Semantic audit: which page owns which query, and what to do about it

What a semantic audit decides — and what it does not decide

A semantic audit connects your pages to real queries, real intents and real performance: impressions, clicks, conversions. Its output is not a list of terms, it is a written decision. On one side a query-to-page map, where each intent has an owner URL; on the other a backlog of trade-offs that says what to optimize, what to create and what to consolidate, without cannibalizing the pages that already work. A finding that leads to none of those decisions has no place in the deliverable. The wider framing — the scope of a diagnosis, when to trigger it, how to read the report — is handled by the SEO audit.

What the analysis connects: pages, queries, intents, entities, performance

A useful semantic analysis does not stop at a list of terms. It observes above all the relations between:

• pages (landing pages, guides, articles, categories) and the queries actually triggered in Google;
• dominant intents (information, comparison, action, navigation) and the formats expected in the SERP;
• lexical field and entities (terms, concepts, brands, pieces of evidence) that make content understandable and rankable;
• performance: impressions, CTR, positions and post-click signals (engagement, leads).

That double reading, “engines plus visitors”, is the central point of a content-oriented audit. A page can be well written and badly placed, correctly placed and never clicked: the two faults call for opposite corrections, and only confronting the two sets of signals tells them apart.

What separates it from a technical audit and from keyword research

Three confusions come up often:

• Technical audit: it checks crawling, indexing, canonicalization and performance signals. It can surface a lot of alerts, some of which have no observable impact if the page receives neither impressions nor traffic.
• Keyword research: useful, but insufficient if it does not lead to decisions on site structure, on the query-to-page mapping and on prioritization.
• Semantic analysis: it confronts the existing content with what the SERPs expect and with the measured results, so as to turn an inventory into an editorial action plan.

What matters is not exporting terms, but building a shared language between your offer, your audience and the engines, with trade-offs your production team can be held to: it has to know which page owns what before writing the next one.

Gathering the data and setting the segmentation

Two sources are enough to start, and they answer two different questions. Search Console says what happens inside Google: queries per page, impressions, clicks, CTR, average position. It serves to spot the pages that are visible but rarely clicked — a title, angle or promise problem — and the pages in intermediate positions, 5–20 for example, often closer to a quick gain than a page that is completely invisible. The interface caps what it displays: go through a working export to sort, filter and group.

Google Analytics answers the other question: what happens after the click? A page that is very visible and weak on business performance usually calls for clarification — intent, value proposition, evidence, call to action — rather than enrichment. That reading decides the priority: a semantic optimization does not have the same value on a page that generates leads as on a peripheral page with no measured impact.

Before mapping anything, set a simple and durable segmentation: the offers actually sold and the maturity levels (discovery, comparison, decision); the personas and the vocabulary they use, often less “marketing” than your internal labels; the countries and languages, since a term that is right in one market can be counterproductive elsewhere; and finally the templates, because fixing one template can improve dozens or hundreds of URLs at once. That is the only lever that changes the scale of the work, and it is spotted at segmentation time, not afterwards.

Date that reading: it serves as the point of comparison for every trade-off that follows. You replay it at moments of rupture — a new offer, a cluster falling away, a SERP that changes its expectation — and, outside those moments, at a pace that follows the speed at which your offer and your SERPs move rather than the calendar.

Mapping: one intent per page, one page per intent

The map is what makes everything else binding. As long as it does not exist, every editorial decision is replayed at every brief, and two writers can target the same query without anyone seeing it.

The mapping rule and the owner URL

The operational principle is simple: a page must carry one main intent and become the internal reference on that intent. You then attach secondary queries around it — variants, long tail, questions — without creating twin pages on the same need. A page can rank on many formulations, but the structure must stay readable: one URL “wins” the subject, the others support it through internal linking and complementary angles.

Three rules make it auditable:

• One main query per URL, the one that defines the expected format and the promise of the page;
• One “owner” URL per intent cluster — avoid two pages targeting “price” and “cost” if the SERP shows the same expectation;
• Documented exceptions: if the SERP clearly separates “definition” from “service”, two distinct pages can be justified.

The third makes the first two workable: a rule with no written exception gets bypassed silently, a documented exception gets discussed. One last point of method: do not aggregate terms by exact words, group them by meaning and by intent.

Validating the dominant intent in the SERP and adjusting the angle

A query that looks logical on paper may expect a completely different format. Before writing, look at what Google ranks: guides, service pages, comparisons, categories. That reading decides three things: the format (service page or educational article), the angle (cost, method, mistakes, examples) and the level of evidence expected. When the SERP is locked on one format and one type of player, the point becomes to target a more precise intent, where better-structured content can take ground.

The angle chosen then feeds into the tags, never the other way round. The title tag reflects the intent, the angle and the promise; the H1 is unique and clear, followed by a consistent H2/H3 hierarchy; the meta description plays mostly on the click, it removes ambiguity rather than repeating the title. A tag that promises something other than the body of the page produces the same symptom as a mapping error: impressions with no clicks.

Secondary subjects: enriching without changing the promise

A page rarely performs on the main term alone. It wins because it covers the subtopics the intent expects: definitions, steps, criteria, mistakes, related questions. The aim is to enrich without changing the promise — a quick answer at the start, self-contained sections next, and different intents moved out to other pages.

The page type sets the limit of the enrichment. A guide covers the whole subject — steps, examples, pitfalls, questions — and acts as a hub page. A landing page answers an action intent: concise, reassuring, structured. A category makes the sorting logic and the selection criteria explicit. An article handles a sub-problem and refers to the page that owns the subject.

Keyword density, finally, is a marker and never an objective. Place the main term where it is natural — title, H1, start of the content — use a rich lexical field, and repeat for clarity (referents, definitions) rather than to “force” the engine.

Conflicts: cannibalization and alignment errors

Two different faults produce the same symptom — a page that never takes off — and call for opposite corrections: one comes from a bad choice of target, the other from a contested one. Before handling either, one rule: do not break what works. Before any change, identify the page that receives impressions, clicks and conversions. That is the precondition of every trade-off, not an end-of-project precaution.

When the page does not answer the expected intent

An alignment error occurs when the page does not match the type of answer expected: a service page targeting a purely informational query; an explanatory article targeting a query where the SERP expects a transactional page; a page that is too general and takes on no clear intent.

Two corrections are possible: either you change the angle to fit the intent, or you move the query to another page and adjust the internal linking to designate the reference page. One signal often settles which: when Google systematically associates the target query with another page on your site, the problem is not the angle but the mapping, and it is the other page you should look at.

Cannibalization: the signals, the proof, the page you keep

Cannibalization appears when several pages target the same intent: the signals dilute and none of them prevails. Three signals give it away:

• queries that “change” landing page over time, or that switch from one page to another between readings;
• two pages that are close in subject and in structure;
• internal linking that sends contradictory signals.

A signal is not a proof, and the alternation itself is not one either. A SERP moves of its own accord, Google tests pages, and two URLs can legitimately answer neighbouring intents. First record the facts over a dated window: the query and the associated URL every week for four to six weeks — a practitioner's order of magnitude, to be extended on a site with a low volume of impressions. A repeated alternation establishes that the split is unstable, not that it is harmful. Three checks close the diagnosis: do the two pages really target the same intent, do the queries concerned belong to the same context, and has the combined performance of the two URLs degraded since they began to coexist? Without an overlap of intent or an observed degradation, the finding is filed and monitored; if the alternation follows a release, it is something other than cannibalization.

That leaves designating the page you keep. Compare in this order: first the query intent and its alignment with the observed SERP, then organic performance — clicks, then impressions — then the conversions genuinely relevant to that intent, then the incoming links, internal as well as external. When the signals contradict each other — one page captures the impressions, the other the conversions — it is intent that decides, not revenue: a commercial page does not automatically recover an informational intent, and forcing it to is the surest way to lose both. Conversions separate two equally aligned pages; they do not overturn a gap in intent. The decision stops there: its page-by-page execution, from the correction plan to maintaining the stock, belongs to the content audit.

Arbitrating: create, optimize, consolidate

A semantic audit easily produces several hundred gaps. The order in which they are handled therefore counts as much as the diagnosis: with no criterion, you treat what is quick to fix rather than what pays.

The volume × position × value matrix

In B2B, search volume alone is a poor arbiter. A low-volume query can be very close to the business, and therefore convert. A robust method combines four criteria: current position (for example 5–20, 21–50, invisible); business value (service page, offer, lead generation); effort (a simple rewrite or a template rebuild); risk (loss of traffic if you touch a page that already performs).

Three cases cover the essentials: position 5–20 and high value → targeted optimizations (alignment, structure, internal linking, enrichment, a clearer title); position 21–50 and medium value → deeper re-optimization, sometimes splitting the content when two intents coexist; invisible and high value → create a reference page and build the supporting internal linking.

What justifies entering through the 5–20 zone lies in how the clicks are distributed: the share of clicks absorbed by the top 3 organic results is 75% (SEO.com, 2026). A few places gained on a query already close to the top of the page therefore do not produce the same effect as the same effort on an invisible page. The SEO statistics bring these benchmarks together with their source and their year.

The three decisions — and the case where you touch nothing

Three decisions, three logics. Optimize: when the page already exists and ranks, but lacks clarity, evidence, expected subtopics or a better angle. Create: when no page carries the intent, or when the SERP expects a format that is absent from your site. Consolidate: when several pages share the same intent. In all three cases, document the owner URL and enforce consistency in the internal linking, failing which the problem rebuilds itself a few weeks later.

The fourth decision is the most neglected: not every problem detected is a problem to fix. If a page meets its objectives — positions, clicks, conversions — a theoretical semantic “anomaly” may be noise. If the adjustments raise the risk (regression, a change in perceived intent), it is better to plan a light optimization or a controlled test. The backlog that comes out of the audit then reads in four columns.

Topical coverage: what your pages are missing

Once the map is in place, it remains to check that each page covers what its intent promises. Two markers reveal the gaps: the recurring questions visible in the SERP, in “People also ask” type sections, which signal the expected subtopics; and the missing angles — cost, step-by-step method, selection criteria, common mistakes, examples, evidence. The exercise applies to your own pages, intent by intent: what is promised, what is covered, what is missing.

Those gaps are then grouped into clusters, that is, into groups of queries brought together by meaning and by intent, never by word repetition. Three sizing rules make the grouping usable: a cluster that is too small, a handful of queries, rarely has structural value; a cluster that is too large, several hundred queries, deserves subdividing into sub-intents; and the grouping must stay verifiable and actionable — which page carries what, and why. A cluster nobody can name an owner URL for is not a cluster, it is a list. Clustering does not create the map: it makes it manageable at scale.

Measuring the same gap against the market is another exercise, with its own tools: a panel of competitors, a comparison grid, a reading of the SERP player by player. That is what the SEO competitive analysis covers, quantifying the coverage gaps against the sites that already occupy your queries.

Making pages usable by generative engines

A well-mapped page can still be absent from generated answers if its structure does not lend itself to extraction. This is the direct extension of the semantic work, not a separate project: what makes a page clear to a reader in a hurry makes it usable by an engine.

Extractable structure: short definition, self-contained sections, lists

Three moves are enough to make a page usable: a short, stable definition at the top, self-contained sections that make sense out of context, and lists wherever the information can be enumerated — steps, criteria, use cases. The test is simple: each block must be quotable on its own, without the sentence that precedes it — which also makes the page readable at a glance.

Two benchmarks support those moves, and they measure two distinct things. Pages structured with an H1-H2-H3 hierarchy are 2.8x more likely to be cited by AI (State of AI Search, 2025). Separately, the use of lists in pages cited by AI reaches 80% (State of AI Search, 2025). The GEO statistics give the other benchmarks, with their source.

Heading hierarchy, consistency and safeguards at scale

A heading hierarchy is not just about reading comfort: it exposes the outline and makes it easier to extract a specific passage. Complete it with simple reliability signals — date the updates where that makes sense, attribute the figures you cite, avoid unverifiable claims.

At scale, two safeguards keep that discipline from being lost: an internal glossary — same definitions, same key formulations — so that two pages do not contradict each other; and mapping rules — one intent, one owner page — to avoid repetition and cannibalization. Holding those rules across several hundred URLs means rebuilding the query-to-page association at regular intervals and surfacing the conflicts: that is the task covered by the audit and mapping module.

FAQ: frequently asked questions about the semantic audit

How would you define a semantic audit applied to SEO?

It is an analysis of a site's content aimed at checking that each page answers a clear search intent, ranks on relevant queries and does not conflict with its neighbours. Its output is twofold: a query-to-page map with one owner URL per intent, and prioritized recommendations.

How does a semantic audit differ from a technical SEO audit and from keyword research?

A technical audit mainly checks crawling, indexing, canonicalization and performance. Keyword research identifies terms, but stays insufficient if it does not lead to decisions on structure and mapping. The semantic audit confronts the existing content with what the SERPs expect and with the measured results, so as to turn an inventory into an editorial action plan.

Which data should be collected before launching a semantic audit?

Two sources are enough to start: Search Console (queries per page, impressions, clicks, CTR, average position, spotting the pages that are visible but rarely clicked and the intermediate positions such as 5–20) and Google Analytics (what happens after the click: engagement, journey, conversions). Combining the two makes it possible to prioritize without relying on a hunch.

How do you attach a main query to each page without cannibalization?

Choose one dominant intent per page, group the variants by meaning rather than by exact words, then designate one owner URL per cluster. Document the exceptions when the SERP genuinely separates two expectations. If two pages target the same intent, consolidate and clarify the internal linking to designate the reference page.

Which signals make it possible to detect SEO cannibalization?

Three signals: queries that change landing page over time, two pages that are close in subject and in structure, and contradictory internal linking. An isolated signal is not enough, and repeated alternation across dated weekly readings is not enough either: Google tests, and two pages can answer neighbouring intents. Check further that the intent targeted really is the same, that the context of the queries is the same and that the combined performance has degraded. The best-performing page is preserved in every case.

How do you align a keyword and its landing page with the real intent of the SERPs?

Look at the SERP for the target query: which formats does Google put forward — guides, service pages, comparisons, categories? Then adjust the page type, the H2/H3 structure, the expected sections and the angle. If Google systematically ranks another page of your site, the problem is the mapping, not the angle.

How do you validate the dominant intent in the SERP and choose the right editorial angle?

Analyse the results — videos, “People also ask”, images, institutional players — to decide the format (service or educational), the angle (cost, method, mistakes, examples) and the level of evidence. If the SERP is locked on one format and one type of player, target a more precise intent where better structuring can take ground.

How do you choose between creating, optimizing or consolidating content after the audit?

Optimize when the page exists and ranks but lacks clarity, evidence, expected subtopics or a better angle. Create when no page carries the intent, or when the SERP expects a format that is absent. Consolidate when several pages share the same intent. In every case, document the owner URL and impose consistency in the internal linking.

How do you prioritize the actions of a semantic audit on a B2B site?

In B2B, volume alone is a poor arbiter: a low-volume query can convert if it is close to the business. Combine current position (5–20, 21–50, invisible), business value, effort (a rewrite or a template rebuild) and the risk of touching a page that already performs.

When should you not intervene, even if the audit detects a “problem”?

If a page meets its objectives — positions, clicks, conversions — a theoretical semantic anomaly may be noise. If the adjustments raise the risk of regression or of a change in perceived intent, favour a light optimization or a controlled test. The finding is filed and dated: it will be reviewed at the next reading.

How do you balance keyword density, readability and SEO performance?

Use density as a marker, not as a target. Place the main term in the structuring zones where that is natural — title, H1, start of the content — then work on clarity, lexical richness and the coverage of the expected subtopics. A readable, structured and complete page lasts longer than content calibrated on a ratio.

How do you adapt a semantic audit to GEO and LLM stakes (generative engines)?

Strengthen extractability: a short, stable definition, well-titled self-contained sections, lists when the information can be enumerated, and blocks that can be quoted on their own. Add reliability signals (update dates, attributed figures) and avoid contradictions between pages through an internal glossary and mapping rules.

Which deliverables should you expect: map, recommendations and prioritized backlog?

At minimum a query-to-page map with the owner URL of each intent, a list of qualified problems (alignment, cannibalization, coverage, tags) and a prioritized backlog. That backlog carries four columns: the situation observed, the decision, the risk taken and the validation criterion that will say whether it was the right one.

How often should the analysis be relaunched and the SEO monitoring adjusted on a B2B site?

Relaunch it first at moments of rupture: a redesign, new offers, a fall in visibility on a cluster. Outside those moments, the right pace follows the speed at which your offer and your SERPs move rather than the calendar — a stable catalogue is rarely reviewed, an offer that changes every quarter forces you to redo the map at the same pace.

Which corrections should be favoured in case of duplicate content and near-identical pages?

Prioritize by performance: first identify the page that already captures impressions, clicks and conversions. Then choose the least risky option — merge and 301 redirect, canonical tag, noindex in specific cases, or a rewrite to differentiate the intent and the content. Avoid changing a page that performs in order to “fix a theoretical fault”.

Continue reading

• The page is well aligned, with no conflict, and it still does not take off: the brake is upstream, and the technical SEO audit covers crawling, indexing, statuses, canonicals, JavaScript rendering and orphan pages.
• Your pages are structured to be picked up and you now have to check whether they are: the AI GEO audit measures share of voice, cited sources and the accuracy of what is rendered back.
• The trade-offs are executed and their effect still has to be proved, cluster by cluster: SEO tracking details KPIs, annotations and the control of gains over time.
• The SERP has changed its expectation and the target has to be revalidated before touching the page again: what a search intent is and how to read it in a results page.