Skip to main content

Press enter or click to view image in full size

I want to keep writing about AI and I want it to remain free to read. Help me do that by subscribing to my newsletter, Future Proof (free and paid tiers available).

There’s been a story doing the rounds this week, amongst podcast folk, which is extraordinary in its gratuitous audacity. It was revealed by the Hollywood Reporter that a company, Inception Point AI, was producing some 3,000 podcast episodes a week, using one simple hack: AI.

The episodes, the article added, were entirely generated using artificial intelligence software. Someone on Inception Point AI’s 4-person team (surely a needless over-staffing, at this point) would type in a prompt like: “create a 15-minute podcast about knitting hosted by a middle-aged woman called Katrina Le Mongé which contains the basic principles of knitting and is intended to be soothing and inoffensive”. A few moments later, the AI spits the episode out, like a goose laying a great, golden shite. “We believe that in the near future half the people on the planet will be AI, and we are the company that’s bringing those people to life,” said Inception Point’s CEO Jeanine Wright, who might sound like a Bond villain (or the guy outside my local Sainsbury’s who yells at pigeons) but is, mysteriously a former COO of the late, lamented podcast monolith, Wondery.

The financial principle behind Inception Point AI is pretty simple: it costs less than $1 to produce each of these episodes, so even if they’re making relative peanuts via programmatic advertising, they can soon wash their face. Imagine you create 3,000 episodes a week at a cost of $0.50 per episode (than $1 figure seems high, if you ask me or my accountant). Then imagine each of these episodes makes $20 in advertising revenue (a figure that would be considered too low to bother with under a conventional podcast advertising model). That’s an outlay of $1,500 for a return of $60,000, i.e. a profit of $58,500. Per week. That’s a pretty premium return in an industry that is facing the squeeze, and it’s hard to compete with if you’re spending $1,000 on each episode (or $10,000 or $100,000). So, maybe the 50-cent episode is the way to go.

Except, of course, it’s not. Because it relies upon being able to game the mechanics. And nobody is incentivised to indulge that, other than the hot shots at Inception Point AI who constructed this genius scheme. The agency who hosts their programmatic advertising will presumably cut them off as soon as they can, as it devalues the core trust they’ve developed with advertisers. Those advertisers, who are already wary about podcasting, will see this as further evidence of podcasters’ collective failure to assure the reality of their audiences. And so it’s pretty clear that the pipeline to even a buck an episode is going to, pretty quickly, be removed. And then it doesn’t sound like such a clever idea.

It is indicative, however, of the short-termism that is, at present, guiding the way that the media interacts with AI. I spent yesterday afternoon at Press Gazette’s Future of Media Technology conference here in London, which was dominated by questions surrounding AI products. There are two main ways that AI intersects with journalism: as a traffic generator and as a creator. The first is a problem; the second, a disaster.

As a search engine, ChatGPT is increasingly competitive with Google. It still has only a tiny sliver of their audience, but it’s growing — and, more importantly, it’s changing the demands and assumptions of a searching audience. The big problem, for publishers, with an AI search is the extent to which information is being extricated from the platform, presented natively in the search engine, and thus avoiding searchers being redirected actually through to the publication (generally seen as a requirement for profitable journalism). Google has long been the biggest driver of news traffic via its algorithmically curated privileging of certain news stories at the top of the page, but increasingly the Google AI summary is occupying that real estate. This presents a real dilemma for publishers who have long relied on this referral pathway.

But, more importantly for this piece, there’s the question of generative AI. Companies like Inception Point AI (and they’re by no means the only offenders) demonstrate the ability of these new technologies to rapidly flood the market, in the way that abusive sweatshop labour changed the fashion industry. There are defences against this. A presentation from the FT Strategies team, yesterday, identified six key manoeuvres: product resilience; star personalities; niche audiences; full-service platforms; networking components; and consistent business model review. (Essentially how Veblen goods providers, like LVMH, grew in the advent of fast fashion). But the core issue is that a lot of journalism can be replaced, quite seamlessly, by AI dross. In part this is an indictment of modern journalism — where the prioritisation of ‘traffic’ created a clickbait culture and a lot of filler content — but it’s also at the heart of the demands of journalism. If I am a consumer who is looking for a podcast that will give me practical advice on knitting, what do I need beyond receipt of that information? It is an increasing reality of journalism in the internet age that it is, all too often, simply trying to predict the questions that potential consumers will ask, and then answer them. AI can do that better.

As I sat in the audience yesterday, surrounded by anonymous executives wearing lanyards bearing the names of companies I’d never heard of, I was struck by a thought. Journalism, as an industry, is irreparably bifurcated. On the one hand, you have content creators (who would, almost universally, hate that term), whether they are columnists or illustrators or podcast producers or video editors. For them, it is meaningful that there is a human involved in the journalistic process (not least because their pension plan is predicated on it). They might use AI tools (a writer might search on ChatGPT, a podcaster might use Adobe Studio Enhance, a video editor might use Veed’s auto-captioning) but they’re just tools. A shiny new chisel with which to carve the Easter Island moai.

On the other road are all the people who are invested in the media as a business or technology product. They create the architecture for this content (which is far more influential than any individual piece of content) and, most importantly, must nourish the financial conditions for journalism to grow. This means that they can’t be blind to the new technologies out there, and the way that money is flooding into the AI sector. As a panellist from a major news agency said, these tools are exciting. They should enable them to significantly increase their output. More reporting, more video, more data, more everything. And they’ll have to do that, because their differentiator (from the Joe Nobody Newswire) is scale. If they want to continue to be a big news agency, they will have to adapt to a world in which AI slop has rapidly scaled up their smallest competitors. If you’re a business development, product or audience executive at a media organisation and you’re not looking to integrate AI into your offering, you’re probably being negligent.

Part of the issue with this bifurcation is that it’s also social. Very few people in the audience appeared to be journalists. Very few journalists or podcast producers or video editors show up at AI events or show much interest in the mechanics or business of the industry their livelihoods are reliant upon. They also work in silos in these big media organisations, interacting only through unglamorous conduits like the *shudder* Managing Editor. The impact of not going to the pub with your Product Development & Operations Director is twofold: firstly, they don’t really care whether they destroy your job (because they don’t know you), and, secondly, they live in the ahistorical vacuum of people who see the media as “just another industry” in which to work.

And it’s the latter part of this, particularly, that I fear we’re getting wrong. The media is not comparable to industries like financial services or accountancy or retail. The industry it most reminds me of is professional football: fragile, vain, prone to seismic disaster but also wildly overinflated in its cultural significance. And the problem with media organisations buying in too credulously to the integration of AI products in journalism, is that it erases the fact that the media industry has already — irrevocably — fucked all of this up before.

Fundamentally, there are three ways to generate revenue from journalism. Direct sales — whether that’s flogging a newspaper in W.H. Smith’s, a subscription to your Substack or tickets to see your journalists playing the Troxy — is the traditional one. Then there’s advertising, which has evolved from print adverts, for anything from Patek Philippe watches to casual sex, to online advertising, which is a vast digital pollutant. And then there’s data — the “oil” of our new age — which can be commoditised. (On the last of these, I have basically been won over by the argument that almost all of Big Data, as a business plan, was bullshit, as suggested in this piece by

James Ball

). So, whatever, you want to run a successful media business, you either sell your product or you sell advertising against it.

Except, you may note that we are no longer living in the Golden Age of Journalism, where writers were paid $15 a word and the list of the world’s richest people was littered with titans of the media industry. We are living in an age where journalism is in terminal decline, where small publications have become the butt of the joke to the extent that The Paper, a spin-off of The US Office, has transitioned from a failing paper company (Dunder Mifflin) as the world’s shabbiest employer, to a failing local newspaper (The Toledo Truth Teller), for modern white-collar desperation. That’s despite the fact that (as was revealed to me reading

Michael Grynbaum

’s history of Condé Nast, Empire of the Elite) just a few decades ago, the giant paychecks offered to Vogue, New Yorker and Vanity Fair writers were subsidised by local newspapers just like the Toledo Truth Teller (now they’re subsidised by an early investment in Reddit, which is also funny in a drably ironic way). So what happened?

The dot-com bubble! The insane over-investment in online media! The commensurate diminution of print sales! The devastation of print advertising! The bursting of the aforementioned bubble! The ongoing and seemingly inexorable decline of online advertising! The installation of an audience belief in free journalism! The rise of social media! The loss of control of referral platforms! The insane over-investment in social media! The eroding or erasure of standards and gatekeeping! The collapse in trust with journalism! The devaluation of the product! The too-late attempt to retro-fit a direct sales model! Cuts! Layoffs! Closures!

And so, it strikes me that a lesson might be learnable from this incident (which happened pretty recently). Short-termism is fine if you’re working in a sector which involves frequent repositioning and pivoting. In financial services, for example, you’re constantly having to iterate and evolve your product, but these changes manifest as fairly minor at the consumer-level. Architecturally, a bank account today is radically different to a bank account in 2000; yet at the consumer side, the differences are pretty negligible. The purpose and perks remain the same. In the media, conversely, changes have disproportionate consumer-side impacts. Going from reading a newspaper to reading a blog on your phone is significant. Going from reading third-party curated reportage to reading AI-generated query response ‘journalism’ is significant. Going from listening to a real person talking about knitting to a disembodied facsimile droning on about an activity it could never do (legs may be coming to the Metaverse, but hands are never coming to AI) is significant.

But, more simply, you still have to make money from journalism. Selling AI slop is harder than selling human slop. It just is. People barely think they need to pay for journalism that several people have slaved over for weeks — assuming they’ll pay anything for AI-generated journalism, is for the birds.

Which means you’re relying on the advertising model. The advertising model which — if history serves as a warning — was decimated by digital advertising just a couple of decades ago. Inception Point AI might think they’ve come up with a clever scheme (I’m sure Charles Ponzi thought he was onto a winner too) by gaming (within the parameters of the rules, it should be said) the way that programmatic advertising works. But this is a heist: get out while you still can. The medium and long term effects will be to drive down the price of digital advertising. Almost every innovation has served to diminish the returns from digital advertising, but none has had the potential to destroy it in the way that AI can. Because AI slop works in two ways here: firstly, it flushes the market with low-value content that advertisers need to avoid, and, secondly, it presents advertisers with a chance to circumvent the product entirely.

If I worked for Nestlé, would I bother paying top dollar for 1,000,000 impressions on podcasts marketed at sleep deprived parents, or would I just auto-generate, for a fraction of the price, that content myself, and bake in all the Nestlé promos? It’s not just that slop will drive down prices (though it will, because many of these shows have zero listeners — they are the subject of the phantom listener phenomenon) but it will also generate more slop as a reaction. Slop begets slop.

And so, I hope that where these two roads diverge in a yellow wood, they might, perchance, merge down the line (in the green wood). Bringing more creators into the business side of journalism will help them understand the challenges facing the industry at this inflection point (perhaps that should be the name of my AI company). But it will also help the rooms full of executives plotting the inadvertent debasement of journalistic standards and practices understand that AI exists on a continuum with the forces that have already turned the global media industry from a roaring tiger into a mewing kitten.

The AI bubble will burst — as surely as the dot-com bubble did before it. But the impacts don’t evaporate. The residual effects of the pivot to online journalism are still being felt and are baked into the future of journalism. The impact of AIification of the industry will persist long after evangelism has turned to scepticism. It is far easier to destroy something than it is to rebuild it. So, journalists, take the computer nerds out for a drink — this round is on us.

Follow me on Bluesky.