EXTRACTIVE SUMMARY
A content gap is a tuple the topical map requires and the property does not render. Competitor gap analysis cannot produce information gain, because its reference set is what competitors already published and therefore what the index already holds. Five gap types exist: tuple gaps, depth gaps, question gaps, sequence gaps, and gain gaps. Each is found by a different method, and only tuple gaps are detectable by the tools that dominate this category. Gaps are prioritised by prominence first, path position second, cluster completion third, and search volume last. Three conditions that resemble gaps are not gaps: absent attributes that failed filtration, attributes belonging to a neighbouring subject, and thin coverage that is actually a depth decision. A gap is closed when it sits on a revenue path or inside an almost-complete cluster, and accepted when closing it would open a cluster the property cannot finish.
What a content gap is here
A content gap is a tuple the topical map requires and the property does not render.
The definition is narrow on purpose and it depends on a prior artifact. Without a topical map there is no required set, and without a required set there is no gap, only an absence that somebody has an opinion about. A property with no map can have its content audited and cannot have its gaps analysed.
This is the first point at which this definition departs from the standard one. The usual definition of a content gap is a query a competitor ranks for and the property does not, which is a statement about competitors rather than about the subject.
Why competitor gap analysis cannot produce information gain
Competitor gap analysis cannot produce information gain, because its reference set is the content competitors have already published, which is by definition content the index already holds.
The method works like this. Export the queries three or four competitors rank for, subtract the queries the property ranks for, and treat the remainder as the content plan. Every item on that plan is a subject at least one competitor has covered, which means every article written from it enters an index that already contains an equivalent document.
Two failures follow structurally rather than through poor execution.
The plan inherits the competitors’ blind spots. Anything none of them covered is invisible to the method, and the attributes none of them covered are precisely where rare and unique coverage is available.
The plan inverts the priority rule. A competitor query export ranks by volume, because volume is the field the export carries. Prominence, which is the property of the subject rather than of the market, does not appear in the data at all.
The method is not worthless. It reliably identifies commercially proven territory, and that is a legitimate thing to know. It cannot identify where a property could be distinctive, and it is sold as though it can.
The five gap types
Five gap types account for what a property is missing, and they are not interchangeable.
Tuple gaps. The map requires a tuple and no URL renders it. This is absence in the simplest sense and the only type that resembles the standard definition.
Depth gaps. An attribute is covered at hub level and has no nodes beneath it. The subject appears addressed and the cluster is one page deep, which reads as opened rather than held.
Question gaps. A URL exists and covers its tuple, and it fails to answer a question type its cluster generates. A procedural page that never states cost, a definitional page that never distinguishes its subject from the adjacent thing it is confused with.
Sequence gaps. A query in a path has no page, so the path breaks partway. The property holds the entry query and the terminal query and nothing between, and readers arrive, progress one step, and leave.
Gain gaps. A URL exists, covers its tuple, answers its questions, and contains nothing the index does not already hold. The page is complete and contributes no reason for the property to be preferred over any other.
Only tuple gaps are detectable by the tooling that dominates this category. The other four require the map, the query network, or a judgment about the index, and none of the three is present in a keyword export.
How to find each type
Each gap type is found by a different method.
Tuple gaps: diff the map against the property. List every tuple in the inventory, mark those rendered as a heading, table row, or page, and the unmarked remainder is the gap list. This is mechanical and it is the one step automation performs well.
Depth gaps: count structural class distribution per cluster. A cluster holding a hub and no nodes, or a property whose root and seed share exceeds its node share, has depth gaps regardless of how many URLs it holds. The distribution is a more honest read on depth than page count.
Question gaps: diff the question inventory against each page. For each URL, list the question types its cluster generates and confirm the page addresses each. The types most often missing are quantitative, because writers avoid committing to numbers, and distinctive, because distinguishing a subject from its neighbour requires knowing the neighbour.
Sequence gaps: trace each query path end to end. Take a path from the query network, walk it query by query, and confirm the property holds a page at each step. The break point is the gap, and breaks on paths terminating in a transactional query are the expensive ones.
Gain gaps: audit core pages against the index. For each core page, name the element it contains that the index does not: an original number, an original framework component, an original comparison, an original observation from practice. Pages where that element cannot be named are gain gaps.
The last method is the only one requiring judgment rather than comparison, and it is the one that finds the gaps a property is least willing to see, because the pages concerned are usually well written.
→ How to run a semantic SEO audit: separating symptoms from causes
How to prioritise which gaps to close
Gaps are prioritised on four criteria in fixed order.
Prominence first. A gap on an attribute whose removal would break the central entity’s meaning outranks every gap on a removable attribute. Prominence is a property of the subject and it does not move with the market.
Path position second. A gap that breaks a path terminating in a transactional query outranks a gap on a path terminating in an informational one. A broken revenue path is losing readers who had already signalled commercial intent.
Cluster completion third. A gap inside a cluster that is otherwise complete outranks a gap inside a cluster that is barely started, because completing a cluster changes how the whole cluster reads and starting a new one does not.
Volume last. Search demand breaks ties between gaps that are equal on the three criteria above. It never promotes a gap past them.
The ordering will produce a queue whose first several items have low search volume, which is the expected result and the point at which most gap analyses get overridden. The override is the volume inversion failure arriving through a different door.
What is not a gap
Three conditions resemble gaps and are not.
An attribute that failed filtration. An attribute the map considered and rejected on relevance is not missing, it was excluded. This is why filtration rejections belong in the delivered map rather than being deleted: without them, every audit rediscovers the same rejected attributes and proposes them again.
An attribute belonging to a neighbouring subject. Adjacent subjects share vocabulary and generate queries that look in scope. Covering them extends the property into territory its source context cannot monetise, and the coverage dilutes the central entity rather than deepening it.
Thin coverage that was a depth decision. A node covered in four hundred words because the attribute is terminal and four hundred words exhausts it is not thin. Thinness is coverage that stops before the attribute does, and the distinction requires knowing what the attribute contains rather than counting words.
The common error in all three is treating an absence as a gap without checking whether the absence was chosen.
When to close a gap and when to accept it
A gap is closed when it sits on a revenue path or inside an almost-complete cluster, and accepted when closing it would open a cluster the property cannot finish.
Accepting a gap is a real option and it is rarely presented as one. A property with capacity for forty articles a year and a map requiring three hundred will not close every gap, and pretending otherwise produces a backlog that functions as a permanent reproach rather than a plan.
The accepted gaps should be recorded as accepted in the map, with a reason, rather than left as unfilled rows. An unfilled row is ambiguous between not yet and not ever, and the ambiguity guarantees the same conversation every quarter.
One gap type should never be accepted. A gain gap on a core page is a page already published and already consuming crawl attention while contributing nothing distinctive, and closing it costs an edit rather than an article.
BRIDGE
Four of the five gap types are invisible without a topical map, a query network, or both. Depth gaps need structural classification. Question gaps need a question inventory. Sequence gaps need path data. Gain gaps need a judgment about what the index holds, made against a defined subject rather than against a general sense of the field.
Most properties commissioning a gap analysis have none of these, which is why they receive a competitor keyword export and a content plan built from what their competitors already wrote. A Holistic Radar authority audit builds the map where none exists, runs all five detections against it, and returns the gap list ordered by prominence and path position rather than by volume, so that building topical authority to rank means closing the gaps that change the property’s standing rather than the gaps a tool could see.
→ Authority audits: one binding constraint, with the evidence behind it
SUPPLEMENTARY CONTENT
Whether competitor data has any role in gap analysis
Competitor data has one legitimate role, which is establishing search supply. Knowing what already ranks for an attribute tells you whether coverage alone will compete or whether the attribute requires information gain to win.
The distinction is that competitor data informs how to cover an attribute the map already contains. It should not determine which attributes the map contains.
How often to rerun gap analysis
Rerun after any publication run of twenty or more pages, and after any map extension. Gap analysis run more frequently reports the pages you know you have not written yet.
The exception is gain gaps, which are worth auditing on a slower cycle of their own, because they appear on pages that already exist and are therefore invisible to any check that counts absences.
Whether AI can perform gap analysis
Automation performs the tuple gap diff and the structural distribution count reliably, since both are comparisons between two lists.
Question gaps are partly automatable, because question types can be classified mechanically once the inventory exists. Sequence gaps require path data that is inferred rather than observed. Gain gaps resist automation entirely, because naming what the index does not hold requires knowing what your organisation knows, and that is not in the corpus.
What to do when the gap list exceeds capacity
Reduce the map rather than carrying an unbounded backlog. Remove the attributes the property will not reach, record the removal as a scope decision, and restate the tuple inventory so coverage is measured against what is actually being attempted.
A coverage figure computed against an inventory nobody intends to complete describes a plan that was abandoned rather than a property that is behind.
→ How to build a topical map: nine steps, two halves, four failure modes