AI models are quietly becoming one of the most powerful forces in reputation management. That was the core message from Medianet's recent webinar, Opening the Black Box Phase 2, hosted by Head of Marketing Mercedes Carrin with insights from Jacquie Hanna, Head of Insights, and Adam Palmer, Media Analyst. This is the very team behind the pioneering research into generative citation analysis.
The session unpacked findings from Medianet's second phase report into how large language models like ChatGPT, Gemini and Claude cite and represent brands. Phase 2 analysed 957 LLM responses and 11,127 hand-coded citations, narrowing in on RACV's My Melbourne Road campaign and Victorian Government entities. Here's what PR and comms professionals need to know.
One of the clearest findings is that LLMs neutralise controversy. Sentiment toward brands like Westpac and government bodies was consistently more positive in AI responses than in traditional media coverage. Negative issues such as branch closures and job cuts were often missing entirely.
Jacquie explained why: Journalists frame stories with editorial judgement, which shapes public sentiment, but LLMs work differently. They favour structured, authoritative owned content over news coverage. In Phase 1 of the research, 45% of messaging in Commonwealth Bank’s LLM responses came directly from the bank's own website, and Phase 2 found the same pattern with Victorian Government sources.
For comms teams managing a crisis, this is a genuine strategic asset. Deep, well-maintained owned content can do a lot of the reputational heavy lifting, balancing out negative coverage elsewhere. As Jacquie puts it:
Traditional media issues typically fade within two to four weeks. LLMs don't work that way. Adam pointed out that once an issue is picked up and indexed by an AI model, it can stay part of that model's responses indefinitely.
He used the 2019 Banking Royal Commission as an example. Years later, it still shows up prominently in LLM responses about banks. By contrast, a 2024 ASIC ruling against RACQ that generated significant traditional media coverage barely registered in AI responses twelve months on. Adam sums it up:
The practical takeaway is about resourcing. Short-term issues in the news cycle need short-term channel management. But anything likely to be indexed by AI models, particularly formal records like royal commission submissions, deserves long-term attention because it may never fully disappear from AI outputs.
The report found that corporate and government-owned websites accounted for 67% of citations, while major metro newspapers made up just 4%. Social platforms, including Facebook, Reddit and YouTube, accounted for around 11%.
Jacquie noted that this is shaped specifically by the Australian media landscape. Much of the country's quality journalism sits behind paywalls and crawl blockers, making it largely invisible to LLMs. Freely available owned content fills this gap.
YouTube stood out as a channel comms teams can actually control. In the RACV research, 75%of YouTube citations came from RACV's own channel. LLMs tend to read video metadata and transcripts closely, so well-titled, well-structured video content becomes genuinely citable. She also noted that view counts don't matter here. Structure does.
Facebook is a different story. Much of the content driving Facebook citations wasn't brand-owned, coming instead from public posts, community groups and third-party pages. Adam's advice was to monitor your Facebook footprint the way you'd monitor press coverage, since you don't control who's talking about you there.
Traditional media rankings don't translate directly to AI visibility. The report compared Youi and RACQ as an example. Youi held just 7% share of traditional media coverage but achieved 26% of LLM responses. RACQ, with 28% media share, appeared in only 15% of LLM responses.
This means comms professionals now need to treat LLM visibility as its own discipline, separate from media share of voice, and measure it with the same rigour applied to traditional coverage.
Adam raised an important nuance around brand mentions. Some LLM responses cited RACV's content verbatim without naming the brand at all. Whether this is a risk depends on the campaign's goal.
For an education-focused campaign like My Melbourne Roads, the message still gets mentioned even without attribution. For organisations promoting a specific product or technology, losing that attribution is a real missed opportunity.
Across the session, three practical priorities came through clearly.
Medianet has made this easier through Medianet Labs, a space inside the Medianet platform giving users early access to tools including LLM Monitoring, which tracks how brands are mentioned and cited across Gemini, Claude and ChatGPT. It's available now to Medianet account holders under the Medianet Labs tab. So if you are a Medianet user, you can head there and start monitoring your brand on LLMs now.
Opening the Black Box Phase 2 report is available to download.