Content Effort: Google’s Quality Criterion, Mapped

The first time Content Effort mattered to me, it showed up as a calendar invite with no agenda. We had shipped a category roundup: two source interviews, a designer, an editor who pushed back, then a model that turned the research doc into about 2,400 words plus a comparison table. It settled at position fourteen, behind three URLs that read like the same URL in different fonts. Our head of content asked me, reasonably, what happened. I said internal links, then freshness, then site authority, all plausible, none of them the reason, and none of them what more than 10,000 quality raters are actually asked to look at. The reason had been sitting on the page the whole time. The interviews had happened. They just were not visible.

So I did what most of us did in early 2025 and read the coverage first. Search Engine Land’s report on the January revision of the Search Quality Rater Guidelines documented a new Lowest rating path for Main Content created with little to no effort, originality, and added value. Then I opened the Search Quality Evaluator Guidelines myself, mostly because I wanted to quote it correctly in a client deck. I decided in about four minutes that this was E-E-A-T 2.0, and I spent the next week telling people that. Then I stopped, because it is not, and anyone who knows the handbook would have corrected me in nine seconds. More than 10,000 raters work from that document, and the September 2025 revision left the effort path in place.

Content Effort is Google’s rater vocabulary for how much satisfying human work a stranger can see on the page. It is not a confirmed ranking factor, it is not experience and expertise wearing a new hat, and it is not “we spent forty hours on this.” This page maps the system: what Google wrote down, what the 2024 leak reporting suggests, and how you turn either into a page-level decision that survives a skeptical review. If you want the clean, citable definition, take what is Content Effort. If you want the argument for why the system holds together, stay.

TL;DR

• Raters already score effort, and there is a reasonable case that ranking systems are learning to estimate the same evidence at scale. The second half of that sentence is my inference, not a Google announcement.
• Visible effort is what survives on the live URL. Production effort is what happened in your document and your calendar. Only the first can be scored by someone who cannot interview the author.
• Three layers get collapsed in most coverage: the rater guidelines criterion, the leaked contentEffort attribute, and Search Central’s people-first questions. Keep them in separate columns.
• Score six on-page signals at 0 to 2 each for a 0 to 12 total, then adjust for page purpose. Word count is not the gate and never was.
• This page argues the system. The definition spoke cites it.

What is Content Effort?

Content Effort is the visible, non-replicable human work that makes Main Content more satisfying than the substitutes already available for that page purpose. The version I brief to a team, because one-sentence versions are the only kind that survive a deadline: if a careful paraphrase of today’s top ten results would still look like your URL, you do not have Content Effort. Blunt, slightly unfair, and I have never had it be wrong.

Google’s wording is older than the industry name, which is worth knowing before you present it as news. The guidelines tell raters that for most pages, Main Content quality is determined by the amount of effort, originality, and talent or skill that went into creating the content. Effort, in the handbook’s phrasing, is the extent to which a human being actively worked to create satisfying content. Notice what that includes: writing, curation, translation, or building functionality. That is broader than most teams assume, because we hear “effort” and picture a writer typing. Curation counts. A tool counts. What does not count is pushing existing copy through a translator or a generator and publishing at volume, which the handbook calls out directly.

So the term is the industry doing what it did to E-A-T in 2014 and to E-E-A-T in December 2022: taking a handbook paragraph and promoting it to a briefing category. I am fine with that. Shorthand is how standards travel between people who have six other things due. The discipline is keeping the claim class honest at each step. The handbook language is documented. The leak attribute is reported by third parties. The ranking weight, if any exists, is unknown. Any deck that blurs those three will be right for a quarter and embarrassing after the next Search Central post, and I say that as someone who wrote the embarrassing deck.

Is Content Effort a Google ranking factor?

No. Google does not publish a ranking factor list containing a field called Content Effort, and quality raters do not move live rankings when they submit a score. Their job is to tell Google whether its ranking systems are surfacing pages that look high quality to real people, the same relationship Search Central has described for E-E-A-T since the extra E post: rater concepts help Google evaluate its systems, and they are not a number you can inspect in Search Console.

What you can say out loud, with a source behind each line:

1. Documented: effort is a Main Content quality criterion, grouped with originality and talent or skill, with bars that change by page purpose.
2. Documented in 2025: little to no effort, plus little to no originality, plus little to no added value is a Lowest path, and the handbook names AI paraphrase and scaled publishing without curation as examples (Search Engine Land).
3. Reported: the 2024 Content Warehouse leak includes an attribute described as a language-model-based effort estimate for article pages (Searchable).
4. Inference, and label it every time: that attribute is plausibly how Google industrializes the rater concept at web scale. Plausible is not confirmed.
5. Adjacent: the people-first questions in Search Central’s guidance on creating helpful content ask whether you added original information or mostly rewrote other sources. Same question, operator English.

The table below is what I put in front of stakeholders, and I walk them through the middle column first, because that column stops a leak blog post from becoming a roadmap commitment.

Read the middle column before the left one, especially in a room where someone is taking notes for a client. I once deleted the status column from our own findings deck because it looked cluttered and I had opinions about whitespace. For two months nobody, including me, could tell which findings were fixed and which were still arguments. That is how “leak reporting” becomes “Google confirmed” between a Tuesday and a Thursday. Nobody has to lie. The column just looks like hedging until you need it.

That story has a practical edge. Operate as if visible effort will eventually be estimated at scale, because the direction of travel is obvious and the work is good work regardless. Do not tell your executive team that Google admitted a ranking factor. The first posture produces pages that hold up; the second produces a correction you issue in front of the same people. The ticket I refuse to file is “optimize for the contentEffort score.” There is no score to optimize, and phrasing it that way guarantees someone spends two weeks reverse-engineering a number instead of adding one fact the web does not have.

How do quality raters evaluate effort?

Raters start with page purpose, not with a labor quota. Then they spend a few minutes actually using the Main Content: reading the article, watching the video, trying the calculator. That is the step almost every SEO audit skips. We grade pages from a crawl and a scroll; the rater uses the thing. If your calculator does not work on mobile, no amount of schema talks its way out of that.

The guidelines ask what effort, originality, talent, or skill would look like for that type of page, which is why a recipe, a mortgage calculator, a product review, a medical explainer, and a joke video are not held to one standard. Direct human work counts, and the handbook’s example of translating a poem is deliberate: the labor is in the judgment, not the volume. Systems work counts too, when the page itself is the system. Unattended bulk conversion of other people’s content does not count, no matter how much engineering went into the pipeline, which is a sentence some of us needed in 2023. On forums, total human participation is allowed to add up, and that provision explains something you have noticed and resented: a long, messy Reddit thread outranking your tidy article that says nothing.

Accuracy joins the test on informational and YMYL pages, and this is the part teams underweight. Effort that produces confident falsehoods is not high Page Quality, so the six-signal instrument further down is not a substitute for subject-matter review on health, finance, legal, or safety topics. The scorecard tells you whether the page has evidence. A qualified reviewer still has to tell you whether the evidence is right.

The 2025 Lowest examples are the other half of the briefing, and they are unusually blunt for a Google document. Chicken-recipe doorway pages, auto-generated affiliate swaps, medical articles that still contain the phrase “as a language model,” paraphrased roundups with nothing added. None of those are Lowest because they are short. They are Lowest because the Main Content shows almost no human work relative to what already exists on the topic. And the handbook says you can credit your sources properly and still be rated Lowest, which is the sentence I would put on the first slide if I only had one.

What is the leaked contentEffort attribute?

Leak reporting describes contentEffort as a large language model estimate of how much effort went into an article page. Searchable argues it is the mechanism that answers Google’s people-first questions at web scale, and Hobo frames the same attribute as a proxy for how much a business invested in a page. Both reads are worth your time, and neither is a Google statement.

Here is the rigorous speculation, labeled as speculation so you can quote me without getting hurt. If you need to score a billion URLs the way 10,000 raters score a sample, the obvious engineering move is to train a model on the rater concept and store the output as a warehouse feature. Linguistic uniqueness, rare facts, original media, and some machine approximation of “would this page exist if the rest of the results vanished?” are things a model can plausibly be taught to look for. That story is consistent with how much of the 2025 handbook revision went to paraphrase and scaled abuse. Consistent, though. Not proven.

Where a careful reader will check you is the gap between that story and what it establishes. It does not establish that the attribute sits in the ranking function at any material weight, or that it is the Helpful Content system under a different label. It also does not mean a third-party crawler can hand you a numeric score for a URL, so if a tool offers you one, you are buying a proxy metric with a confident name. I would rather ship a page with a documented criterion and an admitted inference than a dashboard with a decimal point.

A dedicated teardown of the attribute is a page worth writing later, and it is on the backlog further down. Until it exists, keep contentEffort in the inference column and the rater guidelines in the documented column. Mixing them is the fastest way to lose a technically literate reader, and it is the central flaw in most coverage currently ranking for this term.

Visible effort versus production effort

The reframe worth owning is this. Production effort is hours, cost, interviews, tooling, and editorial process. Visible effort is the residue a stranger can point at on the live URL. Visible effort versus production effort with six on-page signals is the name I use in briefings now, because naming both halves is the only thing I have found that stops the conversation drifting back to timesheets.

And it does drift. I have sat in reviews where the defense of a page was the amount of work behind it, presented as if effort were a receipt you could staple to a URL. Timesheet theater. Nobody means harm by it, because it is genuinely painful to watch a week of real work rank behind something assembled in an afternoon, and the natural response is to talk about the week. The rater cannot see your week.

Cyrus Shepard has written about scoring pages as a rater, and his read matches the handbook closely: a low-effort page is information assembled with little human involvement, and a high-effort page is one where a person decided what mattered and organized it so a visitor could use it. High-effort pages tend to carry first-party facts. Process documents never enter the rating task at all, because the rater cannot see them and would not be asked to.

That split is the roundup from the top of this page, in the order it actually happened. We spent a week on it: real research, twelve tabs open, two interviews, a shared doc with actual arguments in it. A model drafted about 2,400 words from that doc, and an editor cleaned the tone and added a comparison table. The finished URL contained nothing a reader could not have gotten from the three pages already ranking, because every fact in that doc had come from those same three pages and the week had gone into agreeing with them. The production graph looked serious. Meetings happened, a brief existed, people cared. The visible graph was empty, and raters see only the second graph.

The test I run before a draft goes to review takes about ninety seconds. Strip the byline, the screenshots, the proprietary numbers, and the named method. If what remains still functions as a distinct answer because of sequencing, omissions, and judgment, the page has craft. If it collapses into a generic explainer any competitor could publish next Tuesday, you had production theater. Before I approve a draft I ask the writer for one thing: a fact whose source is not a URL. A measurement, a client number they can publish, a transcript line, a screenshot with a date on it. One is enough to change the character of the page, and zero is disqualifying no matter how good the prose is. The prose is often very good, which is exactly why the rule has to be mechanical.

Content Effort vs E-E-A-T, helpful content, and information gain

These concepts overlap heavily and they are not aliases, and treating them as interchangeable is the briefing error that costs a quarter. Watch how it plays out. An SEO lead diagnoses “weak E-E-A-T” on a page that is actually a competent paraphrase, prescribes author bios and credential markup, and ships a beautifully credentialed page that still says exactly what the other results say. The diagnosis was one word off and the remediation was wasted.

E-E-A-T is about reputation, experience, and trust in the creator and the site. Content Effort is about labor evidence inside the Main Content. The two come apart in both directions, which is the useful part rather than a technicality. A trusted medical institution can publish a thin auto-summary that fails on effort while passing on trust. An unknown practitioner can publish a messy, irreplaceable field note that fails on reputation signals while carrying obvious effort. Different failures, different fixes, and the table below is how I keep them straight in a room.

Read the right-hand column in review meetings, because every row there describes a real ticket that someone has filed at my expense. The full comparison, including where scaled content abuse sits as a site-level rather than page-level judgment, lives in Content Effort vs E-E-A-T and helpful content, and information gain as a writing doctrine is mapped in our information gain content framework. Send a writer to the specific one rather than making them guess which failure they have.

What does high-effort vs low-effort look like on the same query?

Hold the query constant and the difference stops being abstract, which is the only way I have gotten this to land with writers who think it is a vibes conversation. Take a commercial-informational search like “best [category] tools for [segment],” where two pages target identical intent and only one shows visible work.

The low-effort version. A 2,800-word listicle that restates vendor marketing pages, uses stock screenshots from press kits, orders the publisher’s affiliate partners in a suspiciously convenient sequence, and closes with a generous “it depends on your needs.” Every claim already exists in the current results. No workflow was run. No pricing is worked at a specific seat count. Swap the product names and the skeleton stands unchanged, which is the tell I look for first. Under the 2025 handbook, this sits close to the paraphrase-plus-affiliate pattern raters are told to mark Lowest when the Main Content is recycled and nothing has been added.

The high-effort version. Same query. The author ran each product through one real workflow, dated the screenshots, worked the pricing at both 10 and 50 seats, named the point where their favorite tool stops being the right answer, and published a failure that appears nowhere in the vendor documentation. This page is allowed to be shorter than the first, and usually is. Its evidence cannot be generated from the results page alone, which is the definition in practice: first-party information, artifacts that are expensive to copy, and added value measured against the substitutes.

The lesson is not “write more reviews,” which is where this argument gets misfiled. It is that the query does not determine effort, the residue on the URL does. A definition page can carry high Content Effort if it disambiguates a term the rest of the web is conflating, which is most of the reason this page exists. A listicle can be low effort at any length, including six thousand words.

How should teams measure Content Effort?

Do not invent a fake Google score, and be loud about that when you present this. What you can score is visible effort evidence, approached the way a rater would: purpose first, then the on-page proof, using six signals at 0 to 2 each for a 0 to 12 total. It is an editorial instrument for making review conversations concrete, so “this feels thin” becomes a specific missing artifact. It is not a claim about Google’s weights.

The zeros are the actual point of that table, and the total is almost a distraction. A number helps you triage fifty URLs, but the column a writer needs is the one naming which kind of evidence is missing from their draft.

Bands, for triage only: 0 to 3 is Lowest-risk evidence, 4 to 6 is low, 7 to 9 is high, 10 to 12 is the top of the instrument. Then apply the purpose modifier, because a forty-second explainer clip and a documentary were never supposed to land on the same raw total. The full procedure, including what to do with each signal when it comes back zero, is in the Content Effort audit checklist, and the runnable version is the audit template.

One clarification prevents the most common misuse, and I have watched it go wrong often enough to lead with it. “Add” never means “add words.” A zero on first-party information gets fixed by a measurement, a teardown, or a primary document nobody else bothered to read. A zero on artifacts gets fixed by a dated screenshot, a working calculator, or a chart built from your own data. A zero on synthesis gets fixed by a decision rule the rest of the results do not repeat. If a ticket could be satisfied by making the draft longer, the ticket was written wrong and the writer will correctly resent it.

Can AI-assisted content score high on Content Effort?

Yes, when the visible work is real. No, when the model produced the entire Main Content and the results page was the entire research process. Google’s handbook is explicit that generative AI tools alone do not determine effort or Page Quality, and the Lowest path is defined by the absence of effort, originality, and added value rather than by whatever tool generated the text.

That is the doctrine to have loaded when leadership asks for a blanket ban, which in my experience arrives about a week after a competitor gets publicly caught. Banning the tool does not raise Content Effort by a single point, because the pages that fail were failing on evidence rather than on provenance. Requiring artifacts does raise it, whether or not a model touched the draft. The policy that works is boring, and boring is the goal: models can draft, humans must contribute at least one thing to the page that did not exist before the draft started, and nothing ships with a zero in the first-party column. The extended argument, including why scaled publishing gets judged differently than a single AI-assisted URL, is in Content Effort and AI-generated content, and the adjacent craft problem of cadence and tells is covered in how to humanize AI writing.

One forward inference, labeled as inference rather than smuggled in as a conclusion. As generation keeps getting cheaper, origin becomes a weaker discriminator and visible effort becomes a stronger one. That is my best explanation for why the 2025 handbook spent its new pages on paraphrase and scaled abuse instead of on prohibition. It is the same economic pressure that turned E-E-A-T into a briefing term once cheap content flooded YMYL results, and this is the labor-evidence version of that story rather than a new one.

What to write next as

Content Effort becomes a real search category

Terms follow a demand curve, and this one has not finished walking it. E-E-A-T, helpful content, and information gain all went through the same sequence: first “what is it,” then “is it a ranking factor,” then comparisons against adjacent concepts, then checklists, then the AI question, then tools and scores. Content Effort is early on that curve, with coverage concentrated in a handful of leak and rater posts, so the list below is extrapolated from those analogs rather than read off a mature results page. I would rather tell you where the guess comes from than present it as data.

The early questions are already answered across this cluster: what the criterion is, whether it is a ranking factor, how it compares to E-E-A-T and information gain, how to audit it, and whether AI-assisted content can score well. Likely next, and not yet written here:

• Content Effort examples, and high versus low effort pages side by side
• Content Effort score and checker queries, which will arrive with tooling
• Scaled content abuse compared against page-level Content Effort
• Content Effort for YMYL topics specifically
• How to recover from low-effort ratings, and where that overlaps with helpful content recovery
• The contentEffort leak explained as a dedicated technical page
• A standalone glossary entry, once enough people search the bare term
• Follow-up questions that only appear once a term circulates: does word count equal Content Effort, how do raters know content is AI, is it part of core updates

Treat that as an editorial backlog rather than a sitemap, and resist linking to slugs that do not exist yet. It is a small trust cost, and careful readers notice it first.

The harder problem is not knowing the standard. It is keeping the standard alive after the person who argued for it moves to another account. Quality language shifts, somebody adds a checklist to a shared document, and by the third sprint that checklist is a tab nobody opens while pages ship on deadline anyway. It decays because it lives in prose that has to be re-argued from scratch every time, by a different person, usually on a Friday.

The durable version is the same standard encoded once as a reusable skill, run as a repeatable workflow, and applied by an agent that can fetch the URL, score the visible evidence signal by signal, and hand a human a result with a dispute trail attached. The judgment stays human. What gets automated is the consistency of applying it, plus the context that would otherwise live in one person’s head and leave with them. The failure mode of editorial standards is rarely disagreement about the standard. It is drift in how twenty writers apply it across six months, and drift is the opposite of anything compound.

Metaflow is where that handoff lives for this framework, and it belongs in a workflow rather than a document because of the drift above. The Content Effort audit template scores specified URLs against the six signals, and the skill behind it holds the rater language stable, so a writer receives “add a first-party artifact to section three” instead of “this feels thin, make it better.” When the diagnosis turns out to be your production system rather than one page, content engineering for non-commodity pages is the companion argument.

Frequently Asked Questions

What is Content Effort?

Content Effort is visible, non-replicable human work in the Main Content that makes a page more satisfying than the substitutes available for its purpose. Google groups effort with originality and talent or skill in the rater guidelines, and sets the bar by page type rather than by length. It is not hours spent, and it is not a published ranking factor. Metaflow keeps that definition attached to the six-signal scorecard so the argument and the audit do not drift apart.

Is Content Effort a Google ranking factor?

Not in any ranking factor list Google publishes. Raters use effort to judge Page Quality, and those ratings help Google evaluate whether its systems work rather than adjusting individual URLs. The leaked contentEffort attribute may estimate similar evidence with a model, and that remains a hypothesis. Your response is the same either way: require visible evidence before a page ships. Metaflow labels the leak column as inference deliberately, so nobody quotes it as fact three slides later.

How do quality raters evaluate effort?

They match the bar to page purpose first, then spend real time using the Main Content instead of skimming it. Effort can be writing, curation, translation, or building functionality, which is broader than most teams expect. Bulk conversion of other people’s content without oversight does not count. Since 2025, Main Content that is almost entirely recycled with nothing added sits on an explicit Lowest path, with AI paraphrase named as an example.

What is the leaked contentEffort attribute?

Leak reporting describes it as a language-model-based effort estimate for article pages. It is not named anywhere in the rater handbook, so mapping it onto the guidelines criterion, or calling it the Helpful Content system renamed, is inference rather than documentation. Keep it in its own evidence column in any audit, so the claim class stays visible to whoever inherits your work.

Can AI-assisted content score high on Content Effort?

Yes, when first-party artifacts, genuine synthesis, and added value are visible on the finished page. No, when a model restates the existing results and a human only adjusts the tone. Google’s handbook states that generative AI alone does not determine effort or Page Quality, so the rating turns on whether effort, originality, and added value are present. That is a question about the page, not the tool that drafted it.

How should teams measure Content Effort?

Score six visible signals from 0 to 2 each: first-party information, non-replicable artifacts, model-changing synthesis, craft matched to purpose, curation fingerprints, and added value against substitutes. That gives a 0 to 12 total, adjusted for page purpose. The audit checklist covers the procedure in full. Running it as a Metaflow skill keeps scoring consistent across writers, which is where most editorial standards quietly fail. Do not publish the result as though it were a Google metric.

Key takeaways

• Content Effort is about visible evidence on the URL, not the hours, meetings, or tooling behind it.
• The rater guidelines, the leaked attribute, and the people-first questions are three separate evidence layers, and collapsing them is what gets teams corrected in public.
• Six signals scored against page purpose beat word count as a review instrument, and every zero gets fixed with an artifact rather than an adjective.
• Where the text came from is not the rating. Paraphrase without added value is what earns a Lowest path.
• The next year or two will bring queries about examples, checkers, YMYL, and recovery, and owning the definition now is cheaper than competing for it later.

Sources

• Google Search Quality Evaluator Guidelines (PDF)
• Search Engine Land: Google quality raters and AI-generated content
• Google Search Central: Creating helpful, reliable, people-first content
• Google Search Central: E-A-T gets an extra E for Experience
• Cyrus Shepard / Zyppy Signal: Content Effort
• Searchable: The contentEffort attribute from the Content Warehouse leak
• Hobo: What is Google’s Content Effort signal?
• Google Search Central: Spam policies for Google web search