What Good Content Is: contentEffort and Google's Soft Metrics

01One line in the leak

"What is good content?" has a surprisingly short machine answer in the leaked Content Warehouse documentation: a single field, contentEffort, described in one line as an LLM-based effort estimation for article pages. It sits in the module QualityNsrPQData, stored as a list of versioned values whose type is defined in Google's quality/nsr code, next to other page-level quality predictors such as chard, tofu, keto, rhubarb and vlq.

That one line is also everything the leak says. It names no inputs, no scale, no weight and no use in ranking, and it limits the field to article pages. A weather forecast, a calculator or a product page may never receive this value. Anything beyond the line is inference, and this article marks it as such.

02Google's own definition of effort

The word the field uses is the first word of Google's rater manual on content quality. The Search Quality Rater Guidelines of 11 September 2025 say that for most pages the quality of the main content can be determined by the amount of effort, originality, and talent or skill that went into it, plus accuracy for informational and YMYL pages. These four are the soft metrics: judgments a trained person makes, not numbers a tool reports.

The definition of effort is about people and purpose, not length or production method. Effort is the extent to which a human actively worked to create satisfying content, and it can go into building page functionality as much as into writing. What counts as enough depends on the page type, and the use of generative AI alone does not decide the level of effort.

03The soft metrics, spelled out

The rater manual gives each soft metric a high and a low end. High effort means well organised, edited and curated; low effort means helpful content mixed with filler, little discussion, or instructions buried under unrelated text. High originality means content unique to the site, own photos or footage, or a first-hand perspective; low originality means summaries of what others already said.

Two calibrations matter more than any single criterion. First, filler is named as its own failure: content that inflates a page without serving its purpose. Second, the bar is relative to the competition, because typical and average pages on a topic are Medium, not High. At the bottom, pages built almost entirely from copied, paraphrased or AI-generated material with no added value are rated Lowest, even when they credit their sources.

04Google's twelve questions for publishers

For publishers, Google translates the same ideas into self-assessment questions. The twelve content and quality questions ask about original information and analysis, completeness, insight beyond the obvious, added value over copied sources, honest titles, whether the page would be bookmarked or cited in print, value compared with other results, spelling, sloppy or hasty production, and mass production across many creators or sites.

Two further passages sharpen what effort is not and what it looks like. Google says outright that it has no preferred word count, and it asks publishers to show the "How": for reviews, how many products were tested, with what results and methods, backed by evidence such as photographs. On the other side of the line sits the spam policy on scaled content abuse, which applies no matter how the content is created.

05Rater judgment and the machine estimate

Rater scores do not rank pages. Google states that no single rating can move a page, and that ratings serve to measure how well search works and to supply examples of helpful and unhelpful results. The rater's soft metrics are therefore not inputs; they are the yardstick against which systems are checked.

The leaked field and the rater manual share one word, and that is where the evidence ends. Nothing in the leak says that contentEffort is trained on rater judgments, and nothing in the guidelines mentions the field. Reading contentEffort as a machine approximation of the rater's effort criterion is plausible, and it is our interpretation.

06The checklist

1. Purpose first: say what the page is for and put the most helpful part at the top. 2. Effort shows in the result: organised, edited, curated, with no filler ahead of the answer. 3. Something original: own data, own photos, own tests or a first-hand perspective; if you draw on sources, add substantial value. 4. Skill fits the task: the explanation works because the author can do the thing.

5. Accurate, and on YMYL topics consistent with expert consensus. 6. An honest title that neither exaggerates nor shocks. 7. Show the How: methods, tests, evidence, and a disclosure where automation did substantial work. 8. Calibrate against the results page: if your page is typical for the topic, raters call that Medium.

9. Do not write to a word count; length is not effort. 10. No scale without value: generating, stitching or paraphrasing pages in bulk is where the spam policy begins, whatever tool is used. Items 1 to 10 restate Google's documented criteria; that meeting them raises contentEffort is not documented anywhere.

07References