Should SDRs write prompts from scratch each time?

No. Reusable templates are faster and more consistent. Most SDRs need 3 to 5 prompts and rotate inputs into them.

Does the prompt need to be long to be good?

No. A 6-sentence prompt with role, source, persona, output, and an evidence rule is enough. Length is a poor proxy for quality.

How do you stop the model from making up details?

Anchor the model to a fixed source and add the rule 'if the source does not contain the answer, say so'.

Where should the team store its SDR prompts?

A shared doc, Notion page, or prompt library inside the sales tool. Versioned, owned by one person, and treated as a sales asset.

Prompting for SDRs: Prompts That Actually Produce Accurate Context

AI output quality is a function of prompt structure, not model intelligence. The 4-part SDR prompt (role, source, persona, output) plus an evidence rule prevents most hallucinations.

Prompt anatomy

Role and offer, source corpus, persona, output structure, and an evidence rule that forces quoted sources.

Reusable templates

Account research, persona signal, opener angle. Three templates cover most SDR work.

Common mistakes

Asking 'write me an email', skipping the source slot, and accepting the first output without a refinement pass.

Storage

Shared doc, Notion, or in-tool prompt library. Versioned and owned.