Google Search's AI Content Faces Consistency and Safety Concerns

Google Search is increasingly veering into AI territory, offering AI Overviews (AIO) and Featured Snippets (FS) as quick answers to user queries. Yet, in a world where accuracy is critical, their reliability is coming into question. An audit of 1,508 baby care and pregnancy-related queries sheds light on this issue. The findings are eye-opening, revealing a convergence problem between these AI-driven features.

Inconsistency in Information

In 33% of cases, AIO and FS on the same search page deliver inconsistent information. This inconsistency raises a pressing question: If these tools can't align on the basics, how can users trust the precision of the information? For a search engine that's often the first stop for knowledge, this is a significant shortcoming.

Lack of Medical Safeguards

Google's AI features, despite scoring high on relevance, fall short in a critical area. Medical safeguards appear in only 11% of AIO replies and a mere 7% of FS instances. Is relevance enough when the safety of the advice can impact user health? The absence of these safeguards calls into question the responsibility Google holds in disseminating health-related content.

Source of the Matter

Health and wellness websites dominate the source categories, but FS often links to commercial sources as well. The AI-AI Venn diagram is getting thicker, but with FS leading users to potentially biased commercial content, this isn't a partnership announcement. It's a convergence that might mislead rather than inform.

Implications for Public Health

This audit isn't just a critique. It's a wake-up call. With AI cementing its role in information dissemination, the gaps in consistency and safety aren't minor oversights, they're substantive threats to public health information access. The call for stronger quality controls in AI-mediated health information is clear and immediate.

Google's AI features, while innovative, need refinement. The compute layer needs a payment rail, but more importantly, it needs solid guardrails. The future of search might be agentic, but only if it’s accountable. If agents have wallets, who holds the keys?