Manual Actions Versus a Bad Week: A Google Manual Action Recovery Playbook

A traffic drop that wasn’t what it looked like

A site I was helping with lost a bit over 60 percent of its organic traffic across four days. The person running it was completely certain they’d been penalised. By the time they called me, they’d already started an emergency link audit.

They hadn’t been penalised at all. What happened was an algorithmic update, and the work they’d already started was aimed at a problem the site didn’t have.

That distinction is the most useful thing I can offer on this topic. A manual action and an algorithmic drop feel identical from inside the business, and they need completely different responses. Let’s separate them properly, starting with the one that’s easy to check.

Manual action or algorithm: why the difference matters

A manual action means a person at the search engine looked at your site, decided something on it broke the guidelines, and applied a penalty by hand. The property that matters here, stated as plainly as I can: manual actions are always disclosed to you. They show up in Search Console, in a report called Manual Actions, with a description of what was found and how much of the site it applies to.

An algorithmic drop has no such notification, because no person is involved. Rankings shift because the system evaluating your pages changed, or because it re-evaluated your pages against criteria that were always there.

Check the manual actions report first

The first move in any traffic emergency takes about 15 seconds. Open Search Console, open the Manual Actions report, and look. If it says no issues detected, there is no manual action on the site. The report is authoritative on that point, which is exactly why it belongs first in the sequence rather than somewhere in the middle. Every hour spent auditing links after seeing an empty Manual Actions report is an hour spent on a theory the evidence already ruled out. I’ve watched people spend a fortnight there.

Reading the report: problem type and scope

If there is something in the report, read the exact wording. It tells you two things that shape everything you do next.

The first is the problem type: unnatural links pointing at your site, unnatural links going out from your site, thin content with little or no added value, user generated spam (usually a comment section or forum nobody’s moderating), cloaking or sneaky redirects, hacked content, spammy structured data markup, or pure spam.

The second is the scope, described as either site-wide or partial. Partial matches name the affected section, which narrows the work enormously. A partial match on a subfolder is a very different job than a site-wide one.

Fixing unnatural links pointing at your site

This is the most common one people actually hit, so it’s worth walking through in full.

Pull your link data from as many sources as you can get, because no single tool sees everything. Start with Search Console’s own export, since that’s closest to what the search engine sees, then add whatever third-party index you have access to. Deduplicate by referring domain and start classifying.

What you’re looking for is links you or someone working for you caused to exist: paid placements, mass directory submissions, comment and forum links dropped at scale, a private network someone sold you, or links from a site you also own that were built specifically to pass authority. The test is intent, not quality. A genuinely low quality site linking to you organically isn’t the problem here.

For each one, try removal first. Contact the site, ask for the link to come down, and keep a record of every attempt, including the date and the address you wrote to. A lot of those emails will go nowhere, and that’s expected. The record matters as much as the outcome.

What you can’t get removed, disavow. Build the file, submit it, and be conservative about domains you’re unsure of. A disavow file is appropriate here, in a manual action for unnatural links, which isn’t true in most other situations.

Hacked content: the penalty is not your biggest problem

If your manual action is for hacked content, the penalty is the least of your worries. You have an active compromise, and cleaning the visible spam pages while leaving the way in open means it comes straight back, usually within days.

The order matters. Find and close the entry point first. Rotate every credential, including hosting, database, and any keys in config files. Then remove the injected content. Then update whatever was out of date. Only after all of that do you request reconsideration. I’ve seen sites go through three rejected requests because they kept cleaning the symptom instead of the cause. Also check for injected content that only appears to crawlers, since that’s the entire point of the technique, and it’s invisible when you browse the site normally.

Thin content: the slowest one to clear

Thin content is the vaguest category and the one people argue about most. In practice it tends to land on sites running large numbers of near-identical pages generated from a template, doorway pages built to catch location or product variations, scraped or lightly reworded material, and affiliate pages that add nothing to what the merchant already published.

The fix is genuinely substantive work rather than a technical change, which is why this is the manual action people take longest to clear. If a page exists only because a keyword existed, that’s the pile to start with.

Writing a reconsideration request that works

This is where people undo their own work. A reconsideration request is read by a person. It needs to answer three things clearly: what caused the problem, what you specifically did about it, and what has changed so it won’t happen again.

Be concrete. Give numbers. We identified this many problem domains, contacted this many, got this many removed, disavowed the remainder, ended the arrangement with the agency that built them, and documented the link policy going forward. The record of removal attempts is the evidence that you did the work, rather than just filing a disavow and hoping.

Be honest about how it happened. If you bought links, say you bought links. Reviewers see a great many of these, and a request that dances around the cause reads exactly like what it is. Straightforward admissions clear faster than elaborate explanations, consistently enough to bet on.

Don’t submit until the work is genuinely finished. A rejected request costs you weeks, and repeated thin submissions do you no favours with the person reading them. Timelines run from a few days to a couple of months, and a first-attempt rejection is common enough that you should plan for it rather than be devastated by it.

When it’s not a manual action: algorithmic drops

Telling the two apart, beyond the Manual Actions report, comes down to pattern. Check your drop date against the publicly announced update windows, which the search engines do announce and the industry tracks closely. Check whether the loss is spread across everything or concentrated in one type of query. Check whether the sites that took your positions are a coherent group, because a broad reshuffle where several competitors also moved looks very different from you alone falling.

Recovery works differently here, and it’s worth being straight about it. There’s no request to submit and no switch to flip. You improve the site against whatever the update appeared to reward, and you wait for reassessment, which typically means waiting for a subsequent update. That can take months. Nobody can tell you what any particular update measured, and anyone claiming to know the internals is guessing with confidence.

Some sites don’t come back

The harder truth is that some sites don’t come back. I’ve watched people spend two years rewriting a property that was never going to recover, when the same effort spent on a new build would have got them further. Knowing when to stop is a real skill, and nobody teaches it. The question worth asking is simple: if this domain didn’t exist and you were starting today with the same budget, would you build it again. The answer is often no, and it’s often been no for a year.

Check the boring explanations first

Before either diagnosis, check the boring explanations, because they’re correct far more often than the dramatic ones.

Did your analytics tracking break. A script that stopped firing after a deploy produces a graph identical to a catastrophe. Did someone push a noindex tag live from a staging environment, which is the single most common self-inflicted disaster in this business. Did robots.txt change. Did a redesign alter URLs without redirects. Did you lose a featured snippet that was carrying a large share of clicks. Is it simply the season, and were you looking at the same week last year.

Check the site in Search Console’s URL inspection tool. Check that pages still return a 200. Check that the site is still indexed by searching for a distinctive phrase from one of your pages. Five minutes of that saves a fortnight of the wrong work.

When the ranking doesn’t move but the traffic still does

A site can have no manual action, no algorithmic issue, and still lose a lot of traffic because the results page changed shape around a query. An answer box appearing above your position, or a new set of shopping results, moves your traffic without moving your ranking at all. Your position report looks fine and your clicks are gone. The performance report shows this if you compare impressions against clicks rather than looking at position alone.

What to do while you wait

Whichever case you’re in, the waiting is where people damage things. Keep publishing. Don’t panic-delete pages on a hunch. Don’t rip out your internal linking. Don’t change your URL structure. Every one of those adds a variable, and when the traffic does move, you’ll have no idea which change moved it. Do the work you’d have done anyway if none of this had happened.

Talking to people who don’t do SEO

If you’re explaining this to a boss or a client, give them the diagnosis and the timeline separately. The diagnosis is usually available within a day. The timeline almost never is, and pretending otherwise to sound in control is how you end up having a much worse conversation six weeks later. Tell people what you know, what you don’t, and when you’ll next have something to report. That lands better than confidence does.

The ten-minute quarterly check

Prevention is dull and it works. Once a quarter, open the Manual Actions report, glance at the links pointing at your site for anything you didn’t expect, and check that nothing has appeared in your index that you didn’t publish. It takes ten minutes, and it turns most of this into a thing that happens to other people.

The order of operations

Check tracking. Check indexation. Check the Manual Actions report. Check the date against known updates. Only then start theorising. Running that sequence backwards costs people weeks, every time.

If you want more breakdowns like this one, the rest of our SEO playbooks live at The SEO Desk.