Prompt to campaign objects
An AI prompt can launch ads when it writes the same account objects the dashboard, API, CLI, and MCP already share. The prompt is the interface. The campaign, creative, and targeting rows are the product.
Published 2026-09-09 · Search job: AI prompt to launch ads
An AI prompt launches ads when it creates real campaign objects on the same account you would use in the dashboard. Cursor, Claude, ChatGPT, Codex, the REST API, the CLI, and MCP are interfaces. They are not separate media-buying systems.
The public site already says this: run ads from your AI chat; one account; API, CLI, MCP, and dashboard; creative tests; real reporting. The June 2026 angle that carries the sentence is ai-chat-control. The builder-facing sibling is builder-tool-workflow. The API-facing sibling is api-quickstart-intent. Those are three promises. They must not share one generic landing page.
What the prompt has to produce
A useful prompt does not stop at copy. It has to produce the objects the account already knows:
- Targeting from a real catalog, not a guessed segment name.
- A creative that can be submitted and associated.
- A campaign with a daily budget and a lifetime envelope that is not silently one day.
- A destination that matches the ad promise.
- A way to pause, copy, and read performance through the same public surfaces.
ztdsp, MCP tools, and the dashboard assistant all talk to that account. A newly registered external-agent account can receive an API key before email verification because an empty account cannot spend. Funding still happens through the hosted prepaid path.
What the prompt must not invent
Freeform model claim_sources are not evidence. Variant copy stays claim-guarded until landing-page claim extraction exists. See Agent-written creatives, deterministic claim checks.
The prompt also must not rotate five angles into one cell. If the user asked to launch from Cursor, the page and the CTA should stay on builder-tool-workflow or ai-chat-control, not drift into a homepage tour.
This is operator-judged product method, not a published win rate for prompt-launched campaigns. The test is whether the objects exist and can be iterated without leaving the public API. Copy a campaign instead of editing it when the next prompt wants a new isolation.
The paid landing-page families already point at this job: prompt-to-campaign, Cursor ad test, Claude / ChatGPT / Codex launch pages. Those pages are product funnels. They are not consumer probes. A prompt that lands a stranger on /sb/sandwich-poll did not launch ads from chat. It launched a traffic study. Keep the destination inside the same promise the prompt just wrote.