HCU Classifier: Why One Bad Page Can Drag Down Your Who...

TL;DR. The Helpful Content classifier is no longer a discrete update. It runs continuously inside Google's core ranking system and emits a sitewide signal. If enough pages on your domain trip the unhelpful flag, even your good pages lose rankings. The fix is rarely "add more pages." It's prune, rewrite, and prove authorship.

What the HCU classifier actually does

When Google announced the Helpful Content Update in August 2022, it described a system that "better ensures people see original, helpful content written by people, for people, in search results, rather than content made primarily to gain search engine traffic." Two phrases in that sentence carry the whole load: sitewide and primarily to gain search engine traffic. The classifier scores pages individually, but the demotion it applies is to the whole site. If a domain accumulates enough pages that look written-for-search-engines instead of written-for-people, every page on the domain takes a haircut, including pages that score fine on their own.

In March 2024 Google folded the system into the core algorithm. There is no more "HCU rollout." It runs every time the core ranking system runs, which is continuously. There is no calendar to wait for.

What signals the classifier looks for

Google's published self-assessment is a checklist of about thirty questions. The questions that matter most for site owners cluster into five buckets.

Content depth. Does the page provide original information, reporting, or analysis? Does it go beyond what is obvious from competitors? Does it summarize others without adding value? Is it visibly mass-produced? Are you writing to an arbitrary word count because you heard Google has a preferred length? (Google explicitly calls this out as a warning sign.)

Expertise. Is there clear sourcing and evidence of who wrote it? Is the author someone who demonstrably knows the topic? Are facts verifiable? Section 2.5.2 of the Quality Rater Guidelines insists on clarity of authorship, which means a real name, a real bio, and ideally a Person schema block linking the author to LinkedIn, Wikidata, or ORCID.

Production quality. Is the page produced well or does it look sloppy? Are there spelling and stylistic errors? Are dates honestly maintained, or are you changing dateModified to look fresh when nothing changed? (Another explicit Google warning.)

People-first posture. Do you have a real audience for this content, or did you enter a niche purely because you thought it would rank? Does someone reading the page feel they got a satisfying answer, or do they immediately search again?

Trust apparatus. Does the site have a primary purpose? Is the entity behind it reachable? Are there visible policies, contact paths, and team pages?

A page that fails all five is obvious. The classifier's real bite comes from pages that fail two or three and don't realize it: thin affiliate roundups, programmatic location pages, AI drafts that nobody read, evergreen guides on topics outside the author's actual expertise.

What you can detect from the HTML alone

The shared engine that powers the Mega SEO Analyzer now runs a dedicated Helpful Content dimension on every audit. It surfaces ten HTML-detectable proxies for the classifier's signals.

Body word count. Under 300 words skews heavily into HCU-demoted corpora across public case studies. Under 600 usually lacks the depth the classifier rewards for competitive queries.
Author byline strength. Three signals checked: rel="author", Person schema with author reference, and a visible "By Firstname Lastname" pattern.
Date apparatus. Both datePublished and dateModified need to be present and honest. A dateModified that you bumped without changing the content is a documented anti-pattern.
First-party experience markers. Phrases like "I tested," "we measured," "in my experience," "photo I took." Google's question is, "Does your content clearly demonstrate first-hand expertise." If you can't say I somewhere on the page, you probably haven't.
AI boilerplate density. Fifteen-plus phrases over-represented in HCU-demoted corpora: "in today's fast-paced," "navigate the complexities," "delve into," "leverage cutting-edge," "unlock the potential," "harness the power," "plethora of," "testament to," "cornerstone of." The point isn't that AI wrote them. The point is that nobody edited them back into human language.
Affiliate density. Affiliate presence is fine. Affiliate links above 25% of all outbound links with no original reasoning ("why this vs the alternative") is a classic demotion pattern.
Site-purpose clarity. Links to /about and /contact from the page tell the classifier the entity is reachable. Their absence is one of the simpler trust-deficit signals.
Title vs H1 alignment. Sub-25% keyword overlap between the title tag and the H1 reads as clickbait. Google's self-assessment is specific: "Does the main heading or page title avoid exaggerating or being shocking in nature."
Citation surface. Outbound links to primary sources, wrapped in <cite> or <blockquote cite="…">. Both humans and AI retrievers reward verifiable claims.
AI-content disclosure. Optional. If present, it reads as transparency. If absent on a page that's clearly AI-generated, it reads as cover-up.

None of these are individually disqualifying. They stack.

How to recover if you've been demoted

A few things to understand first. Recovery is slow. Public case studies in 2024 and 2025 put successful HCU recovery at six to twelve months, and only after the site removed or substantively rewrote the offending pages. Surface-patching (changing a few phrases) has been documented to prolong demotion because the classifier retrains and the patches don't change the underlying pattern.

The recovery playbook that has actually worked for sites that came back:

Inventory the thin and the off-topic. Pull every page under 500 words. Pull every page where the topic sits outside the site's primary focus. Score each one: keep, expand, or remove.
Remove with redirects, not deletions. Pages you're killing should 301 to the closest related stronger page. Naked deletions leave broken links and confused retrievers.
Rewrite the keepers with first-party content. Add the screenshot you took. Add the measurement you ran. Add the quote from the customer. Replace the AI boilerplate with how it actually went.
Fix authorship globally. Every page gets a real byline, a real bio link, and a Person schema block with sameAs to wherever the author has a verifiable identity.
Honest dates only. Stop bumping dateModified on cosmetic changes. The classifier almost certainly checks for date inflation against actual content drift.
Wait. The classifier reassesses periodically. There is no resubmit button. IndexNow pings help with discovery of the changes; they do not accelerate the verdict.

When to run the HCU check

After any content batch ships, especially if AI assisted the drafting.
After a redesign or CMS migration that touched author markup or schema.
Quarterly as routine hygiene.
The first thing when traffic drops and the timing roughly correlates with a core update.

You can run it on a single URL from the HCU Pattern Detector, or get it folded into the seven other dimensions in the Mega SEO Analyzer. The Mega Analyzer treats HCU as a top-tier dimension because, unlike most ranking factors, a fail here drags every other page on the same domain.

Fact-check notes and sources

Google: Creating helpful, reliable, people-first content — full self-assessment question list
Google: Helpful content and Google Search results FAQ — March 2024 "folded into core ranking" announcement
Google: August 2022 Helpful Content Update announcement — original sitewide-classifier description
Google: Search Quality Rater Guidelines (PDF) — Section 2.5.2 authorship requirements

This post is informational and not a substitute for professional SEO consulting. Recovery from a Helpful Content demotion is rarely fast and is never guaranteed. No affiliation with Google is implied.

The HCU Classifier: Why One Bad Page Can Drag Down Your Whole Site