AI for YMYL Content: What the Evidence Reveals

When creating content for Your Money or Your Life (YMYL) topics, those directly affecting health, finances, safety, or general well-being, publishers must recognize that Google enforces significantly higher standards. While artificial intelligence writing tools promise efficiency, they consistently fall short in delivering the accuracy, authority, and firsthand experience that YMYL subjects demand. Evidence shows that AI systems frequently generate unsupported claims and factual errors, making them unreliable for these sensitive areas.

Google subjects YMYL content to intense algorithmic scrutiny. According to its Search Quality Rater Guidelines, pages covering YMYL topics must meet “very high Page Quality rating standards” and “require the most scrutiny.” The guidelines define YMYL as any subject that “could significantly impact the health, financial stability, or safety of people.” For YMYL queries, Google assigns greater weight to factors like authoritativeness, expertise, and trustworthiness. The March 2024 core update, which aimed to reduce low-quality content by 40%, heavily impacted finance and health websites, underscoring this strict approach.

The guidelines create a two-tier quality system. While everyday expertise may suffice for general content, YMYL material must demonstrate “extremely high” levels of E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness). Content failing to meet these standards receives Google’s “Lowest” quality rating. Given these elevated requirements, AI-generated content struggles to compete.

Error rates for AI in YMYL contexts are alarmingly high. A Stanford HAI study from February 2024 tested GPT-4 with Retrieval-Augmented Generation (RAG) and found that 30% of individual statements were unsupported, with nearly half of all responses containing at least one unverified claim. Google’s Gemini Pro managed only 10% fully supported responses. In one instance, GPT-4 RAG provided treatment instructions for the wrong medical equipment, a mistake with potential real-world harm.

Money.com evaluated ChatGPT Search on 100 financial questions in late 2024, reporting just 65% accuracy. The tool sourced answers from unreliable personal blogs, omitted recent rule changes, and failed to caution against risky strategies like “timing the market.” In the legal domain, Stanford’s RegLab analyzed over 200,000 queries and found hallucination rates between 69% and 88% for top AI models. These systems incorrectly stated court holdings at least 75% of the time, with 439 documented instances of AI-generated hallucinations appearing in legal filings.

Men’s Journal published an AI-written health article in February 2023, which Dr. Bradley Anawalt of the University of Washington Medical Center reviewed. He identified 18 specific errors, noting “persistent factual mistakes and mischaracterizations of medical science.” The article incorrectly equated medical terms, suggested unsupported diet-symptom links, and issued unfounded health warnings. Dr. Anawalt described the content as “flagrantly wrong about basic medical topics” yet possessing “enough proximity to scientific evidence to have the ring of truth,” a dangerous combination that misleads readers who cannot easily detect inaccuracies.

Even when AI states facts correctly, it lacks what Google prioritizes: genuine experience. In December 2022, Google updated its evaluation framework to E-E-A-T, adding “Experience” as the first component. The guidelines now ask whether content “clearly demonstrate[s] first-hand expertise and a depth of knowledge,” such as that gained from using a product, providing a service, or visiting a location. This criterion directly challenges AI’s limitations. While an AI might define temporomandibular joint disorder (TMJ) accurately, only a specialist who treats TMJ patients can answer practical questions about recovery timelines, common patient mistakes, or when to consult a specialist versus a general dentist.

Google’s content quality questions further reward originality, asking, “Does the content provide original information, reporting, research, or analysis?” and “Does the content provide insightful analysis or interesting information that is beyond the obvious?” The company explicitly warns against “mainly summarizing what others have to say without adding much value,” which is essentially how large language models operate.

This lack of originality leads to another critical issue: content homogenization. UCLA researchers identified a “death spiral of homogenization,” where AI systems gravitate toward statistically probable outputs, erasing diversity. Oxford and Cambridge studies illustrated this by training an AI on dog breeds, over time, the model produced only the most common types, a phenomenon termed “Model Collapse.” A Science Advances study concluded that while generative AI can boost individual creativity, it reduces the collective diversity of novel content. For YMYL topics, where differentiation and unique expertise are competitive advantages, this convergence is detrimental. If multiple financial advisors use the same AI tool for investment guidance, their outputs will be nearly identical, giving Google and users no reason to prefer one over another.

Google’s March 2024 update specifically targeted “scaled content abuse” and “generic/undifferentiated content” that rehashes widely available information without fresh insights. To determine whether content genuinely originates from the credited expert, Google uses its knowledge graph to verify authors’ credentials. Established professionals, such as doctors with Google Scholar publications, lawyers with bar registrations, or financial advisors with FINRA records, leave verifiable digital footprints. Google analyzes writing style, terminology, sentence structure, and topic focus to create an author signature; deviations from this pattern can raise authenticity concerns.

Building authority requires consistency. Publishers should link author bylines to detailed biography pages listing credentials, jurisdictions, specializations, and verifiable professional profiles. Most importantly, experts should either write the content themselves or conduct thorough reviews, not just fact-checking, but ensuring the voice, perspective, and insights reflect their genuine experience.

The consequences of YMYL misinformation are severe. A 2019 University of Baltimore study estimated that misinformation costs the global economy $78 billion annually. In 2024, deepfake financial fraud impacted half of all businesses, with average losses of $450,000 per incident. Unlike errors in non-YMYL content, which may cause minor inconvenience, YMYL mistakes can lead to injury, financial loss, and eroded trust. U.S. federal law imposes penalties of up to five years in prison for spreading harmful false information, 20 years if severe bodily injury occurs, and life imprisonment if death results. Between 2011 and 2022, 78 countries enacted misinformation laws.

Validation is crucial in YMYL because consequences compound over time. Medical decisions delayed by false information can allow conditions to worsen irreversibly. Poor financial choices create lasting hardship. Incorrect legal advice may result in loss of rights. These outcomes are permanent.

Readers seeking YMYL content want more than textbook definitions. They look for connection with practitioners who understand their situation. They seek answers to questions like, “What do other patients commonly ask?” “What typically works?” “What should I expect?” These insights stem from years of hands-on practice, not training data. When a doctor notes, “the most common mistake I see patients make is…,” it carries a weight that AI-generated advice cannot replicate. Authenticity builds trust, which is essential when people are making decisions about their health, finances, or legal rights.

Organizations producing YMYL content face a strategic choice: invest in genuine expertise and unique perspectives, or risk algorithmic penalties and reputational harm. Google’s introduction of “Experience” to E-A-T in 2022 targeted AI’s inability to provide firsthand knowledge. The Helpful Content Update penalizes content that merely summarizes existing information, exactly how LLMs function. With AI error rates ranging from 18% to 88% in YMYL contexts, the risks clearly outweigh any benefits.

Experts don’t need AI to write for them; they need support in organizing their knowledge, structuring their insights, and making their expertise accessible. This represents a fundamentally different role than content generation.

Looking forward, the true value of YMYL content lies in knowledge that cannot be scraped from existing sources. It comes from the surgeon who knows what questions patients ask before procedures, the financial advisor who has guided clients through economic downturns, or the attorney familiar with which arguments persuade specific judges. Publishers treating YMYL content as a volume game, whether through AI or content farms, face a difficult path. Those treating it as a credibility signal have a sustainable, trustworthy model. AI can serve as a tool within the content creation process, but it cannot replace human expertise.

(Source: Search Engine Journal)