Carl Hendy challenges Google's AI Search preference claims, pointing to data gaps in the Google/Ipsos survey that the industry uses as its main source.
A LinkedIn post by Carl Hendy, founder of Audits.com, has reignited a debate that practitioners in search and digital marketing have been circling for months: when Google says users prefer AI in Search, what exactly is that claim built on?
The post, published in late May 2026, landed at a precise moment. DuckDuckGo had just reported a 30.5% peak increase in U.S. app installs in the week following Google’s I/O announcements, which included a broader rollout of AI Overviews and a more seamless AI Mode. And yet Google’s official position – repeated at industry events and in shareholder communications – is that users are happier with AI in Search, searching more, and generally receptive to the new experience.
Hendy’s argument is methodological, not impressionistic. He traced every public claim about user preference for AI Search back to its original source. What he found, according to his post, was that “every claim that users prefer AI Search traces back to one source. Google.” Internal Search data is Google’s. AI Mode usage numbers are Google’s. The “users are happier” message is Google’s.
The only third-party research Hendy could identify in the public mix was the Google/Ipsos Multi-Country AI Survey 2026, published on January 15, 2026. That survey – the third annual wave, covering 21 countries and roughly 21,000 adults interviewed between September 22 and October 10, 2025 – was commissioned by Google. Ipsos conducted the fieldwork.
According to Google and Ipsos, 66% of respondents reported using an AI tool or application in the last 12 months, up 18% from 2024 and 28% from 2023. Those are meaningful headline numbers. But Hendy’s scrutiny lands on a different question: what the survey actually measured.
“The only third-party research in the mix is Google/Ipsos, and that is general AI research,” Hendy wrote. “It asks whether people use AI tools and how they feel about AI broadly. It never asks the actual question: when you search, would you rather have the AI answer or the web results?”
That gap is significant. The survey documents usage and broad sentiment toward AI tools, including ChatGPT, Gemini, and Claude – it is not a search preference study. A respondent saying they used a generative AI tool in the last year, or that they find AI broadly useful for work, tells you nothing about whether they would actively choose an AI-generated search result over a traditional list of links for a given query.
The topline figures that Google cites most often from the Google/Ipsos “Our Life With AI 2025” report – the prior wave, fielded in autumn 2025 – show that 66% of respondents globally have used an AI tool in the past 12 months. But 21% describe themselves as using AI “a lot.” The headline adoption figure and the intensity-of-use figure are separated by 45 percentage points.
The trust figures are more striking still. According to the survey, 16% of respondents globally trust an AI application or tool “a great deal” to give accurate information. That global average, as Hendy’s annotation of the data makes clear, is pulled upward dramatically by a small number of markets. Nigeria sits at 49%, India at 45%. By contrast, the US records 3%, the UK 12%, and Australia 12%.
“The 16% global average masks a 3% to 49% spread across the 21 countries surveyed,” Hendy wrote in an annotation shared with the post. “A blend, not a verdict.”
The same market-level divergence runs through related questions. Willingness to use AI more – respondents saying they are open to doing more with AI but want to be more confident about how – sits at 57% globally, which sounds encouraging. But the US figure is identical, 57%, despite having among the lowest trust numbers in the dataset. That alignment suggests the question is measuring aspiration rather than satisfaction.
This is the core of Hendy’s argument, and it is a methodological point that applies well beyond this particular survey. When AI answers are embedded in the default search experience – which AI Overviews are, for most Google users without multi-step intervention – usage rises automatically. That rise does not distinguish between users who wanted the AI answer and users who simply received it.
“And usage is not preference,” Hendy wrote. “If AI sits inside the default search experience, usage rises automatically. That does not mean users chose it.”
The point has practical resonance for search marketers and publishers. Google has faced sustained criticism from publishers over the claim that AI Overview clicks are higher quality, with the company citing user engagement signals to defend reduced click-through volumes. Google’s own John Mueller reinforced that messaging in July 2025, though the position drew direct challenges from publishers documenting traffic losses. A Pew Research Center study published in July 2025 found that website link clicks fell by half when AI summaries appeared, a finding Google contested on methodological grounds.
The absence of an opt-out has been a recurring point of tension. As PPC Land documented in May 2024, there is no built-in setting to disable AI Overviews entirely within standard Google Search, though a “Web” filter was introduced as a partial workaround. One commenter on Hendy’s LinkedIn thread made this point directly: “It’s absurd to me that we can’t just toggle off AI search / results. Of course it’s going to look like we prefer AI search when the only way to get rid of it is like five clicks deep in settings that most users never touch.”
At a Google event in Sydney, according to Hendy, the messaging was structured around three propositions: AI is growing Search, people are searching more, and users are happier. Hendy observed that when he looked at the evidence cited, “data sources were blended together. Some references were years old. And the claim that ‘users are happier’ appeared to rest on Google’s own recent shareholder update, rather than an independent source.”
Shareholder communications serve a different purpose than academic or market research. They are written for investors, not practitioners, and they operate within a different evidentiary standard. Using them as the sourcing basis for a user preference claim conflates two distinct categories of information.
Other practitioners in the thread pushed back in similar terms. Lily Ray, VP of SEO at Amsive, described a persistent disconnect between how people talk about AI and how Google repeatedly claims users love these features. Greg Sterling, a long-standing local search analyst, noted that his own survey data suggests the picture is contextual – that people may prefer the conversational interface of AI without trusting or preferring the actual results. “It very much depends on the category,” Sterling wrote.
Whatever the survey data shows about attitudes, recent behavioral data offers a different lens. Following Google’s I/O 2026 announcements on its AI Search overhaul, DuckDuckGo reported U.S. app installs rising 18.1% week-over-week on average in the May 20 to May 25 period, compared to May 13 to May 18, with growth sustained for six consecutive days and peaking at 30.5% on May 25. On iOS, week-over-week growth hit a 33% average, peaking at 69.9%.
DuckDuckGo CEO Gabriel Weinberg characterized the growth explicitly as a response to Google’s approach. “Google is force-feeding AI with no way to opt out,” Weinberg said. “As a result, their results are getting worse, not better. We want to be the place that puts users in charge and allows them to decide how much or how little AI they want.”
App analytics firm Apptopia independently reported a 29% increase in average daily U.S. downloads and a 12% increase globally over the same period. Visits to DuckDuckGo’s AI-free search page, noai.duckduckgo.com, averaged 22.7% week-over-week growth, peaking at 27.7% on May 24.
These are still small absolute numbers. DuckDuckGo’s global search market share stood at roughly 0.9% as of Q4 2024, and the install spikes described above represent a shift from a small base. But as a behavioral signal – users actively seeking an alternative specifically because of AI Search – they are more direct evidence of preference than usage attribution within a default experience.
For search marketing professionals and SEO practitioners, the distinction between usage and preference has direct operational implications. Campaign budgeting, content strategy, and publisher revenue models are all being recalibrated around assumptions about how users interact with AI search surfaces.
Independent research published in April 2026 found that only 15% of users trust AI search platforms “a lot”, with 63% double-checking AI results against other sources and 51% describing results as a “walled garden.” That 15% figure aligns closely with the 16% global “great deal” trust figure in the Google/Ipsos survey – suggesting both studies are capturing the same underlying ceiling.
A broader AI trust study published in October 2025 documented AI awareness reaching 90% and weekly usage nearly doubling to 34%, while trust – particularly for journalism and health – remained persistently low. High awareness and rising usage do not convert automatically into trust or preference.
The market split in the Ipsos data matters for another reason: most of the AI tool usage growth driving the headline numbers is concentrated in emerging markets. Nigeria reports 88% AI tool usage in the past 12 months, India 86%, UAE 84%. These are markets where AI tools may function differently as an information resource – where the infrastructure gap between AI and traditional search is smaller, and where trust in AI accuracy is substantially higher than in Western markets.
The US at 54%, Australia at 63%, the UK at 62%, and Germany at 56% tell a more cautious story. These are also, not coincidentally, the markets that generate the bulk of advertising revenue and where the SEO and publisher industries feel the effects of AI search most acutely.
Hendy notes that the natural vehicle for answering the preference question used to exist: Google Surveys, the company’s own market research tool, was discontinued. “A quick, market-by-market preference test is exactly what it was built for, and exactly what is missing here.”
The study he describes would need to do something the Google/Ipsos surveys have not attempted: show the same query to comparable users with and without an AI-generated answer, and ask which result they prefer. It would need to be market-specific, task-specific, and conducted by a party with no commercial stake in the outcome.
Until that research exists, Hendy’s conclusion holds: the preference claim has one source, and that source sells the product. “Even if usage keeps climbing, that is a rollout succeeding, not a preference being measured,” he wrote. “Until someone independent asks users, by market and by task, whether they would rather have the AI answer or the web results, I am treating this as a rollout and usage story. Not a user preference story.”
Who: Carl Hendy, founder of Audits.com and an SEO and search growth consultant whose clients have included BlackRock, Aviva, British Airways, and John Lewis. His post drew responses from practitioners including Lily Ray of Amsive and Greg Sterling, a local search industry analyst.
What: Hendy published a detailed methodological critique on LinkedIn arguing that every public claim about user preference for AI Search traces to Google as the sole source. His central point is that the Google/Ipsos survey – the only third-party data in the mix – measures general AI tool usage and broad sentiment, not search preference specifically. He further distinguishes between usage growth within a default experience and actively chosen preference.
When: The post appeared in late May 2026, published approximately one week after Google’s I/O 2026 developer conference, at which the company announced a significant expansion of AI Overviews and AI Mode. The underlying Ipsos fieldwork it references was conducted between September 22 and October 10, 2025.
Where: The critique was posted on LinkedIn and engaged with by practitioners globally. The Google/Ipsos survey itself was published in Washington, DC on January 15, 2026, and covers 21 countries across North America, Europe, Asia-Pacific, South America, Africa, and the Middle East.
Why: The question of whether users genuinely prefer AI-generated search results – as opposed to simply receiving them within a default experience – has direct consequences for how publishers set traffic expectations, how advertisers model search behavior, and how SEO professionals prioritize content investment. Without independent, market-specific, task-specific preference data, the marketing industry is operating on claims sourced from the platform with the greatest commercial interest in the answer.
Your go-to source for digital marketing news. Get the latest updates from Google, Meta, Amazon, and The Trade Desk. Stay informed on ad tech innovations, programmatic trends, and policy changes.
Google says users prefer AI Search – but is that actually true? – PPC Land
Leave a Comment
