Google's 'questions to ask yourself' list is a leaked rubric
The question: what does Google's self-assessment guidance for helpful content actually reveal?
The 'Creating helpful, reliable, people-first content' document contains a long list of self-assessment questions. Read closely, it is a near-transcription of what the systems were trained to reward — a rubric published in the second person.
Several questions are unusually specific and worth treating as direct signals. 'Does the content provide original information, reporting, research, or analysis?' privileges primary contribution over synthesis. 'Does the content provide insightful analysis or interesting information beyond the obvious?' targets aggregator content explicitly. 'Would you trust this for an issue concerning your money or your life?' folds YMYL back in. And the meta-question — 'Is the content written by an expert or enthusiast who demonstrably knows the topic well?' — pairs expertise with enthusiast, echoing the experience addition.
The most actionable is the 'search-engine-first' set: was the content made primarily to attract search traffic rather than to help people; does it summarize what others said without adding value; was it produced to hit a word count. These describe the exact failure profile of scaled affiliate content.
Caveat: this is guidance, not a checklist that flips a switch. Answering 'yes' to every question does not guarantee ranking; answering 'no' to several is a reliable predictor of trouble.
What we still don't know: the relative weight the systems place on 'original information' versus competent-but-derivative coverage that is genuinely well-organized.
The question: what does Google's self-assessment guidance for helpful content actually reveal?
The 'Creating helpful, reliable, people-first content' document contains a long list of self-assessment questions. Read closely, it is a near-transcription of what the systems were trained to reward — a rubric published in the second person.
Several questions are unusually specific and worth treating as direct signals. 'Does the content provide original information, reporting, research, or analysis?' privileges primary contribution over synthesis. 'Does the content provide insightful analysis or interesting information beyond the obvious?' targets aggregator content explicitly. 'Would you trust this for an issue concerning your money or your life?' folds YMYL back in. And the meta-question — 'Is the content written by an expert or enthusiast who demonstrably knows the topic well?' — pairs expertise with enthusiast, echoing the experience addition.
The most actionable is the 'search-engine-first' set: was the content made primarily to attract search traffic rather than to help people; does it summarize what others said without adding value; was it produced to hit a word count. These describe the exact failure profile of scaled affiliate content.
Caveat: this is guidance, not a checklist that flips a switch. Answering 'yes' to every question does not guarantee ranking; answering 'no' to several is a reliable predictor of trouble.
What we still don't know: the relative weight the systems place on 'original information' versus competent-but-derivative coverage that is genuinely well-organized.
Google's AI-content stance is about quality and disclosure, not authorship
The question: does AI-generated content inherently lack E-E-A-T?
Google's stated position is that the method of production is not the issue; quality is. Per Search Central guidance, content is rewarded for being helpful, reliable, and people-first regardless of how it is produced, and using automation to generate content primarily to manipulate rankings violates spam policy. The line is intent and quality, not 'human versus machine'.
But the E-E-A-T frame exposes a structural tension for AI content. Experience, the second E, is first-hand involvement — something a model that has not used the product, visited the place, or treated the patient cannot itself possess. Authoritativeness and trust depend on a verifiable responsible entity and external reputation, which the generating tool does not supply. AI can produce fluent expertise-shaped text; it cannot, on its own, produce experience or accountable identity.
The coherent operating model: AI as a drafting and structuring tool under a named, accountable human or organization that supplies the experience, verifies the facts, and owns the claims. The failure mode Google's spam guidance targets is unattended scaled generation with no responsible reviewer — the 'scaled content abuse' policy added in 2024 names exactly this.
Caveat: 'a human reviewed it' is not a magic phrase; the review has to actually add the experience, accuracy, and accountability the frame requires.
What we still don't know: how reliably any system detects AI provenance, and whether detection matters at all if the output is genuinely high quality and accountable.
The question: does AI-generated content inherently lack E-E-A-T?
Google's stated position is that the method of production is not the issue; quality is. Per Search Central guidance, content is rewarded for being helpful, reliable, and people-first regardless of how it is produced, and using automation to generate content primarily to manipulate rankings violates spam policy. The line is intent and quality, not 'human versus machine'.
But the E-E-A-T frame exposes a structural tension for AI content. Experience, the second E, is first-hand involvement — something a model that has not used the product, visited the place, or treated the patient cannot itself possess. Authoritativeness and trust depend on a verifiable responsible entity and external reputation, which the generating tool does not supply. AI can produce fluent expertise-shaped text; it cannot, on its own, produce experience or accountable identity.
The coherent operating model: AI as a drafting and structuring tool under a named, accountable human or organization that supplies the experience, verifies the facts, and owns the claims. The failure mode Google's spam guidance targets is unattended scaled generation with no responsible reviewer — the 'scaled content abuse' policy added in 2024 names exactly this.
Caveat: 'a human reviewed it' is not a magic phrase; the review has to actually add the experience, accuracy, and accountability the frame requires.
What we still don't know: how reliably any system detects AI provenance, and whether detection matters at all if the output is genuinely high quality and accountable.
YMYL is a spectrum of potential harm, not a list of niches
The question: is your page YMYL (your money or your life)?
Most practitioners answer by category — finance, health, legal, news. The guidelines are more precise. YMYL is defined by the potential for harm if the content is wrong, not by topic membership. The 2022 QRG reframed it explicitly around harm: harm to one's self, to others, or to society, including financial, safety, and informational harm.
This matters because it makes YMYL contextual within a page. A recipe blog is not YMYL — until it publishes canning instructions, where botulism is the failure mode. A hobby forum is not YMYL — until a thread gives electrical-wiring advice. The harm potential of the specific claim, not the site's vertical, sets the standard.
The guidelines also describe a gradient: clearly-YMYL, clearly-not-YMYL, and a wide middle. They instruct raters to apply judgment about the degree of potential harm rather than a binary switch. A page that could mislead someone into a small, recoverable mistake sits differently from one that could cause serious injury.
The operational consequence: scan your content for harm potential at the claim level, not the category level. The expertise bar rises with the cost of being wrong.
Caveat: this is a rating framework. The ranking systems' notion of 'higher stakes query' is inferred, not the literal YMYL flag.
What we still don't know: how finely the systems distinguish harm gradients within a single broad topic.
The question: is your page YMYL (your money or your life)?
Most practitioners answer by category — finance, health, legal, news. The guidelines are more precise. YMYL is defined by the potential for harm if the content is wrong, not by topic membership. The 2022 QRG reframed it explicitly around harm: harm to one's self, to others, or to society, including financial, safety, and informational harm.
This matters because it makes YMYL contextual within a page. A recipe blog is not YMYL — until it publishes canning instructions, where botulism is the failure mode. A hobby forum is not YMYL — until a thread gives electrical-wiring advice. The harm potential of the specific claim, not the site's vertical, sets the standard.
The guidelines also describe a gradient: clearly-YMYL, clearly-not-YMYL, and a wide middle. They instruct raters to apply judgment about the degree of potential harm rather than a binary switch. A page that could mislead someone into a small, recoverable mistake sits differently from one that could cause serious injury.
The operational consequence: scan your content for harm potential at the claim level, not the category level. The expertise bar rises with the cost of being wrong.
Caveat: this is a rating framework. The ranking systems' notion of 'higher stakes query' is inferred, not the literal YMYL flag.
What we still don't know: how finely the systems distinguish harm gradients within a single broad topic.
Why outdated content is a trust failure, not a freshness failure
The question: when stale content gets demoted, is recency the signal or is something deeper happening?
The careful reading routes through trust. The QRG flags content that is inaccurate because it is out of date, especially on YMYL topics, as a trust problem. The issue is not the timestamp; it is that the page now makes claims that are no longer true. Recency is a proxy.
— A 2019 tax page citing repealed rules: untrustworthy, not merely old
— A 2010 page on a mathematical proof: old and perfectly trustworthy
— The demotion tracks accuracy decay, not the date field
Evidence: the guidelines tie 'outdated' to YMYL accuracy specifically. They do not penalize age in domains where information is stable.
Counter-evidence: query-deserves-freshness behavior in ranking does reward recency directly for some queries (news, trending). That is a separate ranking mechanism from the quality-label logic, and conflating them causes mistaken 'update everything monthly' strategies.
Caveat: a changed date with unchanged stale content is the worst case, it signals false freshness, which is itself a trust violation.
What we still don't know: how systems distinguish genuinely-stable evergreen content from neglected content that merely hasn't decayed yet.
—
Если тема зашла, посмотри @thesignalnoise
The question: when stale content gets demoted, is recency the signal or is something deeper happening?
The careful reading routes through trust. The QRG flags content that is inaccurate because it is out of date, especially on YMYL topics, as a trust problem. The issue is not the timestamp; it is that the page now makes claims that are no longer true. Recency is a proxy.
— A 2019 tax page citing repealed rules: untrustworthy, not merely old
— A 2010 page on a mathematical proof: old and perfectly trustworthy
— The demotion tracks accuracy decay, not the date field
Evidence: the guidelines tie 'outdated' to YMYL accuracy specifically. They do not penalize age in domains where information is stable.
Counter-evidence: query-deserves-freshness behavior in ranking does reward recency directly for some queries (news, trending). That is a separate ranking mechanism from the quality-label logic, and conflating them causes mistaken 'update everything monthly' strategies.
Caveat: a changed date with unchanged stale content is the worst case, it signals false freshness, which is itself a trust violation.
What we still don't know: how systems distinguish genuinely-stable evergreen content from neglected content that merely hasn't decayed yet.
—
Если тема зашла, посмотри @thesignalnoise
E-E-A-T is a proxy target, not a ranking factor
The question: does improving E-E-A-T (experience, expertise, authoritativeness, trustworthiness) directly move rankings?
The precise answer, per Google's own statements, is no. Google's Search Liaison and the Search Central documentation have been consistent: there is no single E-E-A-T score in the ranking pipeline. The four letters are a conceptual frame written for human quality raters in the Search Quality Rater Guidelines (QRG). Raters never touch the index — their judgments train and validate the systems, they do not directly rank your URL.
What actually exists are many signals that correlate with the qualities raters are told to look for: links, mentions, query-document relevance, site-level reliability signals. E-E-A-T is the human-readable description of the destination; the signals are the road.
Caveat: this distinction is not pedantic. Treating E-E-A-T as a dial you turn produces cargo-cult tactics — bolting an author box onto thin content, stuffing 'as an expert' phrasings. None of those are the signals; they are guesses at what the signals reward, and frequently wrong guesses.
The productive reframe: E-E-A-T tells you what 'good' looks like to a careful human reader. The engineering question is which observable artifacts a human and a machine would both read as evidence of that quality, and whether you can generate those artifacts honestly.
What we still don't know: the relative weighting of trust versus the other three. Google has stated trust is the most important member of the family, but the mechanism by which 'trust' becomes a measurable input remains undisclosed, and likely varies by query class.
The question: does improving E-E-A-T (experience, expertise, authoritativeness, trustworthiness) directly move rankings?
The precise answer, per Google's own statements, is no. Google's Search Liaison and the Search Central documentation have been consistent: there is no single E-E-A-T score in the ranking pipeline. The four letters are a conceptual frame written for human quality raters in the Search Quality Rater Guidelines (QRG). Raters never touch the index — their judgments train and validate the systems, they do not directly rank your URL.
What actually exists are many signals that correlate with the qualities raters are told to look for: links, mentions, query-document relevance, site-level reliability signals. E-E-A-T is the human-readable description of the destination; the signals are the road.
Caveat: this distinction is not pedantic. Treating E-E-A-T as a dial you turn produces cargo-cult tactics — bolting an author box onto thin content, stuffing 'as an expert' phrasings. None of those are the signals; they are guesses at what the signals reward, and frequently wrong guesses.
The productive reframe: E-E-A-T tells you what 'good' looks like to a careful human reader. The engineering question is which observable artifacts a human and a machine would both read as evidence of that quality, and whether you can generate those artifacts honestly.
What we still don't know: the relative weighting of trust versus the other three. Google has stated trust is the most important member of the family, but the mechanism by which 'trust' becomes a measurable input remains undisclosed, and likely varies by query class.
The second E was added to solve a specific failure mode
The question: why did Google add 'Experience' to E-A-T in December 2022, when 'Expertise' was already there?
The two are not redundant. Per the QRG, expertise is formal or demonstrated knowledge of a subject; experience is first-hand, lived involvement with the specific thing being discussed. A board-certified dermatologist has expertise on a drug's mechanism. A patient who took it for two years has experience of its side effects. The guidelines explicitly note that for some topics, the experienced person produces more trustworthy content than the credentialed one.
The addition reads as a response to a measurable problem: the post-2020 flood of competent-sounding content assembled by writers who had never used the product, visited the place, or held the object. Expertise can be faked or rented; first-hand experience leaves harder-to-fake traces — original photos, specific failure cases, idiosyncratic detail that aggregators lack.
For affiliate and review content this is the load-bearing distinction. A spec-sheet rewrite signals neither. A teardown photo with a thumbnail of your own hand, a specific firmware version where a bug appeared, a measured number you took yourself — these are experience artifacts.
Caveat: 'experience' is not universally required. For a YMYL (your money or your life) medical claim, the guidelines still privilege expertise; lived experience supplements but does not replace clinical authority.
What we still don't know: how, or whether, the ranking systems detect first-hand experience at scale, versus rewarding the surface features that correlate with it.
The question: why did Google add 'Experience' to E-A-T in December 2022, when 'Expertise' was already there?
The two are not redundant. Per the QRG, expertise is formal or demonstrated knowledge of a subject; experience is first-hand, lived involvement with the specific thing being discussed. A board-certified dermatologist has expertise on a drug's mechanism. A patient who took it for two years has experience of its side effects. The guidelines explicitly note that for some topics, the experienced person produces more trustworthy content than the credentialed one.
The addition reads as a response to a measurable problem: the post-2020 flood of competent-sounding content assembled by writers who had never used the product, visited the place, or held the object. Expertise can be faked or rented; first-hand experience leaves harder-to-fake traces — original photos, specific failure cases, idiosyncratic detail that aggregators lack.
For affiliate and review content this is the load-bearing distinction. A spec-sheet rewrite signals neither. A teardown photo with a thumbnail of your own hand, a specific firmware version where a bug appeared, a measured number you took yourself — these are experience artifacts.
Caveat: 'experience' is not universally required. For a YMYL (your money or your life) medical claim, the guidelines still privilege expertise; lived experience supplements but does not replace clinical authority.
What we still don't know: how, or whether, the ranking systems detect first-hand experience at scale, versus rewarding the surface features that correlate with it.