Van Badham, a columnist at Guardian Australia, has written that human connection now has to be fought for — and she argues it almost entirely with consumption data rather than sentiment. The sharpest number in the piece comes from the French streaming service Deezer: fully AI-generated music accounts for up to 18% of daily uploads and roughly 0.5% of listens. By the Guardian's account, up to 70% of those listens may have been fraudulent.
The same shape repeats across every market she examines. AI-written books already occupy up to 20% of Amazon's ebook catalogue but produce only 12% of sales and 10% of bestsellers in their genres. In commercial fiction, one study found readers rated human-written books higher than AI-written ones on every parameter examined. Bain found that 62% of consumers would not listen to AI-generated music, that 71% did not want AI in books and 65% did not want it in news and magazines. The one category where Bain found prevailing tolerance for AI content was video games.
Service delivery produces even blunter figures. In sales, telling a customer up front that the outbound call they are on is a chatbot cuts the probability of purchase by 79.7%. Deloitte found in 2025 that 74% of customers preferred a human even for ordinary banking questions. Another international study the same year found chatbots to be the least popular support channel available: only 8% of respondents chose one voluntarily.
Against that, the institutions are moving the other way. This week it emerged that Macquarie University has embedded an AI chatbot into weekly teaching on two compulsory psychology courses. Also this week, the National Tertiary Education Union began strike action at the University of Sydney over the spread of AI. Vince Caughley of the NTEU said the union is seeking binding guarantees that AI will not be used to cut jobs, increase workloads, undermine professional and academic judgement, or replace the work and human interaction the university requires. Badham extends her support to other bodies negotiating AI agreements: the Australian Writers' Guild, Australia's entertainment and communications industry union, the US actors' union, the Writers Guild of America, the Canadian performing arts workers' union, German service-sector and actors' unions, and the alliance of Italian dubbing unions. Her reading is that these organisations are defending not only their members' workplace rights but public consent for their industries.
Badham's objection to the Macquarie course is personal and she says so. She lives with depression and relies on careful help from trained specialists, and writes that she cannot trust the conclusions of a supposed expert whose understanding of adult life was formed while being taught by a machine. If she needs a chatbot to stop a panic attack, she notes, she can already open one in a subscription she pays for, in the middle of the night.
The column opens somewhere gentler: a family trip to the cinema a couple of weeks ago, choc-tops in the Australian tradition, talking through the ads, and a full house falling silent as the film began. That was months after Christopher Nolan's The Odyssey opened, shown on film. Badham suggests its pull is no longer only the story of Greeks bearing gifts or the director's name, but the increasing rarity of a cultural reference that can be shared with strangers at all. Public health authorities, she points out, now treat social isolation as an epidemic.
Where she turns to policy, the argument is that public opinion is moving faster than democratic institutions can act on it. Regulation lags under pressure from the economic and political interests of the largest technology owners. Australia is weighing approaches to AI regulation but has introduced none. On social media, Badham's position is that the country is already too late — nazi ideology, radicalisation, dangerous medical misinformation and addictive products spread alongside the platforms. The Australian government is now considering a requirement that users opt in by default to platforms' use of algorithms, which she thinks deserves support if only as a political choice for stated preference over forced consumption. Elon Musk has proposed an AI Odyssey; anyone who believes the algorithm he bought for $44 billion will avoid that outcome, she suggests, should rewatch a few scenes of A Clockwork Orange at a late screening.
The numbers in this column are better than the frame around them, and they are measuring something narrower than the frame claims. Every figure here is a measurement of disclosed AI. The 79.7% collapse in purchase intent is explicitly conditional on telling the customer first — it describes what happens when a chatbot announces itself, not what happens when it does not. The Deezer split is not a verdict from listeners either. Content that takes 18% of uploads and 0.5% of plays, with up to 70% of those plays possibly fraudulent, was never aimed at listeners; it was aimed at the payout mechanism. That is a fraud story wearing a taste story's clothes, and treating it as proof of human preference hands the other side an easy correction.
The gap nobody in this debate is addressing is disclosure. Every statistic Badham assembles depends on the customer knowing, and the Australian proposal actually on the table governs algorithmic recommendation defaults rather than labelling. A 79.7% penalty for announcing a chatbot is not a deterrent to deploying chatbots. It is a very large, very specific incentive not to mention them.
Badham closes by asking who these products are built for, if the customers of AI robots are not themselves robots. Her own Deezer figure contains an answer she does not take up: some of them are not built for customers at all.