- 01AskChoose one visitor question
- 02ObserveLook for a repeatable signal
- 03ReviewLet a person judge the change
- 04ChangeShip one controlled improvement
- 05ReturnCheck the next version
Start with a real visitor question
The page should answer something a person could genuinely ask. A clear question gives the writer a boundary, gives the reader a promise, and gives later measurement something specific to inspect.
- Write for a person first.
- Use a clear question as the page subject.
- Link the answer back to relevant work and proof.
Separate observation from decision
When feedback suggests a change, record the question, the reason, and the intended result. That context is evidence for a decision, not an automatic instruction to rewrite the page.
A weak result can suggest a useful question. It cannot, by itself, decide the wording, direction, or public claim.
Keep the change reviewable
The safest changes are small enough to understand and easy to reverse: metadata, internal links, structured data suggestions, tests, and documentation. Each proposed change should show the reason, the expected result, and the quality checks before it reaches production.
- An agent may prepare the work.
- CI checks the proposed change.
- A person approves the public release.
Make the next version comparable
The loop is incomplete until the new page can be compared with the previous one. Keep the question, the intended change, and the review decision clear enough that a person can understand what was learned.
Questions people ask
Should every website automate its content updates?
No. Automation is most useful for collecting repeatable evidence and preparing reviewable work. Voice, accuracy, and meaningful editorial decisions still need a responsible person.
What is the first useful version of this loop?
Start with one question, one page, one baseline, and one clearly described change. Reliability matters more than the number of agents involved.