Something Inc.Schedule a free consultation
CONTENT

Content effort is a ranking signal you can audit

Cyrus Shepard published a scoring method for the thing everyone gestures at and nobody measures. The useful part is that it is evidence of distinctive value, not hours logged.

TTTyler TruffiManaging Partner · AUG 15, 2026 · 12 MIN READ
5.3%
of top-three ranking pages that were entirely AI-generated (Ahrefs, 331,000 pages, Jul 27, 2026)
82.2%
of top-three rankings held by pages under 50% AI content in the same study
2x to 3x
impression advantage for low and moderate AI content over high AI content
Aug 13, 2026
publication date of Cyrus Shepard's content effort scoring method on Zyppy Signal
TL;DR · 60 SECONDSCyrus Shepard published two companion pieces on August 13, 2026 arguing that Google evaluates evidence of distinctive value rather than the raw labor a page consumed, and offering a way to score it. Read alongside Ahrefs' study of 331,000 pages, the picture is consistent: authorship method is close to irrelevant and evidence of something the page uniquely contains is close to decisive. Content effort is auditable, and the audit is more useful than another keyword gap analysis because it tells you which pages are structurally incapable of ranking.

Every content team has been told to publish higher-quality content, and almost none of them have been told what quality means in a way that survives contact with a spreadsheet. Content effort is the most useful attempt at that definition anyone has published this year, mostly because it refuses the obvious version.

The obvious version says effort equals work: hours spent, words written, revisions logged. Shepard's framing on Zyppy Signal is sharper than that. What gets evaluated is evidence of distinctive value, meaning the traces a page carries of containing something that could not have been assembled from other pages on the same topic. A team can burn eighty hours producing a page with no such trace, and a specialist can produce one in ninety minutes.

What content effort actually means

The distinction matters because it changes what you optimize. If effort is labor, the response is more resources. If effort is evidence, the response is different inputs: original measurement, first-hand access, proprietary data, primary sources, named expertise, artifacts the writer had to obtain rather than paraphrase.

Think about what a retrieval system can actually observe. It cannot see your calendar. It can see whether a page contains a number that appears nowhere else, a photograph that is not stock, a method section, a named practitioner with a track record, a dataset, a transcript, a specification. Those are the observable proxies for effort, and they are the only ones available to a machine.

INPUTCOSTS A LOT OF HOURSLEAVES OBSERVABLE EVIDENCEVERDICT
Rewriting the top five results in your own wordsYesNoExpensive and invisible
Running a small original test and publishing the methodSometimesYesThe highest-return content investment available
Adding a comparison table built from vendor documentationNoYesCheap, underused, and highly extractable
Interviewing one practitioner and quoting them by nameNoYesFast route to distinctive value
Expanding a 900-word post to 2,400 wordsYesNoThe most common wasted content investment
Publishing the failure case alongside the success caseNoYesRare enough to be a differentiator on its own

Look at the last column. The two rows that consume the most hours produce the least evidence, and they describe how most enterprise content programs spend their budget. That mismatch is the entire finding.

The evidence that effort, not authorship, is what gets scored

The strongest supporting data came three weeks earlier. Ryan Law and Xibeijia Guan published an Ahrefs study on July 27, 2026 covering 331,000 pages, and the results cut directly against the industry's favorite anxiety. Pages that were entirely AI-generated made up 5.3% of top-three rankings. Around 9% were at least 80% AI content. Pages under 50% AI content accounted for 82.2% of top-three rankings.

Under 50% AI content82%
At least 80% AI content9%
Entirely AI-generated5%

Share of top-three ranking pages by AI content proportion, from the Ahrefs study of 331,000 pages published July 27, 2026.

The same study found low and moderate AI content earning roughly two to three times the impressions of high and very high AI content. The authors' conclusion was that Google punishes low quality rather than AI authorship, which is the position we took when we covered why Google never punished AI content as such. Shepard's framework explains the mechanism: heavily generated pages tend to contain nothing that was not already on the web, so they carry no evidence of distinctive value, and the correlation with AI authorship is incidental.

THE REFRAME THAT MATTERSStop asking whether a model wrote it. Start asking what the page contains that a model could not have produced without access to something. That single question sorts a content library faster than any tooling, and it is the question underneath every content program we run.

How to run a content effort audit

The audit is mechanical once you accept the definition. You are scoring each page for observable evidence, not judging its writing. Five passes, in order, on your top hundred pages by business value rather than by traffic.

1DAY 1Score for unique substance
THE MOVES
For each page, list every claim that carries a number, a name, or a specification
Mark each one as sourced externally, sourced internally, or unsourced
Flag any page where every single claim traces back to another public page
DONE WHENYou have a list of pages that contain no information the web did not already have.
2DAY 2Score for first-hand artifacts
THE MOVES
Count original images, screenshots of your own product, and photographs you took
Count quotes attributed to a named person you actually spoke to
Count datasets, exports, or measurements published with a method
DONE WHENEvery page has an artifact count, and most of them are zero.
3DAY 3Score for authorship signal
THE MOVES
Check whether the byline is a real named person with a verifiable track record
Check whether that person has a consistent entity presence off your domain
Replace generic team bylines on your highest-value pages
DONE WHENYour top pages carry a named author an engine can corroborate elsewhere.
4DAY 4Score for extractability
THE MOVES
Confirm the direct answer to the page's core question appears in the first screen
Confirm comparisons render as tables rather than as prose
Confirm the page states a verdict rather than surveying options neutrally
DONE WHENA retrieval system can lift a self-contained, attributable claim from every page.
5DAY 5Triage into three piles
THE MOVES
Pile one: pages with real evidence that need only extractability work
Pile two: pages with a genuine topic and no evidence, which need one new input each
Pile three: pages with neither, which should be consolidated or removed
DONE WHENYou have a ranked plan where every item names the specific input the page is missing.

The triage step is where teams flinch. Pile three is usually larger than anyone expects, and consolidating it feels like destroying assets. It is not. A page that contains nothing distinctive competes against every other page that contains nothing distinctive, which is a competition decided by domain authority alone, and you will lose it to a larger site forever.

What the audit finds on most enterprise sites

Three patterns show up almost every time we run this. The first is that the pages with the most evidence are usually the least optimized: engineering write-ups, support documentation, and internal explainers that nobody in marketing has touched. They contain real specifications and real screenshots and they are structured badly.

The second is that the marketing pages with the most invested hours contain the least evidence. A pillar page assembled from competitor research consumed a quarter of somebody's year and contains nothing that was not already indexed. The third is that the fix is almost always additive rather than editorial: the page does not need rewriting, it needs one input the writer did not have.

1Look in support and docs firstThe highest-evidence content in most companies is already written and sitting in a help center with no headings, no author, and no internal links. Restructuring it is cheaper than commissioning anything new and it usually outperforms.
2One input per page beats one rewrite per pageAdding a single measurement, a single named quote, or a single original screenshot changes a page's evidence profile more than a full rewrite does. Budget inputs, not word counts.
3Refresh beats republishUpdating a page that already carries evidence outperforms publishing a new page that does not, which matches what we found on refreshing existing content for AI citations. The compounding is in the asset, not the calendar.

The catalog and ecommerce version of this is sharper still, because product data is evidence by definition and most of it never reaches a public page in a readable form. That was the core of our print ecommerce engagement: the distinctive material already existed inside the business, and the work was exposing it rather than writing anything new.

The fourth pattern is about ownership rather than content, and it is the one that decides whether any of this survives a reorganization. Evidence-bearing inputs live with people who are not writers. The measurement lives with analytics, the specification lives with product, the screenshot lives with support, the named practitioner lives in delivery. A content team with no standing claim on those people can only produce the kind of page that scores zero, no matter how good its writers are. Fixing the audit findings is usually an access problem before it is a craft problem.

Where content effort stops paying

Take the position seriously enough to state its limits. Evidence of distinctive value does not rescue a page targeting a query with no commercial relevance to you, and it does not overcome a genuinely broken technical foundation. If a page cannot be crawled or rendered, its evidence is invisible and its score is irrelevant.

It also has a ceiling in categories where the answer is genuinely commodity. There is no distinctive way to state a standard conversion formula, and no amount of original photography makes one page about it better than another. In those categories the right move is to stop competing on the informational query and compete on the decision query instead, where comparison and verdict still differentiate.

And there is a timing caveat. Evidence-heavy pages take longer to earn their position because they take longer to accumulate corroboration. Teams in fast-moving categories like tech and SaaS often need a parallel track of quicker comparison content while the substantive assets mature. Running only one of the two is the most common planning error we see.

The work order

Run the five-day audit on your top hundred pages by pipeline contribution. Expect pile three to be a third of them. Then pick the ten pages in pile two with the highest business value and name the one input each is missing: a measurement, a named source, an original image, a specification. Assign those ten inputs as tasks, not as articles.

Do that for a quarter before commissioning anything new. Most content teams discover their library is not too small, it is too thin, and thinness is cheaper to fix than volume is to create. The scoring method is worth reading in full at Zyppy Signal's August 13 piece, and the deeper structural version of this argument is in our guide to content that earns AI citations.

DO THIS NEXTThis week: score your top hundred pages for unique claims and first-hand artifacts, and count how many score zero on both. This month: assign one missing input per page for the ten highest-value pages that have a real topic but no evidence. This quarter: consolidate or remove the pages that have neither, and move your support and documentation content into the main site structure where it can rank.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.

TT
Tyler TruffiMANAGING PARTNER, SOMETHING INC.

Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.

Free consultation

Let us be the last SEO agency you ever work with

A 30 minute call and a free audit of your SEO and GEO position. You keep the findings either way.