contentEffort: the Google leak and content that is hard to copy

In May 2024, thousands of pages of Google’s internal documentation ended up online: the description of the “Content Warehouse” API, in other words the list of data the engine stores about every page. Inbound links, anchors, clicks, freshness: much of it was expected. One attribute went almost unnoticed, though, and it deserves a closer look: contentEffort.

What the documentation actually says

The attribute sits in a module dedicated to a site’s quality signals, next to a similar score designed for forums and discussion pages. The documentation gives no formula and no weighting: only a name and a sentence. So we know that a language model reads article pages to estimate the effort they required. We don’t know on what criteria, with what weight, or whether it was still in use at the time of the leak. The documents were made public by Rand Fishkin and Mike King, and the documentation page that describes it is still available.

Everything that follows is therefore an interpretation, not a revelation. But it is an interpretation that points in the same direction as what Google has said publicly since 2022.

The link with “helpful content”

The Helpful Content System, which Google folded into its core algorithm in March 2024, aims to favour pages written for people rather than for the engine. Its public guidelines ask simple questions: does the page provide original information, analysis or research? Does it go beyond the obvious? Does it show first-hand experience of the topic?

An attribute that estimates effort with an LLM looks like a way to automate these questions at the scale of the web. The irony of our times: a language model judges whether a page could have been written by a language model. And as we all know, the engine still struggles to tell the difference: hollow pages still rank very well.

The real question: can your page be copied in an hour?

“Effort” is a misleading word. A 4,000-word page that rephrases the top ten results took time, but it contains nothing someone else couldn’t redo. Conversely, one paragraph giving the real duration of a migration observed at a client is worth more than three pages of generalities.

The right test fits in one sentence: if a competitor, or an AI given the topic, can produce the same thing in an hour, why would the engine put it forward? Now that anyone can generate an article in seconds, generic content costs nothing. So it is not worth much either.

What makes content hard to copy

Three things, which no tool can invent for you.

Data only you have. Your average lead times, your response rates, what a project really costs with you, what your clients ask for most often. An in-house figure, even a modest one, can’t be found anywhere else. On one condition: it must be true, and its origin must be known.

Lived experience. What happened when the work was done: the step that took three times longer than planned, the mistake a client learned the hard way, the screenshot taken on the day. An AI can summarise a topic; it can’t tell the story of your project.

A clear point of view. Saying when a solution doesn’t fit, who you would advise against it, what you would choose instead. The pages that all look alike are the ones that take no risk. A reasoned opinion, even a debatable one, is unique by nature.

AI remains useful in this work, to structure, proofread, translate, check a source. What it doesn’t replace is the raw material: what you know and nobody else has written. It’s the same logic as for AI-produced content that must be reviewed, sourced and signed.

AI helps, provided you have a method

Asking “write me an article” creates no value. For Jim, who has watched for fifteen years what fellow agencies do and what the market rewards, the difference lies in the method: challenge the topic before writing, challenge the AI too when it answers too fast, go through a real research step, check every source when there are any, and keep a critical, human view from start to finish. Without that questioning, you end up with declarative content: correct sentences that anyone could have written.

That’s also where experience comes in: past cases, what you’ve seen work or fail elsewhere, what you think of how the market is evolving on the topic. No model has that perspective.

How to spot your pages that are too easy to copy

Few companies have ever measured the effect of thin content on their traffic, often because nobody looked. Three checks are enough to start.

In Search Console, pages that are shown but not clicked. A page that gets impressions on its queries but a much lower click-through rate than your other pages, or that slowly slips after each major Google update, is often an interchangeable page. Note the dates of the updates Google announces and compare clicks over the four weeks before and after.

The top ten test. For each important page, open the top ten results for its main query and list what your page says that none of them say. If the list is empty, the page can be copied.

The AI test. Give the page’s topic to an AI assistant, with nothing else, and compare. If its answer covers most of your page, you know what to enrich: an in-house figure, a case, an opinion.

These three checks give you a list of pages to enrich, merge or remove, and a starting point for measuring: take the same figures again three months after reworking them.

What it changes for a team that publishes

Publish less, but publish what only you can write. Before starting an article, one question is enough: what will this page contain that isn’t already in the top ten results? If the answer is “nothing”, better pick another topic, or wait until you have the data or the field experience that will make the difference.

It’s also the first thing we look at in an audit: not just the technical side, but the share of pages a competitor could rebuild identically. Our SEO audit starts there, and says frankly which pages are worth enriching and which can be merged or removed.

Sources

1. Documentation page that describes it (new tab)hexdocs.pm