We asked Google AI about big tobacco. Its answer was missing some controversial history
RNZ
Overall Assessment
The article investigates how Google's AI overviews may reproduce corporate PR by relying on self-published sources, using Philip Morris as a case study. It balances expert criticism with platform responses and acknowledges the dynamic nature of AI outputs. The framing emphasizes systemic risk in AI-mediated information rather than isolated error.
"The fan-out queries and generated results could change, based on the wording of the question, what's happening in the news, who was searching, or at what point in time they searched, Sanders said."
Loaded Language
Headline & Lead 85/100
The article opens with a clear, factual headline and lead that accurately represent the core issue — AI-generated summaries omitting critical context about tobacco companies — without resorting to sensationalism or misleading claims.
✕ Headline / Body Mismatch: The headline accurately reflects the article's focus on Google AI's omission of controversial history in its summary of Philip Morris International, without exaggeration or emotional manipulation.
"We asked Google AI about big tobacco. Its answer was missing some controversial history"
Language & Tone 88/100
The tone remains largely objective, with charged language properly attributed to sources rather than embedded in the narrative, and technical explanations presented neutrally.
✕ Loaded Adjectives: The article uses neutral language to describe the AI behavior, letting expert quotes carry the critical tone rather than editorializing in the reporter's voice.
"The AI overview produced by Google on this occasion was pretty upbeat."
✕ Loaded Language: It avoids scare quotes or dog whistles, and uses precise, factual language when describing corporate practices and AI mechanics.
"The fan-out queries and generated results could change, based on the wording of the question, what's happening in the news, who was searching, or at what point in time they searched, Sanders said."
✕ Editorializing: The article includes strong criticism from experts but attributes it clearly, avoiding editorial endorsement of loaded terms like 'regurgitation' or 'incredibly misleading' without attribution.
""I consider this to be incredibly misleading," she said, after examining example AI summaries provided by the ABC and conducting her own search."
Balance 92/100
The article achieves strong source balance by including expert critics, technical analysts, and official corporate and platform responses, while transparently noting non-responses.
✓ Comprehensive Sourcing: The article includes multiple expert voices with relevant expertise: public health (Freeman), AI/SEO (Sanders, Nagappa), and media research (Kasianenko), ensuring diverse and credible sourcing.
"University of Sydney public health professor Becky Freeman said she was shocked to read an AI overview that was, in her view, "essentially a regurgitation of Philip Morris International's PR materials"..."
✓ Proper Attribution: It includes Google's official response, giving the platform a chance to explain its practices, even when the explanation is not fully satisfying.
"The content and links that are included in AI Overviews are dynamic and change based on the information that is most relevant, helpful, and timely for a given search," a Google spokesperson said."
✓ Proper Attribution: The article notes that companies like Philip Morris, British American Tobacco, and James Hardie were asked about SEO strategies but did not respond, preserving transparency about one-sided sourcing.
"The ABC asked Philip Morris, British American Tobacco and James Hardie if they had used these strategies... None responded to the ABC's questions."
Story Angle 90/100
The story is framed as a systemic concern about AI's reliance on corporate sources, supported by multiple examples and expert analysis, rather than a sensationalized exposé or isolated incident.
✕ Framing by Emphasis: The article frames the story as a systemic issue about AI information reliability, not just a one-off error, by comparing multiple companies and citing research on commercial source prominence.
"What was apparent was that when AI overviews sourced information directly from the company's own websites, it was more likely to adopt a promotional tone."
✕ Narrative Framing: It avoids reducing the issue to a simple conflict between 'tech bad' and 'companies bad', instead exploring the structural incentives for businesses to optimize for AI visibility.
"Now companies are racing to optimise their own websites to ensure AI uses them as the source for information in order to secure a more favourable write-up or future business traffic."
Completeness 94/100
The article provides rich historical and systemic context, explaining not only the tobacco industry's past misconduct but also the broader implications of AI relying on corporate self-representations, while acknowledging the fluidity of AI outputs.
✓ Contextualisation: The article provides substantial historical context about Philip Morris, including its history of misleading the public about smoking risks and court findings, which is essential for understanding the significance of the AI omission.
"Alongside other big tobacco brands, PMI's former parent company (Altria) and US associate (Philip Morris USA) were found by a US court to have misled the public about the health risks of smoking over decades, despite internal research dating back to the 1980s that proved Philip Morris knew there were harms."
✓ Contextualisation: It extends the context beyond Philip Morris to other companies with problematic histories (British American Tobacco, James Hardie), showing a systemic pattern rather than an isolated incident.
"British American Tobacco, for example, was continuing to focus on developing its "A Better Tomorrow" strategy to reduce harms, one AI overview read, linking back to the company's own website."
✓ Contextualisation: The article acknowledges the dynamic nature of AI results, including that Google updated the Philip Morris AI overview after media inquiry, which prevents overgeneralization from a single static snapshot.
"After the ABC contacted Google and used the Philip Morris AI overview as an example, we subsequently received AI overview results from the same query that acknowledged the company's controversial practices..."
Big Tech is portrayed as untrustworthy due to systemic reliance on corporate self-representation in AI outputs
The article frames Google's AI overviews as reproducing corporate PR by relying on self-published sources, raising concerns about integrity and transparency in AI-mediated information. Expert criticism is highlighted, including the claim that the AI output is 'essentially a regurgitation of Philip Morris International's PR materials' and 'incredibly misleading'.
"University of Sydney public health professor Becky Freeman said she was shocked to read an AI overview that was, in her view, "essentially a regurgitation of Philip Morris International's PR materials", with no mention of negative press around the company."
AI is framed as posing a risk to information integrity by amplifying corporate narratives without critical context
The article emphasizes the potential dangers of AI-generated summaries omitting crucial historical and health-related context, particularly when sourcing from company websites. It warns users to 'exercise caution' due to the lack of regulatory guardrails.
"Nagappa said users should exercise caution when using AI and consider why AI might be serving the information that they see - because there are really no rules right now."
Public health is portrayed as endangered by AI systems that omit critical health risks and corporate misconduct
The article repeatedly highlights the omission of public health harms — such as smoking-related diseases, youth nicotine addiction, and asbestos deaths — in AI summaries, suggesting a systemic failure to protect public awareness.
"The proliferation of these products is understood have contributed to rising rates of nicotine use and addiction in young people, slowing generations of progress at reducing smoking harms."
Corporate self-representation in AI is framed as illegitimate when it omits harmful histories and ongoing public health impacts
The article contrasts corporate narratives with known harms and court findings, particularly regarding Philip Morris and James Hardie, to question the legitimacy of allowing companies to shape their public image via AI without critical counterbalance.
"It did not mention that asbestos products still contribute to the deaths of thousands of Australians annually, or that the country has one of the highest rates of mesothelioma, a type of cancer caused by asbestos, in the world."
AI is framed as failing in its role to provide balanced, accurate information when it relies uncritically on commercial sources
The article documents how AI overviews sourced from company websites produce promotional content, and notes research indicating AI gives 'prominence to commercial sources', suggesting a systemic performance failure in information curation.
"Kasianenko said the team's recent unpublished research, which examined searches for conspiracy theories, found AI overviews tended to give "prominence to commercial sources"."
The article investigates how Google's AI overviews may reproduce corporate PR by relying on self-published sources, using Philip Morris as a case study. It balances expert criticism with platform responses and acknowledges the dynamic nature of AI outputs. The framing emphasizes systemic risk in AI-mediated information rather than isolated error.
An investigation reveals that Google's AI-generated search summaries for companies like Philip Morris International and James Hardie often rely heavily on the companies' own websites, omitting well-documented histories of public health harm. Experts warn this could mislead users by presenting sanitized corporate narratives without critical context. Google says the results are dynamic and based on relevance, not corporate influence.
RNZ — Business - Tech
Based on the last 60 days of articles
RNZ
