomes.
Why ungrounded prompts stall progress
Most people who give up on AI prompting do so early. The first few outputs feel vague or slightly off, and the assumption becomes that the tool just isn't very good. But more often, the prompt is the problem, not the model.
Last has seen this pattern enough times to know the fix: "Pasting real context and data upfront keeps you out of that loop. The responses are clearer from the start, which makes you keep iterating rather than abandoning the draft halfway through. It turns trial-and-error into something far more focused."
When you give AI a real foundation, like a segment that exists in your account, or results from a campaign you’ve already run, it stops producing outputs that sound plausible in the abstract and starts producing outputs that are true for your brand specifically.
“That’s the difference between something that appears convincing and something that actually performs,” Last says.
10. Use a prompt as a production run
Most marketers are still using AI as a writing assistant. One tab for copy, another for project management, another for design, another for their email platform. AI drafts something, you copy it across, brief the next tool, repeat. The prompt is just the start of the work.
But once your AI is connected to your tools, a single prompt can be the whole job.
Kinga Dow, AI agent architect and email marketing strategist at kingadow.com, describes what that looks like in practice: "A typical campaign prompt kicks off the copy, the Asana task with brief attached, the Klaviyo campaign build, and the Figma email prototype. Then it logs the deliverables back into Asana for team review.”
“I'm not writing 4 separate briefs in 4 tools,” Dow explains. “I'm writing one production brief and the AI handles the execution across all of them."
The set-up that makes this possible is a one-time configuration: connecting Claude to Klaviyo, Figma, and Asana via the MCP server. After that, the daily work is just the prompting.
Why connected AI turns one brief into a full campaign build
When AI is only a drafting tool, the coordination overhead stays with you. You're still the one moving work between systems, creating tasks, setting up campaigns, and chasing sign-off.
When AI is connected to the tools you actually use, that overhead disappears into the prompt. The marketer's job becomes writing one comprehensive brief instead of manually orchestrating across 4 platforms. The strategic thinking stays human. The orchestration doesn't have to.
Related content
