Prompts

Create, version, draft, and ingest prompts — manually, from CSV, by AI, or from real search demand.

The Prompts section is where you decide what CiteFoundry watches. A good prompt list is the difference between a useful product and a noisy one.

Writing good prompts

You’re not testing whether the model knows your brand. You’re testing whether the model recommends your brand when a real customer asks an open-ended question. So:

• Ask the question a customer would type — usually without your brand name in it.
• Lean toward evaluation-style queries: “best”, “top”, “how do I pick”, “alternatives to”.
• Keep them short and natural. Avoid wall-of-text scenarios.

Creating a prompt

Prompts → New prompt opens a focused editor: prompt text, optional persona, and tags. Save and the prompt is published immediately.

If you want to iterate without affecting live monitors, use a draft.

Drafts

Drafts let collaborators propose changes to existing prompts without breaking the monitors that depend on them. Prompts → Drafts shows the queue. A draft has a base prompt, a proposed new text, and an optional rationale. When you approve a draft, it replaces the published text and bumps the prompt’s version. Past runs still reference the version they ran against — drafts never rewrite history.

Ingesting prompts in bulk

The dashboard supports five ingestion sources, each at Prompts → Ingest:

• Paste — paste a list, one prompt per line.
• CSV upload — prompt, optional tag, optional persona.
• AI generate — describe what your customers are looking for; we generate a candidate list using your brand-book and personas.
• Google Search Console — pick a property and we’ll surface real search queries with non-zero impressions. See the GSC integration page for setup.
• Google Ads — pick a customer and we’ll surface real ad keywords. See the Google Ads integration page.

Every ingestion creates an ingestion job. You can review the candidate prompts, accept the good ones, and discard the rest before they become live prompts. Nothing imports silently.

Versioning

Every published prompt has a version. When you edit the text directly (without drafts), the version increments and the next monitor run uses the new text. Older runs keep a reference to the version they ran against, so historical comparisons stay clean.

Tags

Tags are free-form labels (one tag per concept — evaluation, top-of-funnel, competitor-mention). They’re filterable everywhere prompts appear and useful for slicing share-of-voice by intent.

API

Manage prompts programmatically:

• GET /v1/projects/{projectId}/prompts
• POST /v1/projects/{projectId}/prompts
• POST /v1/projects/{projectId}/prompt-sources/{kind}/start — kick off ingestion from csv, paste, ai_generate, gsc, or google_ads.

See the API reference for full schemas.