An SEO Content Audit That Produces a Usable Editorial Backlog

A content audit for SEO is a review of existing pages that ends with a decision about each page: keep it, verify missing evidence, update a confirmed problem, or investigate an unresolved one. A useful audit also names the person responsible, the evidence behind the decision and the conditions for checking the result.

The deliverable is an editorial backlog someone can work through. A spreadsheet full of traffic figures is only an input. Without a clear reader purpose, a defensible recommendation and an owner, even a detailed inventory leaves the next step undecided.

This guide shows how to build that backlog for an existing website. It includes an original scoring method and a fictional portfolio you can use to practise. The scores are editorial planning aids, not Google scores, forecasts or proof that a page will rank better.

Define the decision before collecting the inventory

Start with a sentence that describes the decision your team needs to make. In a fictional equipment supplier, that sentence might be: “Choose which published buying guides need editorial work this quarter, within the capacity of one editor and two subject reviewers.” That is specific enough to determine what belongs in the audit.

Record the website, included content types, languages, markets, review date and decision owner. Set a cut-off for collecting evidence so the inventory does not keep changing while you score it. Separate that snapshot from the working backlog, which will change as people accept assignments and resolve questions.

Choose the unit of review. Usually this is one published URL with its own reader task. A downloadable checklist attached to a guide can remain part of that guide’s row if both share an owner and outcome. If a download has separate facts, maintenance requirements or distribution, record its dependency so it cannot be forgotten when the page changes.

Agree on what “done” means before starting. A completed audit should account for every URL in scope, identify the limits of its evidence and provide an assigned next step. It does not mean every page has been rewritten. It also does not mean every recommendation has received implementation approval.

Keep the boundary clear. Canada Create’s content creation strategy guide covers the broader planning of audiences, formats and publishing. Its technical SEO audit guide addresses technical tools and checks. Here, technical findings are dependencies to assign. Detailed rewriting, new-topic research and URL merger execution need their own work briefs after the audit selects them.

Build an inventory of eligible published content

Use your publishing system’s list of published pages as the starting point, then reconcile it with public navigation and any approved inventory exports you already have. Each source can reveal omissions. A page absent from a performance report still belongs in the inventory if it meets the scope.

Record the exact public URL, current title, content type, language, market, publication status and accountable owner. Open the page and check what a visitor can actually read. Titles and old exports cannot establish the current body, and a publishing-system status alone does not show whether the public page is accessible.

Separate eligibility from performance

Include content that is published, within the agreed remit and available for authorized editorial review. Keep drafts, private pages, deleted content and material with unresolved reuse rights out of the active content set. If an exclusion needs an audit trail, record only the identifier, reason and responsible person. Do not copy restricted content into the worksheet.

A technical problem does not automatically make a page ineligible. An intended public guide with an access problem may remain in scope as an investigation, with body assessment and scoring held until evidence becomes available. Conversely, a healthy campaign page can be excluded because campaign content is outside this particular audit. Write the reason so those situations remain distinguishable.

Do not silently combine URL variants. Preserve the original URL alongside any confirmed reporting URL, and record how the mapping was established. Differences in parameters, language or location can be meaningful. Use separate fields for the public URL, the URL credited in the report and the evidence supporting that mapping. If the relationship is unclear, leave the reporting URL blank and create a verification task instead of forcing a match.

Write the reader job as an action and outcome. In the fictional supplier, “maintenance” is a topic; “help a facilities manager prepare questions before requesting a maintenance quote” is a job. That distinction lets you compare pages that use similar words but serve different decisions.

Make the coverage count inspectable

Reconcile the number of discovered records, eligible pages, exclusions and unresolved records. In the fictional companion, ten discovered records become eight eligible pages and two exclusions. One exclusion is a draft and the other has an unresolved rights question. Neither contributes to scores or performance summaries.

For a large site, a sample can be useful, but label it as a sample. Record how pages were selected and which sections remain unreviewed. A traffic-ranked sample will disproportionately represent pages that already have visibility; a template-based sample may miss unusual pages. Avoid presenting either as a complete audit.

Collect evidence with enough context to interpret it

Keep source facts separate from your interpretation. A source fact might be a reported click count for a defined URL and period. An interpretation might be that the page needs a clearer explanation. The latter requires reading the page and considering alternatives; it does not follow automatically from the number.

For every material finding, record a source reference, the date checked, the period covered, the relevant filters, a short observation and a limitation. Keep confidential report locations in your internal worksheet. The public-facing article or brief usually needs an aggregate observation, not customer records or account details.

Treat Search Console as a defined observation

Google’s Search Console Performance report documentation defines clicks, impressions, click-through rate and average position. Choose the same search type, country and device scope when comparing periods. Preserve the exact search-type label used by the report, including text-based or multimodal Web traffic where distinguished. Record the selected dates rather than relying on a label such as “last month,” which will mean something different later.

For a fictional quarterly review, you might compare July through September with April through June, then examine the same quarter of the previous year where reliable data exists. These are planning choices, not prescribed SEO windows. Match the comparison to the buying cycle and known seasonality, and document unequal periods or missing history.

A decline is a question to investigate. Google’s guidance on diagnosing Search traffic drops identifies several possible causes, including technical issues, changes in demand and ranking changes. It recommends looking across dates and dimensions. Do not translate every falling line into a writing assignment.

Google also explains that query reporting has omissions and most performance data is assigned to canonical URLs. An empty query list is not proof that nobody searched. A row associated with another URL may require a mapping check before you draw conclusions about the page being audited.

Keep missing values missing

Use a blank numeric cell plus a reason field for unavailable metrics. Suitable reasons include not collected, access unavailable, report incomplete, mapping unresolved or definition changed. Use zero only when the source supports an observed zero for the stated scope. The companion makes this distinction explicit.

Take particular care with exports. Google’s documentation for direct report exports says unavailable or non-number indicators can become zeros in downloaded files. It also describes a 1,000-row limit for representative exported data. Preserve the raw export, check its context and avoid treating every exported zero as verified absence or every omitted URL as inactive.

Do not add an arbitrary small value to a missing denominator. If impressions are unavailable, CTR is unavailable. If impressions are an observed zero, CTR remains undefined for that row. If a comparison period has zero clicks, a percentage change is undefined even when the new count is positive; report the absolute change and both counts instead.

Keep visits, events and qualified enquiries distinct

Google Analytics’ Landing page report describes the first pageview in a session and provides metrics including sessions and key events. Those definitions differ from Search Console clicks. Record the report and filters used, and do not expect the two systems’ counts to be interchangeable.

A key event is not automatically a qualified enquiry. Define the event, whether repeated activity is possible and who checks downstream quality. For a support guide, successful self-service may be the desired outcome; for a buying guide, a relevant enquiry may matter. If that outcome is unmeasured, preserve the gap rather than calling the page unsuccessful.

Read each page against its actual job

Assess the current body before scoring it. Identify the intended reader, what they need to decide, what information supports that decision and where the page leaves them uncertain. Write findings at the level an editor can act on. “Weak content” is not a brief; “the comparison table does not explain which maintenance tasks the contract excludes” is.

Google’s people-first content guidance encourages original value, reliable sourcing and content that helps readers achieve their goals. Use those questions as an assessment lens. The page does not need a predetermined length or a new date merely to look current.

Review business fit with someone who understands the offer. A guide can attract relevant readers while referring to a service the organization no longer provides. Another page may attract little search traffic but answer a question that repeatedly delays purchasing or onboarding. Record those roles and the evidence for them rather than treating traffic as the sole value measure.

Check Canadian relevance where it affects the decision: service territory, currency, units, language and the jurisdiction of cited information. Do not add a city name as decoration. A French-language page and its English counterpart should retain separate reader and maintenance responsibilities where appropriate, even if their subjects overlap.

Separate an outdated claim from an old publication date. A recently edited page can still contain unsupported information; an older explanation can remain useful. Record when the material claim was checked, who checked it and what would trigger another review. A screenshot may need attention after an interface change, while a stable definition may not need the same frequency.

Use AI assistance only with reviewable inputs and a human decision owner. It can organize supplied findings or identify questions to check, but it cannot establish that missing analytics exist or that two pages harm each other. Canada Create’s AI prompts for SEO and content provides broader prompting examples. In this audit, a generated suggestion remains a hypothesis until the supporting evidence is recorded.

Choose one of four decisions before ordering the work

Use the decision label to state what should happen next. Use the priority score later to order comparable work. Keeping these separate prevents a high number from disguising an unresolved question as an approved rewrite.

“Keep” is an active decision. Its rationale should explain what is worth preserving and when reconsideration is warranted. Do not create work solely because an audit is expected to produce a long list of changes.

“Verify” protects the audit from false certainty. Name the missing evidence precisely: the current service coverage, the page-to-report mapping, the meaning of an event, or the current source for a material claim. Give that task a due date. Repeatedly carrying “need more data” forward without an owner is not a decision process.

“Update” means the audit has established enough to brief an editorial change. It does not mean the writer may redesign the page, change its URL or remove useful sections without further review. State the reader problem and the boundaries of the proposed work, then estimate the work needed to meet them.

“Investigate” fits a confirmed decline with an uncertain cause, a technical dependency or a suspected collision between pages. Define the question that would resolve the uncertainty. In a fictional pair of maintenance guides, that might be whether their readers need different decisions despite similar search terms.

If several labels seem applicable, use the earliest unresolved dependency. Missing evidence needed to choose the action normally means verify. Adequate evidence of a problem, but uncertainty about the response, means investigate. If a clearly incorrect statement could materially mislead a reader, escalate a focused correction review immediately while retaining any separate investigation. Urgency should not manufacture permission to make unrelated changes.

Use transparent scores to order comparable tasks

The following rubric is an original heuristic for planning an editorial backlog. It is not a search-engine quality score or a validated predictor of commercial return. Its purpose is to make disagreements visible: two reviewers should be able to identify which judgement caused their priorities to differ.

Rate four factors from zero to three. All four require a written basis. Zero is a supported assessment at the lowest level; a blank means the factor cannot yet be assessed. Do not substitute zero for an unknown factor.

For O, explain why the evidence supports the next task. A decline can support an investigation, while measured readership exposed to a confirmed omission can support an update. Simply having a traffic count does not establish an improvement opportunity or a likely return.

Calculate priority as 100 × (3B + 3R + 2O + 2U) ÷ 30, rounded to one decimal place. Business relevance and reader risk each carry 30% of the available weight; opportunity and time sensitivity each carry 20%. The maximum weighted total is 30, giving a maximum displayed priority of 100.

For the fictional buying guide rated B=3, R=2, O=2 and U=2, the weighted total is 9 + 6 + 4 + 4 = 23. Its displayed score is 76.7. If any required rating is unknown, leave the score blank and state what would make the rating possible.

Record confidence separately as low, medium or high, with a reason. Confidence describes how well the evidence supports the proposed action. Multiplying priority by a confidence percentage would make an uncertain but serious problem appear less important, so this method does not do that. Uncertainty creates a task to resolve.

Likewise, keep effort separate. Estimate hours or a locally defined size only after specifying the next task. A two-hour verification and a twelve-hour update are different commitments. Do not divide priority by hours and let easy cosmetic changes automatically displace important reader problems.

Calibrate the rubric before scoring the whole site

Have two reviewers independently score a small shared sample if capacity permits. Compare their reasons, agree on the meaning of each rating and document the rubric version. The aim is consistent use within your portfolio, not a universal benchmark across unrelated websites.

Order ready tasks within the same decision queue, then apply dependencies, deadlines and available reviewers. A reader-risk rating of three triggers immediate review regardless of the total. A rights hold or unresolved authority issue blocks implementation regardless of how attractive the score looks.

Treat close scores as a discussion, not precision. In the fictional method, increasing O by one point adds about 6.7 points. If that judgement is uncertain, record the plausible range and check whether the order would change. There is no defensible reason to promise that a page scoring 76.7 will outperform one scoring 70.0.

Work through a fictional eight-page portfolio

Every business, page, figure, observation, owner role and outcome in this example is fictional. The portfolio belongs to an invented Canadian commercial equipment supplier. It demonstrates the workflow; it is not a client case, a live integration or evidence of SEO results. The companion’s example URLs use an illustrative domain and are not destinations to visit.

Two updates with different evidence and effort

F01 has 240 reported clicks from 6,000 impressions in the current fictional window. The current quarter contains 92 days and the comparison quarter 91, so these are raw totals across unequal windows, without seasonal adjustment. Its CTR is 4.0%. The comparison period has 300 clicks, so the absolute change is −60 and the relative change is −20%. Those calculations describe the observation. They do not establish that the missing comparison detail caused the decline.

The update is justified by a separate body review and a documented offer comparison. Its proposed task is to add the verified explanation after the scope is approved, while preserving useful existing material. The fictional content editor estimates six hours for the bounded edit and allocated subject review. Its measurement plan records both the reader-facing acceptance condition and later search observations.

F07 scores lower but has a narrow confirmed defect and a two-hour task. The assigned editor can correct the illustrative screenshot mismatch through the normal review process. The audit does not use that small effort to declare the page commercially more valuable than F01.

Uncertainty changes the assignment

F03 falls from 120 clicks to 30, a decrease of 90 clicks or 75%. Its body review finds no material defect. The analyst receives a task to compare suitable seasonal periods and record other possible explanations. Rewriting the guide before that check would confuse an observed change with a diagnosed content problem.

F04 has blanks for clicks, impressions and two ratings. Its offer evidence has not been verified as current. The assignment is to obtain that evidence and establish whether usable metrics exist. The blank score keeps the record visibly unresolved; it is not sorted to the bottom as a low-value page.

F05 and F06 share some terminology and reported query observations. The audit links their records with one group identifier and creates one investigation task. The reviewer must compare the reader jobs and page evidence. Counting two rows does not justify budgeting the same investigation twice.

Keeping a page can be the right result

F02 remains because its checklist serves a documented purpose and the review found no material defect. F08 records an observed zero clicks, with 80 impressions and a documented support use. Its CTR is 0%, not missing. The evidence supports keeping it; the review does not assume all useful content must acquire search traffic.

The eight eligible rows produce two keep decisions, one verify decision, two updates and three investigate decisions. Because F05 and F06 share a task, there are five distinct active tasks. In this fictional plan, their estimates total 16 hours: six for F01, two for F07, two for F04, three for F03 and three for the F05/F06 pair.

Use a bounded checklist for possible consolidation

Similarity is a reason to look more closely, not proof of harmful cannibalization. Two pages can share vocabulary while helping different readers or serving different stages of a decision. A keyword match, an AI similarity score or alternating appearances for a query cannot establish the appropriate remedy by itself.

Before recommending a separate consolidation review, answer these questions:

• Do both current page bodies serve substantially the same reader, task, market and language?
• What useful material, obligations or reader needs are unique to each page?
• Is there reliable query-and-page evidence across comparable periods, with reporting limitations recorded?
• Could a technical or measurement issue explain the observation?
• Would clearer differentiation preserve two useful jobs?
• Who can approve a structural change, and what evidence must they review first?

Record the outcome as distinct jobs, insufficient evidence, or a candidate for a separate consolidation review. Keep “candidate” separate from approval. The audit should identify the question, supporting evidence and owner; it should not contain an automatic merge or deletion rule.

Google’s canonicalization documentation concerns duplicate or very similar URLs and the signals used to indicate a preferred version. That does not make changing canonical settings an appropriate response to every editorial overlap. Any later structural implementation needs its own approved scope and technical review.

Turn the decisions into a backlog people can accept

An inventory row describes a page. A backlog task describes work. Keep them linked by stable identifiers so a single task can cover related pages without losing the page-level evidence.

Each active task needs a decision label, a concise problem statement, an owner, a reviewer, evidence references, an acceptance condition, an effort estimate, dependencies and a due date. Add a status that distinguishes proposed, accepted, blocked and complete. Someone must explicitly accept the task; putting a department name in a cell does not establish capacity.

For a fictional F01 brief, the problem statement is: “The current comparison does not explain the excluded maintenance tasks.” The acceptance condition is: “The subject reviewer confirms the added explanation against the approved offer reference, and the editor records what changed.” The task is complete when that evidence exists, not when a word-count target is reached.

For F04, completion looks different: the offer owner verifies current terms, the analyst records the availability of metrics and the audit owner revisits the decision. A verification task can succeed even if it confirms that historical metrics cannot be recovered. That limitation should then be explicit in the next decision.

Use dependency fields to avoid parallel work that conflicts. If two pages may serve the same job, hold major rewrites until the ownership investigation resolves. If the page body cannot be reviewed because of an access problem, assign that dependency first. If a key reviewer is unavailable, show the block and select another ready task.

Plan against the actual bottleneck. In a fictional team with eight editorial hours and six analyst hours available, the two updates use the eight editorial hours. The verification and investigation tasks require another eight hours of analyst capacity, so two hours must move to a later cycle unless capacity changes. A 16-hour backlog is not a feasible one-cycle plan merely because its total fits an undifferentiated team budget. These fictional estimates include allocated review time within each pool. If subject reviewers have a separate capacity limit, record that constraint too before accepting the schedule.

Keep the decision history. When a verify row becomes update, preserve the earlier label, the evidence received and who changed the decision. When an accepted task is cancelled, record why. This makes the next audit a continuation of known decisions instead of another exercise in rediscovering old disagreements.

Set the measurement plan before work begins

Every task needs a way to judge whether the intended work happened. Where the task aims to affect performance, it also needs a later observation plan. Those are separate checks.

First define acceptance: the missing evidence has been obtained, the contradiction has been corrected, the reader job has been clarified, or the investigation has returned a supported decision. Use an independent reviewer where the consequences or uncertainty justify one. Save the accepted version and the review date.

Then define observation: the metric, its definition, the report, filters, baseline window, change date and follow-up window. Choose a window that suits the available volume and decision cycle. There is no universal number of days after which an editorial change has proved successful.

For the fictional F01 update, record the baseline 240 clicks and 6,000 impressions alongside the comparison figures and scope. After publication, review comparable search observations and any qualified-enquiry measure whose definition is stable. Also verify whether readers can use the corrected comparison. An enquiry increase alone would not prove that the editorial change caused it.

Record concurrent changes that could affect interpretation: a campaign launch, a service change, a measurement correction or another major page revision. Where practical, observe unchanged comparable pages for context. This is descriptive monitoring, not a controlled causal experiment.

Do not demand a percentage improvement from a tiny count without showing the counts. In a fictional follow-up, an increase from two enquiries to three is one additional enquiry and a 50% increase. Both are mathematically true; the absolute numbers make the uncertainty easier to judge.

Close the task only when its acceptance evidence is present. Keep performance follow-up as a scheduled responsibility rather than leaving the editorial task permanently open. If later evidence challenges the original decision, create a new review entry linked to the earlier one.

Use the scored inventory companion

Download the SEO content audit companion kit (ZIP; instructions, blank CSV templates and fictional worked examples).

The companion contains a blank inventory, the worked fictional portfolio, a field dictionary, an evidence register, a scope log and a backlog. The blank files provide headers without fabricated starter values. The 42-field inventory includes separate content type, publication status, reporting URL, mapping basis, overlap group and rubric version fields; these keep the supporting decisions visible when the table is filtered or sorted. The fictional files demonstrate the same fields with explicit assumptions and calculated results.

Import the CSV files with UTF-8 encoding and preserve URLs, identifiers and dates as text where your spreadsheet allows it. Keep an untouched copy of the examples. CSV stores values rather than a working multi-sheet model, so editing a rating will not automatically update a displayed score. The instructions provide spreadsheet formulas and a manual calculation method.

Begin with the scope log, then complete the reader job and evidence fields for eligible pages. Assign a decision before calculating priority. Filter the backlog by accepted status and dependencies; do not simply sort every page by its score and start rewriting.

Use the dictionary’s blank-value rules when importing other reports. Check that an empty cell stays empty, that an observed zero stays zero and that URL joins do not silently remove unmatched pages. Update the evidence dates and explanations whenever a recommendation changes.

The worked arithmetic and file relationships were checked locally using fictional data. No live analytics connection, website change or business result is represented by those checks. Use your own authorized evidence and reviewers before relying on the companion for real editorial decisions.

Run a first review your team can finish

Select a manageable group of eligible pages, agree on their reader jobs and collect only the evidence needed for the next decision. Calibrate the ratings, identify urgent reader problems and assign the ready work within available capacity. Keep unresolved evidence visible with an owner and due date.

A useful audit leaves a clear record of what should stay, what needs checking, what can be improved and what requires diagnosis. Its value comes from decisions the team can explain, accept and revisit.

For help reviewing your editorial backlog, discuss your priorities with Canada Create’s SEO team.