By KYN AI Advisory Team — AI implementation specialists, Singapore
A content agent can draft a 200-article plan if each article passes the same five stages: brief, draft, check, gate and sign-off. The agent does the drafting and the mechanical checks. A named person approves publication. The quality gate is what makes 200 articles safe, because it rejects the weak ones before a human spends time on them.
What does the plan look like before any writing starts?
A 200-article plan is a table, not a document. Each row is one article with fixed fields:
- a working title and the search question it answers;
- the reader (founder, finance lead, engineer);
- the shape (how-to, comparison, data, definition);
- the cluster it belongs to and the two or three articles it links to;
- the external facts it must state, each with a source to check.
Build the table first. Group by cluster, check for overlap against live slugs, and cut rows that repeat a question already answered. A plan with 200 distinct questions beats one with 120 distinct questions and 80 rewordings. If a live article already covers the question, give the new one a different reader or angle and link to the old one. Never retire the old one.
What are the five stages?
- Brief. The agent turns a plan row into a brief: the answer in two sentences, the H2 questions, the facts to verify, the internal links.
- Draft. The agent writes to the house voice rules from the brief, not from a blank page.
- Check. Scripts, not the model, test the draft: word count, banned words, heading structure, link validity, front matter length.
- Gate. A pass or fail decision with reasons, covered below.
- Sign-off. A person reads what passed and approves publication.
Stages 3 and 4 are code. Stage 2 is a model. Keep them separate. A model grading its own draft agrees with itself too often.
What goes in the quality gate?
A gate is a list of rules that return pass or fail. Use rules a script can test:
| Rule | Test | |---|---| | Length | Within the agreed word range | | Answer first | First paragraph under 80 words and contains the answer | | Structure | 4 to 7 H2s, exactly 3 FAQ questions | | Banned words | None of the listed words appear | | Sourced figures | Every percentage, price or deadline has an inline link | | Link validity | Every URL returns a successful response | | Duplicate | Slug and title do not collide with live content | | Meta | Title and description within character limits |
A draft that fails any rule goes back to the agent with the reasons. After two failed attempts it goes to a person, with the log. This cap matters, because an agent left to retry will produce something that passes the rules and still says nothing.
The rules above catch form. They do not catch whether the article is correct or useful. That is the sign-off.
How does a person review 200 articles?
Not by reading each one cold. Give the reviewer a page per article with the brief, the draft, the gate result, and the external facts with their source links side by side. The reviewer checks three things: is the answer right, are the facts what the sources say, and would you put your name on it.
At 10 minutes an article, 200 articles is about 33 hours of review. That is a real cost, and it is the cost that should be planned for. Spread it over the schedule. Do not publish 200 articles in a week. Release in batches of 10 to 20, watch how Search Console reports them, and adjust the brief before the next batch.
How do you keep quality from drifting across 200?
- Keep a style sheet the agent reads every time. Voice rules, banned words, example article.
- Sample. Read 1 in 10 of the approved articles again a month later.
- Track the edit rate. If reviewers change more than a few lines per article, the brief is wrong, not the reviewer.
- Version the brief. When a rule changes, record the date, so you know which batch used which rule.
- Refresh, do not retire. When a fact goes stale, edit the live article in place.
The edit-rate measure is covered in AI content that ranks: edit rates, fact checks, banned phrases.
What does it cost?
The model cost per article is small. KYN's own article engine costs about S$0.06 per article for the draft. That excludes review time, which is the larger cost, and excludes the checks that call outside services. The figure depends on the model and article length.
How do you size the plan to your capacity?
Start from review hours, not from article count. The agent drafts faster than you can read.
| Input | Illustrative value | |---|---| | Reviewer time per article | 10 minutes | | Reviewer hours available per week | 4 | | Articles reviewed per week | 24 | | Weeks to review 200 articles | about 8 to 9 |
The figures are illustrative. Replace them with your own after reviewing the first 10 articles, since the first batch is slower than the later ones. If review hours are 2 per week instead of 4, the same plan takes about 17 weeks. That is a schedule, and it is better to state it up front than to discover it at article 80.
What do you do with a failed draft?
A failed draft is information. Group the gate failures by rule after each batch.
- Many length failures. The brief gives too little to say. Add a required table or worked example to the brief.
- Many unsourced figures. The brief lets the agent state external numbers. Change it so the agent lists claims to verify instead of writing them as fact.
- Many duplicate flags. The plan has overlapping rows. Merge the questions, or give one a different reader.
- Many broken links. The agent invents URLs. Restrict it to a list of approved internal slugs and to external links a person has opened.
Fix the brief, not the individual article, and rerun the failed rows. Fixing the cause once is cheaper than fixing 30 symptoms.
Which articles should a person write from scratch?
Some rows are not suited to an agent draft. Mark them in the plan.
- Pages that rest on a client's own results, since the agent has none to draw on.
- Pages on a rule that changed in the last few weeks, where official sources are still being updated.
- Pages that carry an opinion or a recommendation you must stand behind.
Give these to a named author and use the agent for the outline and the fact table only. A plan with 170 agent drafts and 30 human-written pieces is normal.
Where KYN fits
KYN runs its own article engine this way: plan table, brief, draft, scripted checks, a gate, and a named person approving publication. The agent has its own restricted account, every action is logged with the actor, and publishing waits for a person. For clients, the engine runs in a cloud account in the client's name.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. See marketing systems and how we handle approvals.
Frequently asked questions
Can an agent write 200 articles without review?
It can write them. It should not publish them unreviewed. A wrong figure or a fabricated example in one article puts your name on it, and 200 articles multiply the chance of one.
How long does a 200-article plan take to run?
It depends on review capacity more than drafting speed. Drafting is fast. Review at about 10 minutes an article is the constraint, so plan batches of 10 to 20 over several months.
Will search engines penalise AI-drafted articles?
Google says it ranks content on whether it is helpful, not on how it was produced, as set out in its guidance on AI-generated content. It does warn against producing many pages whose main purpose is to manipulate rankings. The gate and the review exist to avoid that.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
