A study commissioned by the New York Times has concluded that Google AI Overviews contain significant problems with both grounding and factual accuracy. This is not a fringe blog post or a competitor making noise. It is a well-resourced media organisation paying for independent research to scrutinise one of the most consequential changes to search in a generation. For anyone whose brand depends on being cited correctly in AI-generated results, this deserves a clear-eyed response.
The findings point to two distinct failure modes: grounding issues (where the AI produces content not properly supported by the sources it pulls from) and accuracy issues (where the information surfaced is simply wrong). These are not the same problem, and they do not have the same fix. Understanding the distinction matters for how you think about your visibility work.
Grounding Failures Are a Structural Problem
Grounding refers to whether an AI system's output is actually anchored to the source material it references. When grounding breaks down, the AI generates a response that may cite a source but misrepresents or extrapolates from what that source actually says. For brands, this is particularly dangerous. Your content could be listed as a citation underneath a claim you never made - or a claim that directly contradicts your position.
This is not a hypothetical. It is a known behaviour in large language models, and the study suggests AI Overviews are not immune to it. The implication for GEO practitioners is that getting cited is not the same as being represented accurately. These are two different outcomes, and optimising purely for citation volume without auditing what the AI actually says about your brand is an incomplete strategy.
The practical response here involves regular monitoring - not just of whether your brand appears in AI Overviews, but of what is attributed to you. Tools that scrape AI Overview content across a range of queries related to your brand, products, and category are worth building into your reporting workflow. If you find grounding errors, the corrective action is to make your source content more explicit, structured, and unambiguous. The AI needs less room to interpret.
Accuracy Failures Are an Audience Problem
Accuracy failures are different. These occur when the AI Overview states something factually incorrect - irrespective of what the cited sources say. For search users, this erodes trust in AI-generated summaries. For brands, it creates a scenario where misinformation about your products, pricing, policies, or capabilities could be served to users at the exact moment they are forming a purchasing decision.
The concern is amplified by the placement of AI Overviews. They sit at the top of the results page, above organic listings, and carry an implied authority. A user who reads an inaccurate summary about your brand is unlikely to scroll down and fact-check it against your actual website. The AI response is often the final word.
This is why schema markup and structured data are not optional extras in a GEO programme - they are risk management tools. The more clearly and formally your key facts are encoded (product specifications, prices, service descriptions, FAQs), the harder it becomes for the model to generate something plausibly wrong about you. You are not just optimising for visibility. You are reducing the surface area for inaccuracy.
What This Means for How You Brief Clients
If you are advising UK businesses on AI search visibility, the findings from this study change the conversation you need to have with clients. The goal can no longer be framed purely as 'get your brand into AI Overviews'. The goal needs to include 'ensure your brand is represented correctly when it appears'. Those are meaningfully different briefs.
Clients in regulated sectors - financial services, healthcare, legal - face particular exposure here. An inaccurate AI Overview that misrepresents a financial product or a medical recommendation is not just a brand problem. It may carry compliance implications. Proactive monitoring and rapid escalation processes (via Google's feedback mechanisms and Search Console) need to be part of the service offering, not an afterthought.
For ecommerce brands, the priority is keeping product and pricing information accurate and consistently structured across every touchpoint the AI is likely to index - product pages, feeds, structured data, and third-party listings. Inconsistency across these sources gives the model more opportunity to generate something plausible but wrong.
The Competitive Angle Other Platforms Are Watching
It is worth setting this in a broader context. AI Overviews are Google's most visible AI product in search, but they are not the only game in town. Perplexity, ChatGPT Search, and Gemini all face similar grounding and accuracy challenges - this is an industry-wide issue with LLM-generated summaries, not a Google-specific failing. The New York Times study focuses on Google because of its market dominance, but the structural risks apply across platforms.
What this does do is create pressure on Google to improve its grounding mechanisms and source verification processes. That pressure is likely to result in changes to how AI Overviews are generated over the coming months - potentially making highly authoritative, well-structured content even more important as a signal. Brands that invest in content quality now are likely to be better positioned as these systems tighten their standards.
The Practical Takeaways
Start auditing your AI presence with accuracy in mind, not just presence. Run brand-relevant queries in AI Overviews, ChatGPT, and Perplexity regularly. Note what is attributed to you and whether it is correct. This should become a standard part of your monthly reporting.
Invest in clarity at the content level. Ambiguous copy, vague product descriptions, and inconsistent messaging across pages all increase the risk of misrepresentation. Write for humans first, but write with the precision that AI systems need to quote you accurately. FAQ content, structured Q&A, and clear factual statements in body copy all reduce interpretive latitude.
Finally, do not conflate AI search investment with passive optimism about how these systems work. The New York Times study is a reminder that AI-generated results are probabilistic outputs, not verified facts. Treating them as a reliable amplification channel requires active management - not just content creation and hope. The brands that take AI visibility seriously as an ongoing discipline, rather than a one-time technical fix, will be far better placed to handle the failures these systems still routinely produce.