The HCU survival guide

In one week, one of my 200 WordPress sites lost 40% of its search traffic. There was no manual action to appeal, no warning in Search Console, no email. The graph bent, dropped, and stayed down. The immediate question was how many of the other 199 sites were about to do the same.

If you're reading this because one of your graphs just did the same thing, you don't need a definition of the Helpful Content Update. You need to know whether it was actually a helpfulness hit, what to do with every article on the site, and how long the road back really is. The short, unhappy version: it was probably broader than one article, the fix is triage rather than tweaks, and the timeline is measured in update cycles, not weeks.

A short history, because the name is now wrong

The fact that changes your planning comes first: there is no "next HCU" on a calendar. You're not waiting out one bad update. You're being continuously scored. Google retired the standalone Helpful Content Update in March 2024 and folded its signals into the core ranking systems, alongside a new spam policy for scaled content abuse: publishing large volumes of unreviewed, unoriginal content at speed. What's left is core updates, several a year, each carrying helpfulness signals inside them.

The compressed history of how it got here: the HCU shipped in August 2022 as a sitewide signal, a classifier that scored whole sites on whether their content seemed written for people or for search engines, and made everything on a flagged site, not just the unhelpful pages, less likely to perform well. A second pass rolled out in December 2022. The one operators still talk about is September 2023: the update that cut small niche and affiliate sites in half or worse, in many documented cases 70–90% down, with very few full recoveries since. Operators still call a helpfulness-driven drop "an HCU hit" (the May 2026 update is just the latest to earn the label) and the shorthand is fine, as long as the diagnosis treats it as what it now is: a signal inside every core update, not a named penalty you can wait out.

How to tell if you were hit

Before touching a single article, confirm the diagnosis. Much of the "HCU recovery" work operators do is aimed at the wrong problem. Work through this in Search Console, in order:

• Check the panels that name their problem first. Manual actions and security issues each have a dedicated report in Search Console. If either has an entry, you don't have an HCU problem. You have that problem, and it comes with instructions.
• Line the drop date up against a confirmed rollout. Google publishes update start and end dates on its Search Status Dashboard. A helpfulness reassessment lands inside a rollout window. A drop on a random quiet Tuesday points at something you changed: a migration, a redesign, a robots.txt edit, a plugin that started noindexing archives.
• Look at the shape of the decline. The signature of a sitewide quality reassessment is many URLs each sliding a few positions, not one page falling off a cliff. Sort the pages report by click delta over the drop window: if the pain is spread across dozens of URLs, that's the pattern. One page collapsing is usually a SERP feature change or a competitor, not a classifier.
• Compare impressions, clicks, and position separately. The typical helpfulness hit keeps you indexed and visible (impressions often hold or dip mildly) while average position slides a page or more and clicks fall off the front-page cliff. If impressions went to zero, look at indexing, not quality.
• Segment brand from non-brand queries. People searching your name still find you; a quality reassessment shows up as non-brand queries sliding while brand holds steady. If brand queries dropped too, something structural is wrong.
• Check whether losses cluster by topic. If one topic cluster took the hit while the rest of the site held, Google may have reassessed your authority on that topic specifically, which changes the triage: the problem set is that cluster, not the whole site.
• Rule out seasonality with a year-over-year view. Month-over-month graphs manufacture emergencies every January and July. Compare against the same weeks last year before concluding anything.

What helpfulness systems actually target

Google's public guidance is a list of gentle questions: "would someone find this helpful?" The operator translation is blunter. Looking at what actually got hit across the update era, the pattern is consistent:

• Search-first content. Pages that exist because a keyword tool said the phrase gets traffic, written by someone with nothing to add about it. The tell: the article answers its title and contains no information you couldn't assemble from the other nine results.
• Coverage without expertise. Sites that sprawled into adjacent topics because they could rank, not because they knew anything: the outdoor-gear site with a cluster of insurance articles. Sitewide scoring makes this sprawl expensive: your off-topic filler taxes your on-topic winners.
• Volume as a strategy. Publishing faster than anyone could plausibly review is now its own spam policy (scaled content abuse), and it was the underlying pattern in most of the September 2023 casualties. The classifier doesn't care whether the volume came from freelancers or a model.
• The ratio, not just the pages. This is the part the SEO-blog version undersells: some signals still apply site-wide. A site that's 30% genuinely useful and 70% filler doesn't get 30% of its traffic protected. The filler drags on the whole domain, which is why triage, not spot-fixing, is the recovery unit.

What it does not target: AI-assisted writing as a method. Google has said this explicitly, and it matches what got hit and what didn't. Unreviewed generic output at scale is the losing pattern. Reviewed, specific, genuinely useful content survives regardless of what drafted it. The distinction that matters isn't human-vs-AI: it's reviewed and useful versus shipped and hoped.

The audit: finding the problem across a whole site

Recovery starts with an inventory, because the unit of assessment is the site, not the article. On one site this is a hard weekend. Across a portfolio it's not manually possible, which is why I ended up building tooling for it, but the method is the same either way:

• Pull every indexed URL with its performance history. Search Console gives you 16 months of clicks, impressions, and position per page. That window matters: you want each page's trajectory before and after the drop, not a snapshot.
• Score every article on two axes: performance and quality. Performance is mechanical: trending up, flat, decaying, or dead. Quality needs an honest rubric: Does this page contain anything not already in the top ten results? Does it show any evidence someone with real experience wrote it? Is the first screen an answer, or a padded intro built to hold a keyword? Word count is a weak signal in both directions; distinctiveness is the real one.
• The intersection is the risk list. Thin and decaying is the delete pile. Thin but still performing goes on a watchlist: it's living on borrowed time. Genuinely good but decaying is the refresh pile, and often the highest-ROI work on the site. Good and performing gets left alone.
• Apply the one-question test to everything borderline: if this page disappeared tomorrow, would anyone notice? Not "would the traffic graph notice": would a person be worse off. Across a portfolio, the honest answer for a painful share of pages is no. That share is your problem, stated as a number.

The triage: delete, refresh, or keep

Every flagged article gets exactly one of three calls. The discipline is in not inventing a fourth category called "deal with it later."

• Delete - it exists to target a keyword and adds nothing a dozen pages don't already say. Serve a 410 or at minimum noindex it. Redirect only where a genuinely equivalent page exists; blanket-301ing deleted content to the homepage just converts a quality problem into a soft-404 problem. Deleting feels like burning money you already spent. It isn't. The money was spent either way. The only question is whether the page keeps taxing the rest of the site.
• Refresh - the topic is real and the angle is yours, but the execution was thin, outdated, or rushed. A refresh is a rewrite for the reader: answer in the first paragraph, cut the padded intro, add the specifics only you have, update what's stale. It is not: a new date, a synonym pass, three added FAQs, and an author box bolted onto unchanged text. That cluster of tricks is exactly the search-first pattern the classifier was built to catch.
• Keep - it's doing its job. The hardest discipline on this list is leaving winners alone. Mid-recovery, every instinct says "optimize everything," and tinkering with pages that still rank is how you turn one problem into two.

Expect the delete pile to be bigger than feels comfortable. The sunk-cost objection arrives on schedule: every one of those articles cost real money or real hours. But keeping a mediocre page because it was expensive to make is precisely the reasoning that built the liability. The portfolio version of this mistake is keeping a whole mediocre site for the same reason.

Rebuilding: what the road back looks like

The honest news first: full recoveries from major helpfulness hits are rare, and none of them are fast. Sitewide reassessment mostly happens when Google rolls out subsequent updates, typically months apart, so the feedback loop is slow and not yours to schedule. Plan for the graph to ignore your work for a while, then move in steps rather than a slope.

What separated sites that came back from sites that didn't, in the pattern that's now well documented across the update era:

• Pruning alone didn't do it. Deleting the filler stops the bleeding; it doesn't rebuild trust. Sites that deleted a third of their pages and changed nothing else mostly stayed flat.
• The ratio had to actually change. Recovery correlated with the site becoming genuinely more useful on average: fewer pages, better pages, plus new content that demonstrated real expertise, published at a pace review could keep up with.
• Cosmetics did nothing. Date bumping, title swaps, author boxes on unchanged text, "E-E-A-T sprinkles": the classifier reads content, not costume.
• Consolidation punched above its weight. Merging five overlapping thin articles into one genuinely complete page (with redirects from the four losers) reliably outperformed five parallel refreshes of the same pages.

Set expectations accordingly: months of work ahead of a reassessment you can't schedule. It's a bad deal. It's also the only one on the table, which is the strongest argument that exists for not needing this guide twice.

How to not get hit again

Every part of the recovery has a preventive mirror, and the preventive version is cheaper by an order of magnitude:

• Gate publishing. The triage you just did manually, done before publish instead of after a penalty: nothing ships without clearing an originality and usefulness bar. Publish at the speed of review, not the speed of generation: velocity you can't review is velocity you'll delete later.
• Watch the leading indicators continuously. A 40% drop is the last signal, not the first. Indexing status and ranking positions move first, page by page, and they're checkable daily by machine. The difference between finding out in 24 hours and finding out from the monthly traffic graph is the difference between fixing three articles and re-triaging a site.
• Keep the inventory alive. The audit isn't a one-time recovery artifact; it's the operating picture. Every site accumulates decay: the watchlist from your audit is next quarter's refresh list if you keep it current, and next year's crisis if you don't.
• Mind the weakest site. Sitewide scoring means every domain carries its own ratio, and a portfolio means you're running many ratios at once. The site you've been ignoring isn't neutral: it's a liability you're choosing. Running that discipline across many sites is its own problem, which is what the portfolio playbook covers.

FAQ