Prompts

A prompt is a reusable AI generation template — a name, a system instruction, a user message template with placeholder tokens, a default tone, and optional default audience/language — that a content creator sets up once and reuses every time they generate a LinkedIn post from an idea.

This page describes the product-facing behavior. For exact request/response shapes and status codes, see the Prompts API reference. For the read-only MCP surface, see Prompts MCP tools.

Why prompts exist

Without a saved template, every "Generate Post" click would require re-typing the same style instructions ("professional tone, no emoji, end with a question") from scratch. A prompt captures that once, as a system_instruction + user_template pair, so generation becomes: pick a prompt, pick an idea, go.

Fields

Field	Required	Notes
`name`	Yes	Short label, e.g. "Professional Thought Leader".
`system_instruction`	Yes	The system-role message sent to Bedrock — sets voice, constraints, persona.
`user_template`	Yes	The user-role message template. Supports placeholder tokens (below).
`tone`	Yes	One of `professional`, `casual`, `provocative`, `educational`, `storytelling`.
`default_audience`	No	Pre-fills the audience selector in the generate modal.
`default_language`	No	Pre-fills the language selector in the generate modal.

Template placeholders

user_template supports these tokens, substituted at generation time by the GenAI domain:

Token	Filled with
`{idea_title}`	The idea's title
`{idea_description}`	The idea's description
`{topic}`	Same value as `{idea_description}` (alias)
`{audience}`	Generation-time audience override, else the prompt's `default_audience`
`{tone}`	Generation-time tone override, else the prompt's `tone`
`{language}`	Generation-time language override, else the prompt's `default_language`
`{examples}`	Up to 5 matching example posts, each wrapped in `--- Example Post ---`

If your template omits a token, its value isn't lost — it's appended as a trailing context line (e.g. Target audience: Developers) instead.

Setting a default prompt

Each prompt card has a "Set as Default" button (hidden on whichever prompt is already default). The default prompt is pre-selected whenever you open the "Generate Post" modal, saving a click on the common path.

⚠️

A brand-new prompt is never automatically made your default — including your very first prompt in a workspace. You always have to click "Set as Default" explicitly at least once. If you were expecting auto-default behavior on first creation, note that it isn't implemented today; see the API reference for the exact create behavior.

Default-ness is tracked per user, per workspace — not workspace-wide. If a teammate also creates prompts in the same workspace, their default and yours are entirely independent; setting yours never changes theirs.

Who sees which prompts

⚠️

The Prompts page only ever shows prompts you created — even though the page doesn't call this out explicitly, GET /api/prompts is scoped to the calling user. A teammate's prompts in the same workspace are invisible on this page, though they are fetchable individually by id (e.g. by a generate-flow that already knows a promptId). See the API reference for the exact scoping.

Editing and deleting

Only the creator of a prompt can edit, delete, or set it as default — this holds even for a workspace owner who didn't create it.
Editing a prompt only affects future generations; it does not retroactively change posts that were already generated using an older version of the prompt.
Deleting is blocked if it's the last prompt you (the caller) have in the workspace — you always need at least one prompt of your own to generate from. This check only counts prompts you created; it does not look at how many prompts other people have in the workspace.

Using a prompt to generate a post

Click "Generate Post" on a triaged idea.
The generate modal loads your prompts, plus the workspace's audiences and languages, in parallel.
Your default prompt (if you've set one) is pre-selected; audience, tone, and language selectors are pre-filled from that prompt's defaults.
You can swap prompts, override audience/tone/language for just this generation, or tweak the system instruction inline — none of these overrides are saved back to the stored prompt.
Confirming calls POST /api/genai/generate with the chosen promptId plus any overrides.

See Post generation for what happens after the prompt is selected.