The living Keyword Universe: from a keyword to a content calendar

Most keyword tools hand you a spreadsheet. You export a few thousand terms, sort by volume, highlight a few rows, and by the time you have a plan the export is already out of date. The spreadsheet does not know your rankings moved, that a competitor published, or that Google reshuffled a result page overnight.

The Keyword Universe is Maark's answer to that: a living map of the keywords that matter to your site, grouped by intent, mapped to the pages that should own them — updated on a schedule rather than exported once. This post walks the whole chain, from a raw keyword to a spot on your content calendar, and is honest about which parts run on their own and which parts wait for you.

Where the keywords come from

A keyword can enter your universe from several directions at once. Maark pulls from your Google Search Console queries, from crawling your own site and extracting the terms your pages already target, from striking-distance recovery (the queries you rank 11–30 for and could plausibly move up), from keyword expansion and research, and from anything you or a connected assistant import by hand. Volume and difficulty are enriched from live search data.

Every term lands in one table with a status of its own: a keyword is a candidate until it earns its place, then retained, or excluded if it is off-scope, or archived. "Imported" is deliberately not the same as "approved" — a bulk import gives you candidates to curate, not a decision already made.

Clustering is math, not a language model

Here is the part people assume is done by an LLM and is not. Maark groups keywords into intent clusters with an algorithm, because clustering by intent is fundamentally a similarity problem and a deterministic method gives you a stable, explainable grouping.

Two signals decide whether two keywords belong together:

• SERP overlap — how much their top search results actually share. If two queries return largely the same pages, searchers and Google treat them as the same job.
• Embedding similarity — how close the terms are in meaning.

Maark combines them, weighted toward the SERP signal, and runs agglomerative clustering over the result. New keywords are first matched against existing clusters by their result-page overlap, so your clusters accumulate rather than reshuffle every run. Each cluster gets a head term (its highest-volume member) and a dominant intent. Terms too lonely to cluster are kept as provisional singletons rather than forced into a group they do not belong in.

A language model does play a part in the universe — a relevance classifier decides which discovered keywords are genuinely about your business — but the clustering itself is arithmetic you could audit.

From clusters to opportunities

A cluster is a target. The next question is what to do about it, and that is where opportunities come in.

Maark runs a family of generators that each look for one specific, nameable situation and raise an opportunity when they find it: striking-distance keywords, content decay, cannibalization (two of your pages fighting over one query), pages that should rank but are not indexable, coverage gaps where no page owns a cluster, broken internal links, and more. Each opportunity carries its evidence and a priority score built from impact, urgency, confidence, effort, strategic fit, and freshness — so the list you see is ranked by what is actually worth your time, not by raw search volume.

Crucially, opportunities do not act on themselves. Each one becomes a proposal: Maark shows you the evidence and its recommendation, and a person makes the call. That is the default posture of the entire product, and it is a deliberate one.

Onto the calendar

An approved direction can become a content calendar. The calendar proposes a run of topics across upcoming days, one slot per day, ranked by each cluster's opportunity score — and a slot is a promise, not work. Editing, swapping, or removing one costs nothing, because nothing has been produced yet. Only when a slot comes due does it turn into a real content task, after a credit check, ready to move into the pipeline that researches, writes, reviews, and hands you a draft.

One honest caveat: the fully hands-off version of this loop — where the calendar materializes and routes work without you in the seat — is opt-in, per-project, and today runs only on a small allowlist of pilot projects. Every lever is off by default and reversible. For everyone else, the calendar is a plan you drive, not an autopilot you hope behaves.

Why "living" is the load-bearing word

The universe does not sit still between logins. On a schedule, Maark re-crawls changed pages, syncs new Search Console data, forms new clusters as keywords accumulate, re-checks mappings, and re-scores opportunities. When a generator stops seeing a problem it once flagged, that opportunity resolves on its own; when the problem comes back, it reopens. A daily snapshot records where the universe stood so you can see it move over time.

You do not rebuild anything. You review what changed — which is the whole point of a map that keeps up with the territory.

If you want to see the mechanics from the inside, the Keyword Universe help articles go deeper on managing keywords, clusters, and opportunities.