An AI content agent that writes, checks its work and never publishes
An AI content agent is an AI assistant with one fixed job, a fixed process and hard limits. This one writes or improves a web page. You give it a mode (create, rewrite, improve or audit), a topic and any first-hand material. It writes a draft, checks it against this site’s facts and a quality checklist, fixes what fails, and hands back files plus a report. It never publishes, pushes code or places ads. A person decides what goes live.
Go to the agent briefHard limits
Written into the agent's own brief.
- It never publishes, pushes to git or places ads.
- It edits page components only when told to apply the copy.
- A missing fact becomes a marked question for the owner, never a guess.
- It reports what it could not verify.
Copy the agent brief
Nothing to fill in. Copy it as it is.
Role: create, rewrite, improve or audit pages for maulik.dev. Input: the mode, the topic and route (at most 5 pages per run), and any first-hand material from the owner. Loop:1. Brief: who is reading, and what do they need to do after?2. Write with the site's content skills (plain, specific, no invented facts).3. Score the draft against a content quality checklist.4. Check titles, descriptions and structured data.5. Fix what fails and check again, until the page passes or a stop condition is reached. Hard limits:- It never publishes, pushes to git or places ads.- It edits page components only when told to apply the copy.- A missing fact becomes a marked question for the owner, never a guess.- It reports what it could not verify.
How to use it
- 1Define the agent once, with its role, loop and limits written in a file. In Claude Code that is a Markdown file in the .claude/agents folder, with a short header that names the agent, describes it and lists the tools it may use.
- 2Give it one small job at a time, with the source material it should use. This agent is limited to five pages per run.
- 3Read its report first. It lists what it could not verify and what only you can supply.
- 4Review the draft yourself, then decide whether to publish.
A good result looks like
A draft that matches your facts, a short report that is honest about gaps, and no surprises, because nothing was published without you.
Stay safe
- Review every draft before it goes live. The agent is a writing assistant, not an editor-in-chief.
- Do not give an agent permission to publish or push on its own.
- Check names, numbers and prices yourself. Google’s own guidance says to fact-check and review AI-generated content before you publish it.
If it goes wrong
4 known problems, each with a fix
- 01
Problem: It fills a gap with a plausible fact.
Fix: The agent’s rules say to mark gaps and ask. Check the report for unanswered questions.
- 02
Problem: The writing becomes smooth but generic.
Fix: Give it first-hand material, because specifics are what lift a page.
- 03
Problem: It tries to do too much in one run.
Fix: Keep runs to a few pages.
- 04
Problem: It scores its own draft well and the page is still weak.
Fix: The checklist score is advice, not proof. Read the page as a stranger would, and ask what it tells the reader that the top search results do not.
Where I have used it
- Used on this siteClaude Code subagent · 22 Nov 2026
Used on this site to draft the four AI guides, then to rewrite their openings in plain language. The output was reviewed by a person and edited.
Questions
Can an agent write content that ranks?
It can draft useful pages quickly, but ranking depends on the page being genuinely useful, accurate and specific. That is why this agent keeps a checklist and asks for first-hand material. I have no ranking results for pages from this agent yet, so I do not claim any.
Why does it never publish?
Publishing is a decision with consequences. Keeping a person in that step catches mistakes and keeps responsibility clear.
Is AI-written content allowed in Google Search?
Google’s guidance (checked 2026-11-22) says generative AI can help with research and structure, but the content must still meet its spam policies. Generating many pages without adding value for users can break its scaled content abuse policy. Google also says to fact-check and review AI-generated content before publishing.
Can you build an agent like this for my business?
I build AI integrations, bots and automations, and the same ideas apply: one clear job, hard limits, and a person approving anything that goes out. Describe the job on the estimate page.
Build this for you
Want this built into your own workflow?
Tell me what you do by hand today. I will say what is safe to automate, what needs a person, and roughly what it would take. I reply within 24 hours.