The difference between a sales professional who uses AI effectively and one who does not is almost always prompt quality. Both are using the same model. One gives it a vague task and gets vague output. The other gives it a specific, structured brief and gets copy that is 80 percent there on the first pass. Prompt engineering is the skill that separates the two.
1. Define the role
Start every prompt with a role definition: "You are an expert B2B copywriter specialising in cold email outreach for software companies targeting sales leaders." This is not just stylistic. It conditions the model's perspective and vocabulary. The same request produces noticeably different output when prefaced with a relevant role versus no role at all. Be specific about the domain expertise, not just the job title.
2. Specify the audience in detail
Not "write a cold email for a SaaS company." Write: "Write a cold email targeting a VP of Sales at a UK-based Series B SaaS company with 50 to 150 employees, using Salesforce as their CRM, who has recently expanded their sales team and is now dealing with inconsistent pipeline reporting." The more specific the audience definition, the more specific the copy. Vague audience descriptions produce audience-agnostic copy that resonates with no one in particular.
3. Give it the one thing you want the reader to believe
Every piece of sales copy has one job: move the reader from where they are now to believing one specific thing. Name that thing in your prompt. "The reader should come away believing that their current cold email approach is leaving 70 percent of potential replies on the table, and that there is a better system available to them." This single sentence shapes everything the model writes toward that outcome.
4. Anchor it to specific claims
Provide at least one specific, verifiable claim to anchor the copy. A generic claim ("we improve pipeline") produces generic copy. A specific claim ("we helped a 40-person SaaS company in London book 22 qualified calls in their first month using our cold email infrastructure") produces copy that reads like evidence, not marketing. Even if you do not have a specific case study, you can provide a specific mechanism: "anchor the copy around the fact that most cold email fails because of deliverability problems, not copy quality."
5. Name the format and constraints
Tell the model exactly what you need: "Write a 3-email cold sequence. Email 1: under 80 words, one CTA, question format. Email 2: under 60 words, different angle from email 1, focused on the cost of inaction. Email 3: break-up format, under 40 words." Without format constraints, models default to their own sense of appropriate length, which for sales copy is usually too long. Specify word count, structure and CTA type explicitly.
6. List what to avoid
Explicit negative constraints are as important as positive instructions. List the cliches and patterns your copy must avoid: "Do not use: game-changing, transform, unlock, skyrocket, leverage, seamless, robust, powerful, cutting-edge, innovative. Do not start any sentence with 'I'. Do not use em dashes. Do not open with a compliment. Do not close with a sales pitch in email 1." These constraints eliminate the most common AI output patterns that make copy read as machine-generated.
7. Include an example of good output
The single most powerful improvement to any prompt is one good example. Provide a cold email or piece of copy that represents the quality and tone you want. The model will pattern-match to it. "Here is an example of the style and tone I want. Match the sentence length, directness and structure but write entirely new content." This one addition typically improves output quality more than any other prompt element.
Your AI prompt is your creative brief. The quality of the output is bounded by the quality of the brief. Treat prompt writing as a skill worth investing in.
Frequently asked questions
What is prompt engineering?
Writing structured instructions to an AI model that produce reliable, high-quality outputs for a specific task. It involves specifying role, context, audience, format, constraints and examples.
How do I stop AI from writing generic sales copy?
Give it specific inputs: a precise audience definition, a specific claim to anchor the copy, explicit tone and format constraints, and a list of cliches to avoid. Generic inputs produce generic outputs without exception.
Should I use Claude or GPT-4o for sales copy?
Claude for conversational, nuanced copy with better tone control. GPT-4o for structured output formats and faster processing. Both produce strong sales copy with good prompts.