Quicktion

AI Body Prompts: Examples and Best Practices

The body prompt is the first input in the AI Processing section on a destination's edit page. It tells AI what to do with each email's body before the email is saved to your destination.

This article covers the kinds of prompts that work well, with copy-paste examples for the most common use cases.

How body prompts work

When AI Processing is enabled with a body prompt, every email that lands on your destination goes through one Vertex AI (Gemini) call before it's saved. The model rewrites the body according to your instructions and the rewritten version is what lands in your destination.

Your prompt is the contract. AI does exactly what you ask — no extra summarization, no restructuring, no helpful additions. If you don't tell it to do something, it won't.

Summarize an email in N bullets

Best for: newsletters, long announcements, threads.

Summarize this email in 3 bullet points. Each bullet should be a complete, factual sentence (no headers, no introductions).

You can ask for more bullets, less bullets, or a paragraph instead. Be specific about format.

Extract action items with deadlines

Best for: client emails, project updates, meeting recaps.

Extract every action item from this email. Format each one as:
- [DEADLINE] [WHO] [WHAT]
If no deadline is mentioned, use "no deadline". If no owner is mentioned, use "unassigned".
Return only the list — no introduction, no summary.

Translate to English

Best for: customer feedback in mixed languages, international teams.

Translate this email to English. Preserve the original structure (paragraphs, lists, links). Do not summarize, do not omit details.

You can replace "English" with any target language.

Strip signatures, disclaimers, and footers

Best for: any email where the actual content is buried under boilerplate.

Remove all sales signatures, disclaimers, marketing footers, "sent from my iPhone" lines, and unsubscribe links. Keep only the main message body. Preserve the original formatting (paragraphs, lists, links).

Extract key facts as a structured note

Best for: research emails, lead inquiries, customer reports.

Rewrite this email as a structured note with these sections:
- TL;DR (one sentence)
- Key Facts (bullet list)
- Open Questions (bullet list, or "none")
Do not include greetings, signatures, or thread history.

Best practices

  • Be specific about format. "Summarize this email" leaves the model guessing about length and structure. "Summarize in 3 bullets, each one factual" gives a clear contract.
  • Tell it what NOT to do. If you don't want extra commentary, headers, or disclaimers, say so explicitly.
  • Iterate. Send a test email, look at the result in your destination, then tune the prompt. Most prompts get one or two refinements before they nail the style.
  • Don't ask for facts that aren't in the email. AI will leave them blank rather than invent — but a sloppy prompt can confuse the model.

What about extraction prompts?

The body prompt is for rewriting the body. Extraction is a separate input — see AI property extraction for how to fill destination properties from the email content.

You can use both at once, or just one. They're independent.

Was this helpful?