Content audit: deciding page by page and holding an update rhythm

What a content audit decides, page by page

A content audit systematically examines the content already published in order to decide, page by page, on the most useful action — update, consolidation, deletion — combining an “engines” reading (visibility) and a “visitors” reading (usefulness, engagement, conversion). Its two deliverables: a documented inventory and a prioritized action plan. A finding that feeds neither is of no use.

It is not to be confused with the analysis that precedes it. The semantic audit connects queries, intents, SERPs and positions in order to designate the owner URL of each intent. The diagnosis of the existing content bears on the page “in the text”: structure, freshness, evidence, clarity, role in the journey. The data points to the page to work on, the editorial side says what to change in it.

Which URLs enter the inventory, and which ones you do not open

A useful inventory is not limited to the blog: it covers the pages that carry business value — offers, landing pages, categories, guides — and those that create SEO side effects: tags, filters, pagination, archives. Keep actionable fields: URL, template, content type, stage of the journey, objective, target query, date of last update, decision status.

That leaves what the inventory does not settle: where the in-depth reading stops. Separate the pages to audit in depth — those that generate impressions, clicks or conversions, plus the strategic pages with no traffic — from the pages to audit at surface level: long tail, weak legacy, generated families, qualified in batches and by template. The switching criterion is twofold: a URL moves into depth as soon as it receives impressions on a query you want to hold, or as soon as it carries a stage of the journey. Without that line, the method does not hold beyond a few hundred pages.

The four statuses and their validation criterion

To avoid a “theoretical” audit, define stable decision rules from the outset, and keep only four: keep, update, merge, delete. But a status says what you are going to do, never how you will know it was the right decision. So add a next action column and a validation criterion column, which turns a table of findings into a verifiable commitment. That criterion is chosen along with the action, not once and for all: a merge is judged on the stability of the combined clicks, a deletion on the absence of loss elsewhere, an update on position and post-click signals as much as on CTR. And a page reworked to be picked up by generative engines is validated neither by the impressions nor by the CTR in Search Console, which do not see those citations.

The data to gather, and what each one answers

A robust audit rests on measurable data and a qualitative grid. The difficulty is not to “measure everything”, but to connect each figure to an action. Three sets of data are enough, each answering one question.

Google Search Console answers the first: what happens inside Google? Impressions, clicks, CTR, average position, queries per URL. It isolates two profiles — the pages seen a lot and chosen little, and those that stagnate in intermediate positions. Google Analytics answers the other: what do visitors do after the click? Engagement and conversions per landing page. That pair connects acquisition and business performance without multiplying tools.

Data alone is not enough to qualify a piece of content. The same page can be visible but rarely chosen — a promise problem — or generate little traffic while converting strongly — a discoverability problem. So document, for each URL kept for in-depth review: the dominant intent (information, comparison, action) and the expected format; the stage of the journey targeted and the next logical action; the perceived value — does the content solve a problem, does it bring evidence? Six axes make that reading repeatable: audience, value, storytelling, format, engagement, conversion.

That leaves the technical layer, where the aim is not to redo a full audit. You record only the signals that weigh on how content is understood and indexed: pages indexed when they should not be, titles and H1s duplicated by a template, crawlable filter pages, missing tags. They become critical when they generate similarity at scale. Useful content can underperform if the template penalizes it: in that case it is the template you fix, not the text.

Reading the indicators: where the lever is

A good audit does not look for “the best pages” in absolute terms, but for the best levers: where a realistic editorial optimization creates a measurable gain, with no needless risk on the stable pages. Two readings follow one another: what the indicators say about a page, then the order of treatment.

High impressions, low CTR: what to rule out first

A page with many impressions and a low CTR suggests a promise mismatch: a title that is too generic, an angle unsuited to the intent, or better-structured competition in the SERP. The ratio on its own does not demonstrate it, and four checks come before any rewrite. Average position first: a low CTR in position 15 is not a title problem, it is a ranking problem. Device next, since mobile and desktop display neither the same SERPs nor the same rates. Then the split between brand and non-brand queries, which flattens the average when the two are mixed together. Finally the composition of the SERP — AI Overview, rich results, ads, question blocks — which pushes the blue links down whatever your tags say. Once these four causes have been ruled out, the promise mismatch becomes the most likely hypothesis, and the most profitable profile to handle, since only the front door has to be redone.

Two benchmarks help in choosing what to test. The impact of an optimized meta description on CTR is +43% (MyLittleBigWeb, 2026). The average CTR for a title containing a question stands at +14.1% (Onesty, 2026). These are not “recipes”, but leads to test when the page is already visible. The SEO statistics bring these benchmarks together with their source and their year.

Expected impact, effort, risk: where to start

The most robust prioritization combines three dimensions:

• Expected impact: the business value of the page — leads, sales, role in an offer.
• Effort: a light update, a structural rebuild, or a template fix.
• Risk: a page that already performs (caution) against a stagnating page (low risk).

One pragmatic rule settles it: start with the pages that are “close” to a threshold — intermediate positions, existing visibility — and that support a clear intent. That avoids scattering effort on subjects where the site has no traction yet. The third criterion is the one people forget: on a page that already meets its objectives, a theoretical anomaly is often noise; you file the finding and review it at the next reading.

Editorial quality, measured

Quality in SEO is not a question of “literary” style, but of usefulness, understanding and trust. It is measured with a short, repeatable grid, applied to every page kept for in-depth review: without it, two reviewers deliver two verdicts.

Clarity, structure, evidence, depth

In an editorial audit grid, systematically check:

• Immediate clarity: does the answer or the promise appear right at the start?
• Structure: a logical H2/H3 hierarchy, lists where useful, paragraphs that breathe.
• Evidence: concrete examples, sourced figures, unambiguous definitions.
• Depth: does the content cover the essentials without digressing?

To calibrate depth, the average length of an article in Google's top 10 is 1,447 words (Webnyxt, 2026), and recommended formats exist depending on the intent served (Backlinko, 2026). The aim is not to reach a quota, but to check that the page has the level of explanation expected for its type. The same grid qualifies thin content, which does not come down to a word count: it is content that does not answer the intent sufficiently. Three signals: insufficient coverage, a vague promise, an absence of evidence.

Subject coverage and editorial consistency

A page can “talk about the right theme” and still be incomplete. Look for three gaps: the missing angles (definitions, prerequisites, steps, common mistakes, limits), the grey areas (undefined terms, implicit assumptions, undemonstrated promises) and the recurring questions prospects ask before acting. The exercise applies to your own pages: what is promised, what is covered, what is missing.

The diagnosis is more accurate when you connect the page to its place in a semantic cocoon: an owner URL carries the main intent, the satellites address sub-problems, and the internal linking prevents twin pages. Consistency, for its part, is noticed when it is missing: an informational page that pushes a CTA too early, a conversion page with no evidence, an unstable tone. Check the promise, the vocabulary used against your personas, and CTAs suited to the stage of the journey.

Duplication and cannibalization: qualify before acting

Most problems come less from “bad writing” than from a multiplication of URLs covering the same intent. The symptom is quick to observe: several close pages share a query and none of them prevails. Before fixing, qualify the cause — real duplication, template similarity, legitimate segmentation: a reflex consolidation destroys as much value as duplication does.

Duplication, similarity, legitimate variations

Duplication can be literal — copied blocks — or structural: same H1s, same sections, same angle. Similarity is not always a problem: local pages, product variations or sector pages are legitimate if they bring specific value. The rule that settles it: if the page changes neither the intent nor the useful information, it risks competing with another URL.

The riskiest areas are often “generated” rather than written: tags, filters, URL parameters, pagination, printable versions. They produce hundreds of near-identical pages, with identical titles. In the inventory, mark those families as “templates” so as to decide at the right level: fix one template rather than rewrite 200 pages.

Reading by query and by URL: the two traps

Two traps come up often:

• Creating a new page for a lexical variant when the SERP intent is identical. The right action almost always consists in enriching the existing URL.
• Stacking up near-identical content because each piece “has a little traffic”, which ends up preventing any one page from becoming the reference.

The cross-reading “query to URL” and “URL to queries” is enough to decide between consolidating and segmenting: you segment only if the expectations are genuinely separate and stable. A SERP that distinguishes two formats justifies two pages; a variation in vocabulary justifies none.

The correction plan: rewrite, merge, canonicalize, de-index, segment

Once the cause is identified, choose the least risky action:

• Rewriting: genuinely differentiate the angle, reinforce the evidence, clarify the promise.
• Merging: combine two pages that are too close, keep the better one and redirect the other.
• De-indexing, depending on the context: utility pages, parameters, pages with no search value.
• Canonicalization: if several variants have to exist for the user, but not for the index.
• Segmentation: create two pages only if the intent is distinct and stable.

A consolidation ends with four checks, which separate a correction from a regression: the internal linking no longer points to the old page; the anchors designate the reference page; canonicals and redirects do not contradict each other; title, introduction and sections reflect the promise that was kept.

Delete, merge, optimize: the decision rules

The most sensitive decisions bear on deletion and consolidation: badly executed, they create traffic losses; well executed, they simplify the site and reinforce the authority of the key pages.

Delete a page when its value is nil or negative: obsolete content with no lasting intent, very weak content that brings neither qualified traffic nor conversion, a strict duplicate of another URL. Before deleting, check five points: does it receive impressions or clicks? does it have useful internal links? has it received external links that deletion would lose? does it serve a stage of the journey? does it have a use outside SEO — a service page, a legal notice, a landing page used in a campaign, an address given out to clients? A single positive answer is enough to prefer a consolidation or an update. And a technical page indexed by mistake is a matter for noindex, not for deletion: it has to leave the index, not the site.

To merge, start from the best-performing page — or the one best aligned with the intent — and bring into it the best of the other: sections, examples, FAQ, evidence. Then redirect the old URL to the page you kept when it no longer has a role; update the internal linking; harmonize title, H1 and sections to avoid recreating an ambiguity. The objective is to consolidate the signals on a single owner URL.

Optimizing, finally, follows a simple logic in which the order counts: clarify the promise and the intent, structure (headings, lists, blocks), enrich (evidence, examples, missing angles), then clean up (redundancies, digressions, obsolete passages). Check the basics — title, meta description, heading hierarchy, internal links, image attributes — and document the before and after, failing which the real effect will stay unverifiable.

Industrializing: workflow, automation and refresh rhythm

Automation speeds up the inventory, the detection of patterns and the prioritization. But the final decision stays editorial: pages with near-identical titles can be legitimate, or reveal harmful duplication, and only a human reading settles it. That is the part of the work that cannot be delegated.

From diagnosis to post-publication control

An audit is only worth something if it leads straight into production. The minimum viable version comes in four stages:

• 1. Diagnosis: table and statuses.
• 2. Brief per priority URL: intent, structure, evidence, CTA, elements to remove.
• 3. Production: update, rewrite, consolidation.
• 4. Monitoring in Search Console and Analytics, before and after, over a comparable period.

The “over a comparable period” is part of the rule: comparing a back-to-school month with a summer month proves nothing. That control answers a single question — has the change taken hold on the target query? — the audit having confined itself to setting realistic objectives.

At high volume, automate first what can be made objective: high impressions with a low CTR, groups of URLs with the same title or the same H1, old pages not updated, segments with no performance. Start with the quantitative, then refine with the qualitative. A standard grid reduces the noise: simple statuses, plus the two columns that change everything, priority (impact × effort × risk) and action (exactly what will be done). That avoids audits that “observe” without triggering any production.

Classifying the existing content and triggering the refresh

The refresh has to become a system, not a reaction to urgency. It starts with a classification, the lifespan of a piece of content deciding how it is handled by default:

• Evergreen: lasting subjects (definitions, methods, guides). The best candidates for a regular refresh.
• Seasonal: comes back every year. Refresh planned before the season.
• Event or announcement: perishable by nature. Plan an expiry, a consolidation, or a transformation into evergreen where possible.
• Comparison: quickly obsolete (prices, features, versions). Strict governance of dates and evidence.

Four triggers then pull a page out of the calendar: a lasting fall in impressions, positions or CTR in Search Console; a drop in engagement or conversion in Analytics; dated data, expired offers, obsolete examples; the appearance of new intents in the SERP. Outside a trigger, three rhythms combine: weekly for the quick wins (titles, metas, internal links), monthly for targeted refreshes (missing sections, FAQ, evidence), quarterly for the structural projects. That is how you avoid accumulating editorial debt.

One safeguard to finish: changing a date or adding two sentences is not enough. A refresh that counts changes the alignment between the intent and the page, adds the missing sections and evidence, improves the internal linking and reworks the promise. Executing that backlog in series — the pages of the action plan rewritten one after another, under a human validation that keeps the last word on the evidence — is the task of the content production module.

Making content usable by generative engines

With the rise of generative engines, performance is no longer limited to the click: impressions can rise while traffic falls. The share of searches ending without a click (zero-click) reaches 60% (Squid Impact, 2025), and the click-through rate for the first position in the presence of an AI Overview is 2.6% (Squid Impact, 2025). A page can therefore be read without being visited: you then assess its ability to be summarized correctly, as much as to be clicked.

Three moves make a page usable, and they are the ones that make it readable for a reader in a hurry: a short, stable definition at the top; self-contained blocks — steps, criteria, use cases, a mini FAQ — that make sense out of context; lists wherever the information can be enumerated. Two benchmarks confirm it: pages structured with an H1-H2-H3 hierarchy are 2.8x more likely to be cited by AI (State of AI Search, 2025), and the use of lists in pages cited by AI reaches 80% (State of AI Search, 2025).

That leaves “machine” readability, which does not mean writing for a robot but reducing ambiguity: define the terms, name the concepts explicitly, avoid vague pronouns, keep a stable organization. Add attributed evidence whenever you put forward a figure: 66% of users trust AI outputs without checking their accuracy (Squid Impact, 2025), which shifts the burden of verification onto the source page. All of that fits in one “citability” column, and auditing it on the first pass saves rewriting twice. The GEO statistics give the other benchmarks, with their source and their year.

FAQ: frequently asked questions about the content audit

What is an SEO content audit?

It is a structured analysis of the existing pages, often in the form of an inventory, which combines quantitative criteria (visibility, traffic, engagement, conversions) and qualitative ones (intent, clarity, evidence, structure) in order to decide what to keep, update, merge, rewrite or delete. Its two outputs are a documented inventory and a prioritized action plan.

What is the difference between an editorial audit, an on-page SEO audit and a semantic audit?

The editorial audit judges the value and the quality of the pages “in the text”: usefulness, structure, evidence, promise, CTA. The on-page check verifies the page elements — tags, hierarchy, internal linking, images. The semantic audit connects queries, intents, SERPs and positions in order to designate the page that owns each intent and to arbitrate the opportunities. The three follow one another: frame, decide, execute.

How do you assess editorial quality during an editorial audit?

Use a short, repeatable grid: a clear promise right at the start, a readable structure (H2/H3, lists), up-to-date information, evidence (examples, sourced figures), and a fit with the journey — what should the reader do next? Also check the consistency of the tone and the presence of CTAs suited to the stage. The same grid qualifies thin content, which is judged on the answer it gives to the intent, not on the word count.

How do you detect keyword cannibalization?

The symptom is read in Search Console: several close URLs appear on the same query and share impressions and clicks, without any of them prevailing. Once the conflict is established across several dated readings, the question becomes one of cause — literal duplication, template similarity, or legitimate segmentation — and it is the cause that determines the action: enrich the existing URL, merge, canonicalize or segment.

Which signals point to a duplicate content problem?

Identical titles and H1s on several URLs, page structures that are strictly similar, text blocks recurring at scale (templates), and pages that are not distinguished by any useful information. The generated areas — tags, filters, parameters, pagination, printable versions — are the most frequent sources, and they are handled at template level rather than page by page.

When should content be deleted?

When the page no longer has a lasting intent (a past event, a discontinued offer), brings neither qualified traffic nor conversion, or exists only because of a technical generation (parameters, duplicates). Before deleting, check whether it receives impressions or clicks, whether it carries useful internal links, whether it has received external links, whether it serves a stage of the journey and whether it has a use outside SEO — a service page, a legal notice, a page used in a campaign. A single positive answer is enough: a consolidation or an update is then preferable. A technical page indexed by mistake, for its part, is handled with noindex rather than with deletion.

How do you choose between an update and a full rewrite?

Update if the intent is the right one but the page lacks freshness, evidence or expected sections: that is the best effort-to-impact ratio. Rewrite if the promise is vague, the structure unsuited, or if the page has to change angle to fit the dominant intent. A full rebuild is reserved for pages that are badly aligned, too superficial or structurally inconsistent.

How do you merge content without losing traffic?

Keep the strongest URL, or the one most legitimate on the intent, bring into it the best content from the other page, then redirect the old URL to it. Update the internal linking and harmonize the tags — title, H1, subheadings — to make the owner page clear. Finally check that canonicals and redirects do not contradict each other: that is where the expected gains of a consolidation are lost.

Which KPIs should be followed after trying to optimize the priority pages?

In Search Console: impressions, clicks, CTR, average position and the queries triggered per URL. In Analytics: engagement and conversions on organic landing pages. Always compare over an equivalent period, before and after, and keep an eye on visibility in impressions as much as on clicks: with enriched SERPs, a page can gain in presence without gaining in sessions.

How often should content be audited?

Adjust according to the size of the site: a quarterly check on the business-critical pages and a broader audit once or twice a year. Between two passes, three rhythms hold the maintenance — weekly for the quick wins, monthly for targeted refreshes, quarterly for the structural projects. And four triggers pull a page out of the calendar: a lasting fall in the signals, a drop in conversion, expired data, a change of format in the SERP.

How do you run an automated audit at scale with a small team?

Standardize the statuses (keep, update, merge, delete), automate the surfacing of patterns — duplicated titles and H1s, low-CTR pages, risky templates — then reserve the human time for the pages kept for in-depth review and for the consolidation decisions. The key is to attach every URL to a dated action and to its validation criterion, otherwise automation produces volume, not decisions.

How do you build a GEO-optimized content approach in from the audit stage?

Add a “citability” column to the grid: a short definition present, self-contained blocks, steps listed, attributed evidence, disambiguated terms. Well-structured pages, with a clear heading hierarchy and lists, are picked up more by generative systems. Auditing those elements on the first pass saves rewriting the same pages twice a few months later.

How do you improve visibility in AI answers after an audit?

Consolidate the owner pages by intent, structure the information into reusable blocks (definitions, lists, FAQ), reinforce the evidence and the disambiguation, then measure what is actually visible: the presence of your pages as a source in the generated answers, recorded on a dated panel of questions. The impressions and the CTR in Search Console do not measure those pick-ups and therefore cannot validate these moves. The rest belongs to the execution of the action plan: these lines are planned like any other, with a validation criterion per page.

Continue reading

• The inventory is clean, the pages are aligned and subjects are still missing: the gap is then measured against the market rather than against yourself, which is what the SEO competitive analysis covers — a panel of competitors, a benchmark and the subjects they cover and you do not.
• The inconsistencies recorded page by page repeat from one piece of content to the next: this is no longer a page problem, and it is the editorial strategy — promise, tone, overall consistency — that has to be settled before fixing the next one.