SEO Recovery
Before anything can be recovered, somebody has to diagnose it
A fall in organic traffic has at least five very different causes, and they produce almost identical graphs. Treatment before diagnosis is how businesses spend a quarter fixing something that was never broken. We diagnose first, and we say when a decline is not reversible.
Presenting symptoms
The graph looks the same whatever caused it
People describe a decline in one of a handful of ways. The description narrows the possibilities more than most people expect, which is why the first conversation is about the shape of the fall rather than about tactics.
- It fell off a cliff on one specific day.
- A single-day collapse points towards something discrete: a deployment, a robots rule, a certificate failure, a manual action, a tag removed, or a domain change. Gradual causes rarely produce a vertical line. This is the most diagnosable shape and often the most fixable.
- It has been sliding for eight months.
- A slow slide points towards competitive displacement, ageing content, an accumulation of small technical decisions, or a category where demand itself has moved. It is the hardest shape to attribute and the one most often blamed on the wrong thing.
- One section fell and the rest is fine.
- Localised declines are the most useful of all, because the boundary tells you where to look. A blog that fell while the product pages held, or one language that fell while the others held, narrows the search dramatically.
- Impressions are steady but clicks are down.
- This is almost never a ranking problem. It points at what surrounds your result: more advertising, a features block, an answer given directly on the page, or a title and description that stopped matching what people wanted.
- The tracking tool says nothing changed.
- Which means either the measurement is wrong or the tracked keywords are not the ones that were producing the traffic. Both are common, and both are checked before anything is concluded about the site itself.
Differential diagnosis
Five causes that produce similar graphs
Recovery work goes wrong when a proposal is written for one of these while the site is suffering from another. The evidence differs even where the symptom does not.
| Dimension | Evidence that points to it | How quickly it moves | What is realistic |
|---|---|---|---|
| Manual action | A notice in the manual actions report. It is either there or it is not — there is no inference involved. | Nothing changes until a reconsideration request is reviewed, which takes as long as it takes. | Reversible in principle, provided the underlying issue is genuinely removed rather than hidden. |
| Algorithmic change | Timing near a broad update, movement across many pages at once, and competitors moving in the opposite direction. | Positions can move again at any point, and often not until a subsequent update. | No fixed route back. The work is to make the pages genuinely better for the people arriving on them. |
| Self-inflicted | A deployment, migration, redesign, template change, plugin update or robots rule dated to the fall. | Usually recovers within weeks of the specific defect being corrected. | The most reversible category, and the reason dating the decline precisely comes first. |
| Competitive displacement | Your pages hold their content quality but specific competitors now occupy the positions you held. | Slow in both directions. It arrived over months and it leaves over months. | Winnable in some markets, not in others. Sometimes the honest answer is to compete elsewhere. |
| Measurement artefact | The site is fine, rankings are unchanged, and the fall is confined to one report, one property or one date range. | Immediate, once the tagging, filter, consent configuration or property setup is corrected. | Nothing was ever lost. This possibility is checked first precisely because it is so cheap to rule out. |
Method
How the investigation actually runs
Validate the measurement
Before anything else. Tag presence, consent configuration, property and view settings, filters, bot exclusion, subdomain coverage, and whether the reporting date range does what the person reading it assumes.
You get: A written confirmation that the decline is real
Date the decline precisely
To the day where possible. Then lay the change log alongside it: deployments, content edits, migrations, campaign changes, announced updates, and seasonal patterns from previous years.
You get: A timeline with the fall and every candidate event on it
Segment until the boundary appears
By template, page group, query group, device, country and landing page. A decline that is everywhere means one thing; a decline with a clean boundary around it means something quite different and much more useful.
You get: Segment analysis identifying what is and is not affected
Rule out the things that can be ruled out
Manual actions and security issues, indexing and crawl status, response codes, canonical and robots directives, and whether the affected pages are still being served and rendered as intended.
You get: A list of causes eliminated with the evidence for each
Look outward at the result pages
What the results for the affected queries look like now compared with before: who is present, how much of the page is taken up by features and advertising, and whether the question is now answered without a click.
You get: A read on whether the opportunity itself has changed
Write the diagnosis and say what is worth doing
The most probable cause, the evidence for and against it, what we would do about it, in what order, and where we think the effort will not repay itself.
You get: A diagnosis document with a prioritised plan or an explicit recommendation to stop
Terms of engagement
What we commit to, and what we refuse to say
We will
- Check that the decline is real before proposing anything at all
- Show the evidence for the diagnosis, including the evidence that argues against it
- Tell you when the most likely cause is something your team did
- Recommend against spending when we do not think the return is there
- Separate what is knowable from what is inference, every time we present a finding
We will not
- Promise that traffic returns, because the outcome is not ours to decide
- Claim to know the contents of an algorithm update that has never been published
- Sell a retainer as the answer before the cause has been established
- Submit a disavow file as a reflex because links are the easiest thing to blame
- Present a recovery that coincided with a subsequent update as proof that our work caused it
The uncomfortable part
Some declines are not penalties and do not reverse
The word recovery implies that traffic is sitting somewhere waiting to be reclaimed. Often it is not. Search results change shape, and when they do, the level of traffic a page used to earn can simply stop existing.
The common versions are recognisable once you have seen a few. A query that used to send people to an article is now answered on the results page itself. A commercial category consolidates around two large aggregators who occupy the positions above you. Intent moves to a format you do not publish, so the demand still exists but it is going to video, or to a marketplace, or to somebody who built a free tool. None of these are punishments and none of them are undone by cleaning up your title tags.
This matters commercially, because chasing a level of traffic that is no longer available is expensive and demoralising. The right response is usually to accept the new ceiling for those queries and move the effort somewhere the demand still converts. We would rather tell you that in week two than bill you for a year of trying.
Questions
What people ask when the graph is falling
Can you guarantee our traffic comes back?
No, and we would be careful with anyone who says otherwise. Recovery depends entirely on the cause, and some causes have no route back.
What we can commit to is a diagnosis supported by evidence, an honest statement of what we think is achievable, and a recommendation not to spend where we do not believe the spending will pay.
How do you know it was an algorithm update?
Usually we do not know, and neither does anyone else. Updates are announced but their contents are not published, so attribution is inference from timing and pattern rather than fact.
What we can establish firmly is what was not the cause: a manual action either exists in Search Console or it does not, a deployment either happened that week or it did not, and measurement either changed or it did not. Ruling things out is the honest part of the work.
Our rankings look the same but traffic is down. What is happening?
That pattern usually means the result page changed rather than your position on it. More paid results, a features block, an answer shown directly, or an aggregator taking the top slots all reduce clicks at an unchanged position.
It can also mean your tracking tool is checking a location or device mix that no longer reflects your audience. We check both before drawing any conclusion.
What if we caused it ourselves?
That is one of the better outcomes, because self-inflicted damage is the most reversible kind. A redesign that dropped internal links, a template change that removed headings, a robots rule shipped by accident, a migration with broken redirects — all of these have specific fixes.
We will tell you plainly if that is what we find. There is no diplomatic version that helps you.
Is disavowing links part of this?
Rarely, and never reflexively. The disavow file is a tool for a specific situation — usually a manual action relating to links, or a genuinely unusual pattern of links you or someone acting for you created.
Used casually it can remove links that were helping. We would rather establish that links are the cause before touching them.
How long does the diagnosis take?
For most sites, one to three weeks. Validating measurement and dating the decline is quick; segmenting a large site and reviewing competitive movement across a lot of queries takes longer.
We report what we find even when the finding is that nothing is wrong with the site.
Get the diagnosis before you commission the cure
Tell us roughly when it started and what changed around that time. We will tell you what we would check first, what we think is most likely, and whether this needs an investigation at all.
If we don't deliver the work we agreed to deliver for reasons within our control, you don't pay for the undelivered work. Read our guarantee
Related
Where to go next
- the outcome this work servesThe problem framed from the business side rather than the diagnostic side.
- the causes explained in plain terms
- a scheduled audit rather than an investigationAn audit asks what could be better; a diagnosis asks what changed.
- moves handled before they cause damageSelf-inflicted declines usually start here.
- measurement you can rely on
- crawling and indexing diagnostics
Last updated · Reviewed by Zubair Afzal