Everyone has read the same AI blog post. It defines the topic, lists five benefits, warns that it is not a silver bullet, and concludes that the future is exciting. It is grammatically perfect and completely interchangeable with a thousand others. The instinct is to blame the model. The model is not the problem.
Generic output is what you get when the input is generic. “Write a post about workflow automation” contains no customer, no situation, no opinion and no constraint, so the draft returns the average of everything ever written on the subject. That average is, by definition, unremarkable.
Start from a question someone actually asked
The best source of content topics is not a keyword tool. It is your inbox, your support queue and your sales calls. Real questions carry the specificity that makes writing useful: they contain a situation, a constraint and usually a wrong assumption worth correcting.
- Pull the last fifty enquiries and group them by the question underneath the question.
- Note which ones you answer the same way every time — those are pages.
- Note which ones expose a misunderstanding — those are your strongest pages.
- Map each one to the service or product page it should eventually lead to.
The brief that changes the output
A brief that produces non-generic writing names things the model could not invent. It takes ten minutes and it is the difference between a draft you rewrite and a draft you edit.
- The exact question, in the customer’s own words.
- Who is asking, and what they have already tried.
- The position you are taking — including what you are willing to say is a bad idea.
- Two or three specifics only you know: a constraint, a failure you have seen, a rule you follow.
- The internal page this should link to, and the next step you want the reader to take.
That last line matters more than it looks. Content with no destination is a cost centre. Every post should have an obvious, non-desperate route to the page that explains how you help.
Keep the review step real
Speed of drafting is the least valuable part of the system. The valuable part is a named reviewer who checks three things: that the claims are true, that the specifics are yours, and that the voice sounds like a person who has done the work. A review step that always says yes is not a review step.
This also protects you from the quiet failure mode of AI content, which is confidently stating something you cannot support. Numbers, results and client outcomes need a source. If there is no source, the sentence does not ship.
Make it a system, not a sprint
Publishing consistently beats publishing brilliantly once. A workable rhythm has a backlog of real questions, a template brief, a drafting step, a named reviewer, and a rule for which internal pages each piece links to. That is what AI content and email engine sets up — the drafting is accelerated, the editorial judgement stays with a person.
Next step
If your content reads like everyone else’s, start by collecting fifty real questions rather than fifty keywords. When you have them, tell us what should run better and we will map them to the pages that should exist — and to the ones you already have. You can also browse DigLab services to see where content sits in the wider system.