Prompts

Create reusable AI actions you can run on messages or drafts.

Prompts are reusable AI actions you can run on messages or drafts. Unlike the assistant (which is conversational and multi-step), prompts execute a single instruction and return a result.

Use prompts for quick, repeatable tasks: fix grammar, translate, summarize, generate a reply from a template.

Creating a prompt

Go to Settings > AI > Prompts and click Add prompt, or create one directly from the prompt toolbar on any draft.

The prompt editor has these fields:

Title and emoji. Name your prompt and pick an emoji for easy identification in menus. Missive assigns a random emoji by default.

Model. Pick a specific provider and model, or choose Auto to use your default model. The model picker groups models by provider (Anthropic, Gemini, OpenAI), with older models available under a Legacy submenu.

Instructions. The instruction text the AI will follow. Use @ mentions to add context (see Adding context below).

Options:

Option	What it does
Share with	Keep the prompt personal, share with a specific team, or share with your entire organization.
Open response in	Modal: temporary result, not saved to history. Sidebar: persistent, saved for 30 days and re-openable from the sidebar.

When a prompt includes a Current draft context reference, the open mode defaults to Modal. Otherwise it defaults to Sidebar. You can always override this manually.

Adding context

The instructions field supports @ mentions to pull in context when the prompt runs. Type @ or click the Add context button to see available options.

Template contexts

These resolve to the current conversation, message, or draft when the prompt runs:

Context	What it provides
Current conversation	The full conversation the prompt is triggered from
Current message	The specific message the prompt is triggered on
Current draft	The text selected in your draft (or the full draft content)

Searchable contexts

Type @ then navigate to a section to search and pick specific items:

Section	Shortcut	What you can reference
Conversations	⌘1	Search your mailbox or recent conversations and attach a specific one
Calendars	⌘2	Pick a specific calendar or All calendars for availability context
Responses	⌘3	Pick a specific canned response or All responses for documentation context

You can add multiple context references in a single prompt. For example, a prompt could reference the current conversation, a specific calendar, and all your canned responses.

Running prompts

From a draft

The draft toolbar shows your saved prompts as quick-action buttons. Built-in actions like Ask AI, Translate, and other defaults appear first, followed by your custom prompts under a More menu.

Click a prompt button to run it. The AI processes your instruction with the referenced context and returns the result in a modal or sidebar, depending on the prompt's open mode.

From the assistant sidebar

In the assistant's input field, type @ and navigate to Prompts to insert a saved prompt's instructions directly into your message. This lets you use prompt templates as starting points for assistant conversations.

Built-in prompts

Missive ships with default prompts for common tasks: fix grammar, improve writing, adjust tone, reply positively or negatively. These work out of the box with no setup.

Custom prompts are where things get interesting. You can combine @ context references with specific instructions to build one-click actions tailored to your business.

Example prompts

Prompt	Instructions
Billing reply	Read @Current conversation and draft a reply to the customer's billing question. Use @Billing FAQ as your source of truth for policies and pricing. Match the language of the incoming message.
Handoff summary	Summarize @Current conversation for a colleague taking over. Include: who the customer is, what they need, what's been done so far, and what the next step should be.
Follow-up nudge	Read @Current conversation and draft a friendly follow-up email. Reference what was last discussed and ask if they need anything else.
Meeting proposal	Read @Current conversation and @All calendars. Draft a reply proposing 2-3 available time slots for a meeting next week.
Translate and reply	Read @Current conversation. Draft a reply in the same language as the customer's last message. Use @All responses for reference.

The key pattern: each prompt pins the exact context it needs with @, so the AI has everything up front. Your team clicks once and gets a draft grounded in real data.

Use instructions to define your team's tone and persona. Prompts should focus on the task, not the behavior.

Writing better prompts

Getting good results from AI is about giving clear instructions. Here are practical tips for writing prompts that work well in Missive.

Write like you're explaining to a colleague. Use natural language. Describe what you need, what the output should look like, and any constraints. The clearer you are, the better the result.

Give examples of what you expect. If you want replies in a specific format or tone, include an example in your prompt. For instance: "Reply in 2-3 short paragraphs. Start by acknowledging the customer's issue, then provide the solution."

Pin the right context with @. A prompt with @Current conversation and @Billing FAQ attached will outperform a vague "reply to this email" every time. The more relevant context you provide, the less the AI has to guess.

Test with different conversations. Before sharing a prompt with your team, run it on a few different conversations. Try common cases and edge cases. A prompt that works perfectly on a simple question might struggle with a long, multi-topic thread.

Keep prompts focused on one task. A prompt that tries to summarize, translate, and draft a reply all at once will produce worse results than three separate prompts. If you need multiple steps, use assistant instead.

Ask the AI to explain its reasoning. Add "explain your reasoning" to a prompt when you're testing it. This helps you understand why the AI made certain choices and whether its logic is sound. Remove it once you're satisfied.

For more on prompt engineering, see Anthropic's guide and OpenAI's guide.

Prompts vs assistant

Prompts are quick, single-shot actions. They run one instruction and return a result. Great for repetitive tasks you do the same way every time.

Assistant is conversational. It can take multiple steps, search your data, ask clarifying questions, and build on previous messages. Use it when you need back-and-forth or when the task requires reasoning across multiple sources.