Start from one prompt
Paste one complete launch brief into your agent. The prompt maps dashboard, CLI, MCP, REST, funding, creative, campaign, reporting, and webhooks. Payment still happens on the hosted page.
Published 2026-09-09 · Search job: prompt to campaign
You can start a campaign from one prompt when the prompt is a complete launch brief, not a one-line wish.
Adprime Zero's public brief lives on Get started. It tells the agent to discover the service, pick dashboard or CLI or MCP or REST, collect the landing page and offer, store the API key as a secret, request a hosted prepay link, submit creative, create targeting, and read performance with freshness fields. /lp/prompt-to-campaign is the paid funnel for the same idea.
Why one prompt works only if it is long
A short "launch ads for my startup" prompt makes the model invent routes, pixel URLs, and provider IDs. The June 2026 reporting gaps got worse when guessed fields were treated as truth. The get-started prompt is long on purpose. It names the live surfaces so the agent does not have to hallucinate them.
Copy it whole. Then add four lines of yours: the website, the one promise, the geography, and the daily budget. If you want a $20/day cell with stricter "do not invent Beeswax IDs" language, use the Cursor launch prompt. If you want Claude, ChatGPT, or Codex wording, use the AI ads prompt pack.
What the prompt must forbid
- Collecting cards, bank details, or login codes in the chat.
- Embedding
ak_...keys in browser tags. - Inventing conversion pixel URLs.
- Treating
generated_atas provider freshness. - Using display names as segment keys.
- Pausing the last active creative.
- Declaring victory from CTR on remnant display.
Those rules are already in the public brief. If you shorten the prompt, keep them.
What happens after you paste
The agent discovers ztdsp agent-guide or the API root. It creates or connects the account. It stops at a funding URL when the wallet is empty. You complete payment. It submits the ad and waits on review. You look at any destination that landed in requires_review. It activates when balance and review are ready. It reads the first report.
That is a campaign workflow, not a generated media plan. If the agent returns only copy and no ids from the product, the prompt did not execute.
Which tool gets the paste
Cursor and Codex can run the CLI. Claude Code can too. ChatGPT and Claude chat need MCP or a human to run the emitted commands. The cluster pillar is Launch ads from Cursor. The account attach page is Run ads from an AI agent.
Open signup if you do not have an account yet. Then paste the brief in the tool you already use.