.png)
Language models like GPT-4o and Claude have revolutionized the way sales teams craft their outreach messages. But most teams are using them incorrectly. They enter a vague prompt, get a generic message, and are surprised when the results don’t materialize.
Sales AI only generates messages that convert if it’s given the right instructions. It’s the prompt that makes the difference between an email your prospect deletes without reading and one that triggers a response.
1. Why GPT-4o and Claude outperform other models for lead generation
GPT-4o (OpenAI) and Claude (Anthropic) are the two models best suited forAI-powered sales prospecting in 2026 for three reasons: their understanding of complex business contexts, their ability to generate natural variations of the same message, and their capacity to follow precise instructions regarding tone, length, and structure.
Claude stands out for its tendency to produce more understated and less "salesy" messages, which is often an advantage in B2B prospecting, where overpromising undermines credibility. GPT-4 excels at quickly generating numerous variations and personalizing content based on structured data.
2. The Structure of an Effective Sales Prospecting Prompt
A poorly crafted prompt forsales prospecting AI consistently generates generic messages. A good prompt contains five elements.
The role. Provide some context for the role. "You are a B2B sales representative specializing in [your industry]. You are prospecting [target persona]."
The prospect's background. First name, job title, company, industry, company size, and at least one recent development (funding round, hiring, news). The more specific the background, the more relevant the message will be.
The purpose of the message. "Draft an initial cold email with the goal of securing a 20-minute meeting." Not "write a sales email."
Guidelines. Length (under 100 words), tone (professional but direct, not salesy), structure (a personalized opening line, a value proposition, a question), and what not to do (no list of features, no "I hope you're doing well").
The personalization variable. "Use this signal for the headline: [prospect-specific signal]."
3. Examples of prompts that work
Message for an initial contact:
"You're a sales representative at Scal-IA, an agency specializing in automated B2B lead generation. Write a cold email of fewer than 80 words to [First Name], [Job Title] at [Company], which has just raised [amount] in a Series A round. The opening line should mention this funding round and connect it to the need to streamline sales outreach to support the company's growth. End with a simple question that invites a 20-minute conversation. Be direct—no empty opening lines."
Prompt for a follow-up:
"Write a follow-up email of no more than 40 words for a prospect who hasn't responded to an initial email sent 5 days ago. The approach should be different from the first message. Offer valuable content (a guide, a concrete example) rather than a direct sales pitch. Do not use the phrase 'I'm following up.'"
4. Integrate prompts into an automated workflow
Sales prospecting AI reaches its full potential when prompts are integrated into an automated workflow. Clay allows you to call GPT-4o or Claude directly within a table column, using prospect data as dynamic variables. Each row generates a unique message, personalized based on the contact’s actual context.
This process, combined with a large-scale personalization strategy, makes it possible to generate hundreds of personalized messages without any manual effort. Our article on personalization and AI prompts in lead generation details the Clay configurations for this workflow.
5. Prompting mistakes that ruin results
Asking for a "hard-hitting" message. Models interpret "hard-hitting" as "salesy and attention-grabbing," which achieves exactly the opposite of what you want in cold calling.
Do not set a length limit. Without a limit, the models generate messages that are too long. Always specify a maximum number of words.
Forget the rules. Explicitly list what the message must not contain: "no list of features, no exclamation marks, no 'I hope this message finds you well'."
Don’t stop at just one version.Sales prospecting AI can generate 10 variations of the same message in just a few seconds. Don’t stop at the first version. Test them on your ICP and keep refining them.
Conclusion
GPT-4 and Claude are powerful tools forAI-driven sales prospecting, but their effectiveness depends entirely on the quality of the instructions you give them. A well-structured prompt generates messages that your prospects feel are tailored specifically to them. This is exactly what our article on how to use AI for lead generation describes. To avoid the most common mistakes when using AI for lead generation, our article on frequent errors in B2B lead generation remains a useful resource.
