The instinct when content is not working is to publish more of it. That almost never helps. What helps is deciding what each page is for, and being willing to have far fewer of them.
Key takeaways
- Every page should have one job and one audience. Pages that serve everyone serve nobody.
- Topic clusters beat isolated posts, depth on a narrow subject outranks breadth across many.
- Updating an existing page usually returns more than publishing a new one, and takes a fraction of the time.
- A publishing schedule you cannot sustain is worse than a slower one you can.
On this page
Deciding what a page is for
Content earns its place by doing one of four jobs: attracting, convincing, enabling or retaining. A page that cannot be assigned to one of them is usually being written because it is Tuesday.
| Job | Audience | Success looks like |
|---|---|---|
| Attract | People with the problem | New visitors from search |
| Convince | People comparing options | Enquiries and demo requests |
| Enable | People ready to act | Fewer support questions, faster decisions |
| Retain | Existing customers | Repeat purchase, lower churn |
Most struggling content programmes are entirely “attract” and wonder why traffic does not become revenue. There is nothing on the site for the person who arrived, got interested and needed a reason to trust you.
Depth beats breadth
Search engines and answer engines both reward evident expertise on a narrow subject over shallow coverage of many. Twelve thorough pages about one thing will outperform sixty about everything.
The practical structure is a cluster: one substantial page covering the whole subject, surrounded by pages answering specific questions within it, all linked to each other and back to the centre. The links are not decoration, they are how the engine works out that the pages belong together.
- Pick a subject you can credibly claim to know more about than most competitors.
- Write the central page, the one you would send someone who asked about the whole topic.
- List every specific question a real customer has asked you about it. Those are your supporting pages.
- Link every supporting page to the centre, and the centre to each of them, using descriptive wording.
- Only start a second cluster once the first is genuinely complete.
Updating beats publishing
An existing page that ranks on the second page of results is worth more attention than a new page that ranks nowhere. It has already proved the engine considers it relevant; it just needs to be better.
Look for pages gaining impressions without clicks, pages ranking between eleven and twenty for terms you care about, and pages with figures or examples that have quietly gone out of date. Each is a fast, low-risk improvement.
When you update, change the substance rather than the date. Refreshing a timestamp on unchanged copy is transparent, and it is one of the few things that will actively erode trust with both readers and engines.
A page that ranks eleventh is not a failure. It is a page that has already done the hard part.
Serhii Yelbaiev, Well Web Marketing
Where AI helps and where it quietly hurts
Generative tools are excellent at structure and drafting, and poor at the thing that makes content worth reading: specific, first-hand knowledge. Used for the first, they save real time. Used for the second, they produce pages that are fluent and worthless.
The failure mode is not detection. It is sameness. If a page contains only what a model already knew, it says exactly what every competing page says, and there is no reason for anyone, human or machine, to prefer it.
- Reasonable: outlining, rewriting for clarity, drafting variations, summarising your own material, generating FAQ candidates from support tickets.
- Risky: writing the substance of an expert page, inventing statistics, producing volume for its own sake.
- Non-negotiable: a human who knows the subject checks every factual claim before publication.
If your content operation is the bottleneck rather than the strategy, our content engine handles the production side while keeping the expertise and the fact-checking with people.
How much, how often
Choose the cadence you can sustain for a year, then halve it. Content compounds only if it continues; a programme that produces eight excellent pages and stops has bought nothing lasting.
One good page a month, maintained and linked, comfortably outperforms four rushed ones. It also leaves time for the updating work, which is where most of the return actually comes from.
Set a retirement rule at the same time as a publishing rule. Pages that have earned nothing in eighteen months and answer no live question should be improved, merged or removed, leaving them in place dilutes everything around them.
Frequently asked questions
As long as the question requires. Length correlates with ranking mainly because thorough answers tend to be longer, not because word count is rewarded. Padding a short answer makes it worse.
Whatever you can sustain indefinitely alongside updating what already exists. For most small teams that is monthly, not weekly.
Improve, merge or remove, in that order of preference. Anything with links or rankings should be redirected rather than deleted outright.
Not for being AI-written as such. It gets ignored for being unremarkable, which is the usual result when a page contains nothing the model did not already know.
Sources and method
Based on our own content programmes for small and medium clients, measured in Google Search Console and site analytics. Last reviewed February 2026.