Why Your AI Emails Sound Generic (And What to Do About It)

You ask AI to write a follow-up email after a sales call. You get: "Hi [Name], I hope this email finds you well. I just wanted to follow up on our conversation..." You rewrite it. You always rewrite it.

The problem isn't the AI. Claude, ChatGPT, and Gemini are all capable of writing emails that sound like a real person. The problem is that "write a follow-up email after a sales call" gives the model nothing to work with except the most average interpretation of what a follow-up email looks like.

AI fills gaps with averages. A vague prompt produces average output. The fix is giving the model the specifics it needs to write the actual email you want.

The five things missing from most email prompts

Every time you get a generic AI email, one or more of these is missing from your prompt:

Who is reading it. Job title, company size, what they care about, what they said last time. Without this, AI writes to nobody in particular.
What the relationship stage is. Cold outreach, second touchpoint, long-term client — each requires a different tone.
What you want the reader to do. If you don't specify the CTA, AI picks something vague like "let me know your thoughts."
Tone constraints. "Warm but direct" and "formal and brief" produce completely different emails. AI won't guess which you want.
Length. Unspecified length = longer than it needs to be. Give a word count.

Before and after

Here's the same request, vague vs specific:

❌ Generic prompt

"Write a follow-up email after a sales call"

Output: "Hi [Name], I hope this email finds you well..." — needs full rewrite.

✅ Specific prompt

"Write a 150-word follow-up email from a B2B SaaS account executive to a VP of Marketing who attended a demo yesterday. They mentioned their main pain point: inconsistent AI outputs across their content team. Reference that specifically. CTA: book a 30-minute second call to walk through the enterprise tier. Tone: warm but momentum-driving. First-name basis."

Output: email ready to send, right tone, right context, right CTA.

The second prompt is longer to write — but you write it once, and the email is done. The first prompt takes ten seconds and then thirty minutes of rewriting.

Why this matters at scale

Anthropic's June 2026 Economic Index found that 93% of AI conversations produce a concrete artifact — an email, a document, a piece of copy. Most people are using AI to create things they then have to fix. The fix is upstream: better input, better output, no rewriting.

The same report found 86% of professionals report significant time savings with AI — but only when it's well-directed. The time savings collapse when you spend twenty minutes editing a two-paragraph email that should have taken AI thirty seconds to get right.

The prompt checklist for AI emails

Before you hit send on your AI prompt for an email, verify you've included:

Recipient context (role, company, what they said or did)
Relationship stage (cold, warm, existing relationship)
Exact CTA (not "follow up" — the specific action)
Tone (two adjectives is enough: "direct and warm" or "professional and brief")
Word count or length constraint
First name vs title (if it matters)

This takes ten extra seconds to include. It saves you the rewrite.

Automate the specificity

You don't have to build this prompt from scratch every time. Prompt Optimizer takes your rough description and rewrites the input with all of this specificity baked in automatically. You type "follow-up email after yesterday's demo, VP Marketing, they cared about consistency" — it outputs the structured, specific prompt that gets you the email on the first draft.

The model doesn't change. The prompt does.