SEO Content Strategy Built on Topic Clusters and Bottom-Funnel Pages

It starts with the full query set, classified by intent

The first artifact is a query set, not a calendar. We export the complete page-and-query table from Google Search Console across the full 16 months it retains, pull the ranking keyword sets of the four or five domains that own your category in Ahrefs, add your internal site-search log where one exists, and merge everything into a single deduplicated file.

That file is then classified twice. The first pass assigns coarse intent: informational, commercial investigation, transactional, navigational. The second pass assigns the label that decides what actually gets built — comparison, alternative, pricing, use-case, integration, definition, how-to, troubleshooting, template. The query "crm for nonprofits" and the query "what is a crm" sit inside the same topic and belong on different pages doing different jobs.

Each classified query is then matched against a URL you already have. Three gap classes fall out of that match:

• No URL owns the query. A page has to be commissioned.
• A URL exists but answers a different intent. Rewriting beats publishing.
• Two or more URLs compete for the same intent. That is cannibalization, and it is resolved in on-page SEO by consolidation before anything new is written.

The output is a ranked list of pages to create, rewrite, merge or retire, with the query evidence attached to every line.

Bottom-funnel pages are commissioned before top-funnel volume

Most content plans open at the top of the funnel because that is where the search volume is. The mechanics argue for the opposite order.

Someone typing "asana vs monday" or "netsuite pricing" or "inventory software for wholesale distributors" has already accepted that the category solves their problem and is now choosing a vendor. The distance between that landing and a demo request is one decision. Someone typing "what is project management" is five decisions away, and four of them can be made on another company's site.

So the first cluster commissioned is the bottom-funnel set, in a fixed order:

• Comparison pages — you against each named competitor searchers actually pair you with.
• Alternative pages — the "[competitor] alternatives" query, which is a churn signal arriving on your domain.
• Pricing and cost pages — including the ones you would rather not write, because the alternative is a third-party aggregator answering the question for you.
• Use-case and segment pages — the vertical, company size or job title the product is bought for.
• Integration pages — one per connected tool, where the connection is real.

These pages are also cheaper to rank. They need fewer referring domains than a head term does, which keeps the link building budget from propping up a definition post. Top-funnel volume comes afterwards, once the cluster it feeds has somewhere to hand readers on to.

Pillar and cluster architecture, and the internal links that bind it

Topic clusters are an architecture, not a tag. Each cluster has one pillar URL that owns the head term and covers the subject end to end, plus a set of cluster pages that each own exactly one sub-intent from the classification pass.

The links are what make it a cluster rather than a folder:

• Every cluster page links up to its pillar from inside the body, with a descriptive anchor that reads as prose.
• The pillar links down to every cluster page contextually, in the section where that subtopic is discussed, never as a footer list of twenty links.
• Sibling pages link laterally only where a reader would genuinely follow the path — a comparison page to the relevant pricing page, not to every sibling it has.
• Anchor text varies across the set. Fifteen identical anchors pointing at one URL is a pattern, not a recommendation.

Two rules govern the build. One intent, one URL, so the cluster never competes with itself. And no page ships without its inbound links already added to the existing pages that should point at it, in the same deploy. A page published with no internal links has no crawl path from the URLs that already carry authority, and it will sit unranked while everyone waits for it to "age".

The cluster map is maintained as a live document, so you can see which URL owns which intent at any moment, and so can the next person who joins your marketing team.

What is actually inside a content brief

A brief is a specification. Ours runs to several pages per article and contains, at minimum:

• Target URL and status — new, rewrite or merge, with the redirect decision made in advance.
• Primary query plus the secondary queries the page must also satisfy, drawn from the classified set with their real volumes.
• An intent statement in one sentence: what the reader is trying to decide, and what the page has to resolve for them.
• A SERP audit — what ranks now, in what format, at what depth, and which of those results are aggregators rather than vendors.
• Required entities and subtopics the page has to cover to read as credible on the subject.
• The first-hand inputs we need from you: the number, the screenshot, the process step, the objection your sales team hears every week.
• Internal links in and out, with anchors and destination URLs written down.
• Schema type, title, H1 and meta description drafts.
• What to leave out, which is usually the padded definition section.

The brief is written for a subject-matter expert because a generalist writer can only recombine what already ranks. Recombination produces the average of page one, and the average of page one has no reason to outrank page one. Google's people-first content guidance asks whether a page demonstrates first-hand expertise, and first-hand expertise cannot be summarized out of other people's articles.

Where AI belongs in this process, and where it does not

We use language models, and we are specific about where.

Used for: clustering a 40,000-row query export into candidate topics far faster than a person can; deduplicating near-identical queries; summarizing what ten ranking URLs cover so the brief can name the gaps; producing a first-pass outline scaffold a consultant then rewrites; checking a finished draft against the required entity list; turning a technical explanation into a plain-language version for review.

Not used for: publishing. Not at volume, not with a human skim, not with a rewriting pass layered on top.

The reason is mechanical rather than moral. A model produces a synthesis of text that already exists, which means it produces the consensus of the pages currently ranking. That output carries no proprietary number, no screenshot of a process nobody else runs, and no opinion a practitioner would defend in a meeting. Google's guidance is explicit that content is assessed on whether it is helpful and made for people, however it was produced, so the risk is not detection — it is sameness. A page that says what forty other pages say has no argument for being ranked above them.

This is the same position set out in what seothumb.com refuses to do, and it is why our content calendar is shorter and the pages on it are longer.

The refresh cycle: how decay is detected and what happens next

Published content depreciates. SERP layouts change, competitors update, a product name changes, a statistic ages out. Decay is detected on four signals, reviewed monthly:

• Query-level click trend per URL. Trailing 28 days against the same period a year earlier in Search Console, so seasonality does not read as decline.
• Position band movement. A URL sliding from position 3 to position 7 still "ranks" and has lost most of its clicks. Band changes are flagged before the traffic loss is visible in a monthly total.
• CTR drop at a stable position. Usually a new SERP feature above you, or a title that no longer matches the query. That is a title rewrite, not a content problem.
• Impression growth with flat clicks. The page is being surfaced for queries it does not answer, which means the brief called the intent wrong.

Each flagged URL gets one of four dispositions: rewrite against a fresh brief, expand to cover the sub-intents it is now being shown for, consolidate into a stronger sibling with a 301, or retire where the topic no longer supports a page of its own.

Refresh capacity is reserved in the production calendar rather than fitted in when there is room. New pages always feel more urgent than old ones, and without a reserved share of monthly output the refresh queue loses every month until the decay is large enough to be a crisis.

How content is measured: assisted conversions by cluster, not sessions

Sessions are the wrong unit for content. One top-funnel post can triple traffic to the blog and touch no pipeline at all, and reporting that as a win is how content programs get cancelled in year two.

Measurement is built per cluster instead:

• Every URL in the map is tagged to its cluster with a content grouping in GA4, applied by URL-path rules so new pages inherit it automatically.
• Conversion paths are exported to BigQuery, which is where multi-touch paths can be queried properly rather than sampled in the interface.
• The reported figure per cluster is assisted conversions and last-click conversions side by side, with non-brand organic clicks split out from brand, surfaced in Looker Studio.
• Conversions are valued at your close rate and average order or contract value, so each cluster carries a revenue number rather than a traffic number.

That produces sentences worth acting on: the comparison cluster assisted 34 opportunities and closed 6; the definitions cluster drove 40% of blog sessions and assisted 2. The second one gets cut, or rewritten to hand readers into the first.

The same reporting frame runs across every service track, and the model behind the numbers is documented on the SEO pricing page before the first invoice.