When an article loses search traffic, my first job is to work out what changed. Rewriting it immediately can waste a useful page, hide the cause of the drop and make the next comparison harder.
I use content decay for a sustained loss of useful search performance over time. An SEO regression can be more abrupt: a page, template or tracking change is followed by worse results. Neither label explains the cause. Lower demand, a broken page, a different mix of searches and an incomplete report can produce similar-looking charts.
This is the review I use to turn that uncertainty into a sensible next step. It includes a decision table, a fictional example and a blank content review worksheet. The aim is to protect useful content while finding changes I can actually justify.
Start with the question, not the percentage
“Traffic is down” needs a source, a page group and a period. I write a statement such as: “Google web-search clicks to these advice pages fell between two complete four-week periods.” That is specific enough to investigate. “The blog has lost 40%” might mix search with social traffic, partial dates and unrelated pages. In analytics I separate organic search, referral and social traffic: the end of a social campaign is not, by itself, an organic-search loss.
I also record what the content is meant to do. An article that answers a support question, attracts relevant enquiries or helps someone choose a product has value beyond a raw visit count. A high-volume query that never suited the service is not automatically the traffic I should spend most time recovering.
Choose periods that deserve to be compared
I normally start with equal, complete periods containing the same number of weekdays: four complete weeks against the preceding four, for example. That is a starting point, not a universal rule. A low-volume site needs more time; a short event may need a tighter window.
I then check the same stage of the previous season. If an event moved by a fortnight, I compare the weeks before each event as well as the calendar dates. Comparing last year’s peak with this year’s quiet preparation period would create an alarming percentage with little practical meaning. My seasonal comparison guide covers that approach in more detail.
The property, search type, country and device filters must stay consistent. I record the time zone and wait until the reporting source considers the selected days complete. I do not quietly include today in one period or switch from all devices to mobile halfway through the review.
Google’s guide to investigating search-traffic drops recommends looking at a longer history, comparing similar periods and checking the affected pages and search types. I use that as context before deciding that an older article has deteriorated.
Mark missing evidence before sorting the losers
A blank cell is not a measured zero. It can mean that a connection failed, the date range was unavailable, the page was outside the selected scope or an assessment has never run. I keep those states visible. Otherwise a spreadsheet sorts the least understood pages straight to the top of a deletion queue.
I record both the period being measured and the date the report was collected. An assessment saved last winter is not a fresh view of this month just because I exported it today. A completed export proves that rows were written; it does not prove that every underlying measurement is complete.
Search Console also omits some query information for privacy and does not show every query row. Its dimension and filtering guidance explains why query totals and filtered views can differ. I do not label an article worthless because it is absent from a filtered query export.
A confirmed zero is different from missing data, but it is not automatically a verdict on value either. It may be expected outside the season or for a low-volume support page. My worksheet distinguishes complete, partial, unavailable and not assessed. Where confidence is low, the next action is to obtain or check evidence, not to manufacture a score.
Separate a search loss from a measurement or delivery problem
I check the page before judging the writing. Does the normal URL load? Is the expected article present on a phone? Is it still intended to be indexed? Do the canonical, robots settings and internal links point to the right place? A successful page response can still contain an empty template, an old cached version or the wrong content.
If a whole section drops after a release, I look for a shared cause: routing, templates, rendering, canonicals or measurement. I sample affected and unaffected pages. Editing fifty introductions will not fix a template that has accidentally removed their main content.
I keep search clicks and site analytics alongside one another without expecting identical totals. A change in consent or tracking can alter analytics while search visibility remains similar. Google explains these different measurement systems in its Search Console and Analytics comparison guide. A disagreement gives me another question to test.
Find the part of the audience that changed
For a page that still works, I split the review into a few manageable checks. I avoid changing several filters at once, because that makes it difficult to tell which split explained the difference.
- Pages: is the loss concentrated in one article, a topic or a whole template?
- Queries: did the page lose the searches it was written to answer, or did a temporary, less relevant query disappear?
- Devices and countries: is the decline broad, or concentrated in an audience with a different experience or demand pattern?
- Search type: am I comparing web results with web results, rather than combining a changing mix of image, news or other surfaces?
- Clicks and impressions: did appearances fall, the share of appearances turning into clicks fall, or both?
Stable impressions with fewer clicks sends me towards the queries, result presentation and competing answers. Falling impressions sends me towards demand, indexing, visibility and coverage. These are investigative routes, not diagnoses. Average position can change when the query mix changes, so I inspect important queries rather than treating one blended number as the explanation.
My guide to using search data in editorial decisions helps turn those findings into a manageable review list.
Prioritise what is useful and at risk
I sort by the absolute loss as well as the percentage. A fall from two clicks to one is 50%; a fall from 500 to 350 loses far more visits. I then add relevance, existing enquiries or other useful outcomes, editorial importance and the confidence of the evidence.
Remaining valuable clicks are a reason to be careful. Before combining or retiring a page, I inspect the questions it still answers, useful links pointing to it and any audience that would lose its only clear route to that information. A page can need improvement while still doing an important job.
I start with a small set where the problem is clear enough to act on. Uncertain cases stay on a research list with a named next check and owner.
Choose a recommendation before applying an action
A descriptive bucket, such as high traffic, seasonal or low activity, tells me what a page looks like in the report. It is not an instruction: two seasonal pages may need different treatment. I record the recommendation, its reason, the proposed live action and the person approving it separately. A label should not silently change a URL.
On smaller screens, scroll the tables sideways to see every column.
| Decision | When it may fit | Check before acting |
|---|---|---|
| Keep and monitor | The answer remains useful; demand is seasonal or the change is small. | Record the next relevant review period and protect useful routes into the page. |
| Update | The same intent remains, but facts, examples or the answer need work. | List the reader’s unanswered questions and preserve information that still earns useful visits. |
| Combine | Two pages serve substantially the same need and a stronger destination is possible. | Map what each contributes. Build and check the replacement before removing a route to it. |
| Redirect | A page has moved or has a clear, relevant replacement. | Test the destination, response, internal links and the journey from the old URL. |
| Archive | Historical information remains useful but should be clearly dated. | Explain its period and current relevance. Decide indexing separately. |
| Noindex or remove | There is a specific reason to exclude the page from search or withdraw it. | Assess reader access and links. Choose the intended response and record the consequences. |
| Gather evidence | The data is missing, stale or outside the right comparison. | Fix the scope or collection problem and rerun before making a content decision. |
An archive is an editorial treatment, not an automatic search directive. Equally, keeping a page available with noindex is different from removing it. A crawler needs access to read a noindex instruction; I explain that distinction in robots.txt and noindex.
I do not redirect unrelated retired articles to a homepage for convenience. Google’s redirect guidance distinguishes permanent moves from temporary ones. I choose a relevant destination and test it rather than assuming that any redirect preserves the old page’s purpose or performance.
If I genuinely withdraw a page with no suitable replacement, an appropriate 404 or 410 response may be the right outcome. I check its remaining value and routes into it before making that decision.
A worked example: three drops, three different decisions
This example is invented for illustration. It does not describe a client, a real site or a claimed result. Imagine three advice articles reviewed over two complete four-week periods, using the same property, web-search type and filters.
| Article | Earlier clicks | Recent clicks | What the next check finds |
|---|---|---|---|
| Annual event checklist | 800 | 320 | The event is later this year. Equivalent weeks before the event recorded 340 clicks last year. |
| Choosing a suitable size | 400 | 240 | Demand looks similar, but the page lacks an answer now present in competing results. |
| An older setup guide | 90 | Unavailable | The report connection failed. The URL still receives visits in another measurement source. |
The event checklist is down 60% against the preceding period, but only about 6% against the comparable pre-event period. That second comparison changes the question. I would check its dates and usefulness, then monitor the coming demand; the first percentage does not justify retirement.
The sizing article has lost 160 clicks, or 40%. I would inspect the relevant queries, confirm the page works, then improve the missing explanation while preserving what already helps readers. That is an update hypothesis, not a promise of recovery.
The setup guide has no usable recent click figure from this report. I would repair the collection and review the content. Entering zero would invent a 100% loss and create a false reason to remove it.
None of these observations has applied a redirect, changed indexing or deleted an article. The review produces decisions to check, not permission for a script to act on every row.
Make an update that answers the problem you found
I turn the diagnosis into a short brief: who the page helps, the unanswered question, the inaccurate or missing material, what should stay and how I will check it. That might mean a better comparison, a corrected example, clearer ordering or a genuinely useful explanation near the decision.
I avoid padding an article to a target length or changing its date to make it look newer. Google’s people-first content guidance explicitly cautions against superficial freshness changes. The reader should be able to tell what became more useful.
For a combined page, I verify the replacement first: its content, status, canonical, indexing intention, useful links and mobile experience. I keep a record of the original URL and a recovery route before changing anything difficult to reverse. My older-content review guide covers that wider editorial handover.
Save a baseline and define the next comparison
Before publishing, I save the exact dates, filters, export time, affected URLs and measurements. I note the change, its release date and any other changes that could affect the result. An assessment date and an article’s publication date belong in different fields.
I choose the next review window in advance, allowing for reporting completeness, demand and the time search engines need to revisit changed pages. There is no fixed number of days that guarantees a reliable verdict. I also confirm the intended change is actually served before waiting for performance to move.
At the review, I rerun the same scope, check useful outcomes and compare with similar unchanged pages where that comparison is credible. More clicks after an update is an observation. It becomes stronger evidence when the intended queries improve, the comparison is fair and other explanations have been considered; it still does not prove the edit caused every change.
For a fuller measurement plan, use my guide to checking a content refresh.
Use the worksheet as a record of the decision
Download the blank CSV worksheet and import it into your spreadsheet tool. It contains column headings and one empty row, with no invented measurements or client information. Use one row per page and comparison. Leave unavailable measurements blank and explain them in the evidence-status and limitation columns.
Record the comparison dates and filters first, then the measurements, recommendation, reason, owner, approved action and review date. Keep a link to your saved evidence in your own private copy. Do not overwrite the original row when you reassess: save a dated copy or add a new row so the decision history remains understandable.
- Are both periods complete and meaningfully comparable?
- Have I marked missing or stale evidence and checked the normal page?
- Which audience or query change explains the loss?
- What useful traffic, information or links must I protect?
- Is the recommendation separate from an approved live action?
- Is any replacement ready, relevant and tested?
- Have I saved the baseline, owner and next review window?
If the findings point to a wider problem, I can help with the technical SEO and content review as well as the implementation. The useful outcome is a clear decision with evidence behind it, followed by a change we can check.