LLMs skew positive on brand sentiment and quietly neutralise controversy for comms teams, new research finds.
Medianet
-
The second phase of the original Medianet research found that LLMs mentions are predominantly favourable towards Australian banks, even while key issues that usually drive negative coverage, such as job cuts and branch closures, dominated on traditional media.
-
Major Australian newspapers only made up 4% of all citations, in part because Australian publishers operate behind paywalls or have blocked AI crawlers. Models drew directly from corporate content and surfaced on-message material consistently.
-
The research urges media professionals to treat LLMs as stakeholders with their own and distinctive rules while also highlighting the increased value of journalism to hold power to account.
Organisations managing difficult issues now have a reputational ally they didn't ask for. Medianet’s new release of the landmark research titled “Opening the Black Box” finds that large language models (LLMs) consistently neutralise negative coverage by turning to owned content over journalism and producing responses that read more like information bulletins than news.
The Australian study was the first to apply traditional media analysis rigour to the outputs of LLMs. This second release deepens the original findings across two connected research phases spanning 1,357 LLM responses and 11,127 hand-coded citations across ChatGPT, Perplexity and Google AI Summaries.
The neutralising pattern held across sectors and organisation types. In Phase 1, Westpac returned 97% favourable in LLM responses and CommBank 100%, while traditional media coverage of both banks for the same period sat at 30–60% favourable, dominated by reporting on job cuts, AI restructuring and branch closures. In Phase 2, the Victorian Government, frequently the subject of contentious news coverage, produced only 14% unfavourable AI mentions against 50% unfavourable in traditional media on the same topics.
“Journalists frame a topic and brands respond. LLMs flip the dynamic. They do the balancing work themselves and no comms team asked them to. For organisations managing complex policy issues or long-running reputation challenges, that is a significant and largely unrecognised strategic asset,”said Jacquie Hanna, Head of Insights at Medianet Insights.
The research identifies the mechanism driving the neutralisation. LLMs are not drawing on the journalism that shapes traditional media sentiment, they are drawing predominantly on the content produced by brands, or “ owned content”.
Across 11,127 manually re-coded citations, corporate and government owned content accounted for 67% of everything the models cited. Instead, major Australian national and metropolitan newspapers only accounted for 4%. This supports what was observed on Phase 1 where 45% of CommBank’s AI mentions contained messaging drawn verbatim from the bank’s own website. The models were not balancing journalism against corporate messaging. In large part, they were only seeing the corporate messaging.
That dynamic is specific to the Australian market where major publishers including News Corp mastheads, Nine titles and the ABC operate behind paywalls or have blocked AI crawlers from accessing their content. The result is that the independent journalism that holds organisations to account in traditional media is, in many cases, invisible to the models.
“The neutralisation finding has two sides. For brands managing difficult issues, LLMs may be doing reputation work that no comms team could replicate. But for the public relying on AI to understand those same organisations, the picture they receive is stripped of the scrutiny that earned media provides. That is a finding with implications well beyond communications strategy.” said Amrita Sidhu, Managing Director of Medianet.
The research also surfaces significant divergence in AI share of voice compared to traditional media standings. Insurance company Youi held 7% of traditional media share of voice but 26% of LLM responses. RACQ moved the other way, from 28% in traditional media to 15% in LLMs. In Phase 2, 137 LLM responses cited RACV content without mentioning RACV by name, showing that brand messaging had become authoritative enough that the model was treating it as a source rather than a brand asset.
When looking at the themes and sentiment towards brands in LLM responses, the research identified a reputational lag that represents again an opportunity and a risk. LLMs operate on a different news cycle where content takes longer to surface, but will be remembered forever. While some media issues for financial organisations did not show in LLM responses, others did despite of them being more than six year old, such as the 2019 Royal Banking Commission.
“A brand that looks clean in AI responses during a crisis may be drawing false reassurance. An issue the team considers resolved may still be live in AI long after traditional media has moved on”, added Sidhu.
The research concludes that LLMs should be treated as a distinct reputational channel requiring its own monitoring framework, one that accounts for the neutralisation tendency, the slower news cycle, and the dominance of owned content as a citation source.
Opening the Black Box - Phase 2, is available for download here.
About the report
Opening the Black Box analysed 1,357 LLM responses and 11,127 individually hand-coded citations across two connected research pilots conducted in November 2025 and March 2026, covering ChatGPT, Perplexity and Google AI Summary. The research included organisations drawn from the financial services, automotive and government sectors.
About us:
Medianet is Australia’s #1 media intelligence and communications platform, providing integrated solutions for media outreach, monitoring and media analysis. insights. Medianet provides award winning, expert and actionable business insights for some of Australia’s top organisations, applying AMEC-accredited methodology across more than 40 years of Australian media intelligence.
Contact details:
Amrita Sidhu
Managing Director, Medianet
+61 481 177 686 [email protected]
Jacquie Hanna - Head of Insights and Strategy, Medianet Insights
+61 428 226 214
[email protected]
Mercedes Carrin
Head of Marketing, Medianet
+61 430 729 397 [email protected]
