The Content Agent
Drafts the campaign work from your Vault and the claim you picked, and stops at review.
The writer inside Campaign Agents. It takes the claim you picked in evidence, reads your Vault, and produces something a human can review.
What It Writes
Drafts land in the review pipeline as blog-shaped pieces, or in the CMS as entries. The work on the Artifacts screen is article-shaped:
| Kind | The claim behind it |
|---|---|
| Comparison and "alternative" pages | The claim is a buying criterion: "routing reliability, not price" |
| Category and positioning pieces | The claim is how the category should be framed |
| Buyer guides | The claim is a question nobody has answered publicly |
| Blog posts | The claim is an argument worth making at length |
| Evidence updates | An existing page is right but under-proved |
For email, the sequence itself lives in your ESP and is driven from a Surface workflow. See the email sequence guides.
How a Run Goes
Agree the claim first
Which claim it is answering, who it is aimed at, the format, and what would count as success.
It reads the Vault
Your positioning, your voice and your proof are whatever the Vault says they are. See folders and structure.
It drafts
A full piece, structured, aimed at the claim rather than the topic.
It stops
Drafts land in review. Nothing publishes without a person.
Getting Good Output
Give it the claim, not the topic. "Write about lead routing" produces a survey. "Write the comparison page for buyers who care about routing reliability over price, using the closed-won evidence behind that claim" produces an argument.
Say what success looks like. A page built to rank in an answer engine is structured differently from one built to convert paid traffic.
Fix the Vault, not the draft. The same problem in a third draft is a missing rule. See correcting an agent.
Assume it will write a claim your Vault asserts but your product does not do. Nothing in the pipeline checks a claim against the product, which is what the review step is for. Treat Vault accuracy as a publishing prerequisite: see Vault and agents.
SEO and Answer Engines, in One Pass
- Search wants the page to be about one thing, structured, and linked from places that already rank.
- Answer engines want the claim stated plainly, near the top, in a form that survives being quoted out of context.
Track the prompt the piece answers in AI Visibility before you publish.
Driving It Yourself
The whole pipeline is on MCP, loaded
with ?tools=content: paste_blog to create from Markdown, get_blog and
get_blog_versions to read, request_blog_changes and quick_approve_blog for
review, toggle_blog_publish to ship.