Case Study: Recovering 45,000 Monthly Visits After a Google Spam Update
· Royking Niba
A sports-media property lost a large share of its organic traffic to a Google spam update, and roughly 45,000 monthly organic visits came back over a window that ran from August to October. The work was not clever. It was classification first, so that I knew I was dealing with an algorithmic loss and not a manual action, then cutting the manufactured link layer and the thin pages it existed to prop up, then leaving the site almost entirely alone for weeks while Google recrawled and reassessed. The restraint was the hard part and the part that made the difference.
I cannot name the property, the update or the URLs, and I am not going to invent a traffic graph to make the story tidier. What I can give you is the sequence of decisions in the order I made them, including the ones I refused to make. If you are looking for evidence that recovery from a spam update is possible, the sequence is the evidence. The number is just the receipt.
What the loss looked like on arrival
The shape of the drop told me more than the size of it. Traffic did not decay over months. It stepped down inside a short window and then sat flat at the new level, which is what an algorithmic reassessment looks like and what a technical fault or a seasonal dip does not.
Three things were true at the same time, and the combination is diagnostic. The site was still fully indexed, so nothing had been removed. Rankings had not vanished, they had slid, which means the pages were still eligible and were simply being valued less. And the loss was uneven: some sections of the site were barely touched while others were gutted. A site-wide problem produces site-wide damage. This was selective, and the selection itself was a clue about what the update had decided it did not like.
The owner arrived convinced he had been penalised in the punitive sense, that somebody at Google had looked at the site and pressed a button. That is almost never what has happened, and believing it sends people straight into the wrong remedy.
Ruling out a manual action before doing anything else
This is the first thirty seconds of the engagement and it decides the entire plan. Open the manual actions report in Search Console. If it is empty, there is no manual action, and no amount of intuition overrides that. This site’s report was empty.
The distinction is not academic, because the two problems have opposite exit routes. A manual action ends when a human reviewer accepts a reconsideration request. An algorithmic demotion ends when the system recrawls, reassesses and revalues the site, and there is nobody to appeal to. Filing a reconsideration request against an algorithmic loss achieves nothing except giving the client a false sense that a clock is running. I have written the full separation of the two in manual action versus algorithmic penalty, because getting this wrong wastes the first month of any recovery.
| Signal | Points to a manual action | Points to an algorithmic loss |
|---|---|---|
| Manual actions report | Names an issue and a scope | Empty, no matter how bad the drop |
| Indexation | Pages or the site deindexed | Everything still indexed, just ranked lower |
| Drop profile | Often abrupt and total for the affected scope | A step down, then a flat new level |
| Distribution of damage | Matches the stated scope of the action | Follows a content or link pattern you have to find |
| Route out | Fix, document, request reconsideration | Fix, wait for recrawl and reassessment |
I also checked the Search Status Dashboard and the Search Central blog to confirm the timing lined up with a ranking systems update rather than an indexing incident on Google’s side. That is a two-minute check that occasionally saves you from rebuilding a site that was never broken.
Segmenting the loss before touching anything
You cannot fix a site-level number. You fix the segments that produced it, and you can only do that once you know which segments those are.
So the first real piece of work was cutting the loss apart. I split performance by page template, by content type, by the age of the page, by whether a page had ever attracted a link that nobody paid for, and by query type, separating brand and navigational demand from head commercial terms and from long-tail informational demand. Then I compared each segment’s before and after against itself rather than against the site average.
The pattern that fell out was clean enough to plan against. The damage concentrated on a specific class of page and on a specific class of query, and the pages that survived had something in common that the pages that died did not. I will not pretend I can generalise that pattern to your site, because the whole point of segmenting is that the answer is local. What generalises is the refusal to act before the segmentation is done. Every mass action taken on day one, every bulk deletion and every blanket disavow, is a decision made without this information.
What the backlink profile and the content set actually showed
Two separate problems, one shared cause. The link profile carried a manufactured layer, meaning links that existed because somebody had arranged for them to exist, sitting alongside a genuine layer earned by the property’s actual sports coverage. The two layers were distinguishable, which matters, because the entire remedy depends on being able to separate them rather than treating the profile as one undifferentiated mass. The tests I use to draw that line are in toxic backlinks, and none of them are the toxicity score your tool is showing you.
The content set had the mirror-image problem. A body of pages existed to hold the anchors and to catch query variants, not to be read. Sports publishing makes this easy to do at scale and easy to justify to yourself, because fixtures, results, tables and tag archives generate near-identical pages by the thousand and each one looks defensible in isolation. Read Google’s spam policies next to a site like that and the match is not subtle. The manufactured link layer and the manufactured page layer were the same strategy expressed twice.
The decisions, in order
Order matters more than speed. Doing the right things in the wrong sequence produces a site that changes constantly and never gets assessed in a stable state.
- Stop the input. Before anything is cleaned, the activity that created the problem stops completely. Cleaning a profile while acquisition continues is not a recovery, it is a treadmill, and it also destroys your ability to read whether anything you did worked.
- Inventory and classify the link layer domain by domain, into manufactured, genuine and unclear. The unclear pile is left alone. Guessing in either direction costs you more than it gains.
- Pursue removals first wherever a real contact existed on the linking side. Removal is the honest fix. It is slow, most requests go unanswered, and it is still worth doing for the ones that land.
- Disavow the remainder at domain level, using the disavow links tool, scoped strictly to the manufactured layer and nothing else. My full reasoning on scope, domain versus URL entries, and why most disavow files are far too large is in the disavow file explained.
- Sort the content set into keep, consolidate and remove. Keep is anything with genuine demand and a reason to exist. Consolidate is the near-duplicate mass. Remove is the pages that only ever existed to hold a link or a keyword.
- Consolidate with redirects, not deletions, following Google’s guidance on consolidating duplicate URLs, so that whatever equity those pages held lands on the surviving page instead of evaporating.
- Fix the template that produced the mass. This is the step people skip. If auto-generated pages caused the problem, removing this batch without changing the generator means the site rebuilds the problem on its own within a year.
- Rebuild internal linking around the survivors, so that crawl priority and internal authority point at the pages that are meant to carry the site rather than at an archive that no longer exists.
- Stop. Then hold the site steady and let it be recrawled.
What I deliberately did not do
This is the interesting half of the case, and it is where most recoveries are lost. Every item below was proposed, by the client or by somebody advising him, and every one was declined for a reason.
| Not done | Why not |
|---|---|
| Reconsideration request | There was no manual action. There is nothing and nobody to reconsider. |
| Disavow the whole profile | It would have removed the genuine links holding up the sections that survived, turning a partial loss into a total one. |
| Delete the archive wholesale | Bulk deletion throws away the pages that were still earning, and it produces a crawl signal of chaos at exactly the moment you want stability. |
| Mass rewrite everything | Rewriting pages that should not exist is expensive theatre. The thin pages needed removing, not improving. |
| Publish heavily to show freshness | Volume was the original problem. More of it is not a defence, and Google’s guidance on helpful content is explicit that content made for search engines first is the thing being demoted. |
| Chase new links during the window | Acquiring links while a link-driven demotion is being reassessed muddies the only clean signal you have. |
| Redesign, or treat Core Web Vitals as the fix | Performance work is real work, but it was not why the site lost rankings, and doing it here would have changed hundreds of variables mid-diagnosis. |
| Migrate the domain | The escape hatch that is not one. It carries the profile with it, and it discards the genuine authority that made recovery possible. |
The waiting problem
Once the work is done, nothing visible happens for weeks, and this is where clients fire their consultant and start over with someone who promises faster. The delay is structural. Google has to recrawl the changed URLs, process the redirects, pick up the disavow file on its own schedule, and then reassess the site under a ranking system that reevaluates on its own cadence rather than yours. None of those steps report progress to you.
The way through it is to agree, before the silence starts, on what counts as evidence during it. I set the expectation at the outset that traffic is the last thing to move, not the first, and that judging the work by traffic in week three is judging a photograph before it is developed. Then I report weekly on the things that do move: crawl activity on the changed URLs, the redirect chain resolving, the shrinking count of indexed thin pages, impressions before clicks, and average position drifting within page one before any click volume follows it. I send that note in the weeks when nothing moved as well, and I say plainly that nothing moved. A client who hears from you only when there is good news will assume the worst during every silence, and he will be right to.
The other half is honesty about the ceiling. I told him at the start that I did not know whether the site would return to its old number, that some of the old number had been bought and would not come back, and that the goal was to recover the part of it that had been earned. That conversation is uncomfortable once. Promising a full restoration is uncomfortable every week for three months and then terminal.
What came back, and what did not
Roughly 45,000 monthly organic visits came back across the August to October window. That is the number I stand behind and I am not going to dress it up with a percentage or a curve.
It did not arrive as a switch flipping. Recovery from an algorithmic demotion is granular: individual queries recover at different times and to different positions, and the site climbs in uneven steps as sections are reassessed rather than in one clean line. Pages with real demand and a genuine reason to exist recovered. Pages that had only ever ranked because of the manufactured layer did not recover, and were not supposed to, because most of them no longer existed. Some queries came back at lower positions than before and stayed there, which is the honest arithmetic of removing an advantage the site should not have had.
What the site got in exchange is a profile and a page set it can defend at the next update. That is the part clients undervalue at the time and appreciate a year later.
What generalises to your site, and what does not
Four things transfer to any spam update recovery I have run, in sports media or anywhere else. Classify the loss before you plan the remedy, because the manual and algorithmic routes out are different and mutually useless. Stop the input before you clean anything. Segment the damage instead of acting site-wide, since blanket actions are how partial losses become total ones. And change the system that manufactured the problem, not just its current output.
Three things do not transfer. The timeline is not a rule: this ran August to October, but recovery is gated by recrawl and by when the relevant system next reevaluates, and neither of those is on your calendar. The outcome depends entirely on what is left underneath. This property had a genuine audience, genuine editorial output and a genuine link layer sitting under the manufactured one, so there was something for the recovery to land on. A site that is nothing but the manufactured layer has no recovery available to it, only a rebuild, and anybody telling that owner otherwise is selling him a retainer. And a manual action needs the documented, appealed path rather than this one.
If you are at the start of this and still working out which kind of loss you are looking at, begin with the diagnostic sequence in my guide to Google penalty recovery. Do the classification properly and the rest of the plan writes itself. Do it carelessly and you will spend your first month solving the wrong problem, which is the most expensive thing you can do with a demotion.