Context Window 76
Greetings from Portland, where I’m writing this on the Thursday of IBPA Publishing University—or, as my body clock is telling me, about 04:30 London time on Friday… It’s been a long day, but lots of good conversations with publishers have helped to sharpen my thoughts on this week’s news.
The long-awaited Fairness Hearing on the Bartz v. Anthropic settlement was held this afternoon, and a ruling is expected imminently. The Authors Alliance has a good instant analysis of things, particularly the various objections that were raised around particular types of book and author being excluded from the proposed settlement. This may matter as much as a learning point for future litigation as for the case in hand.
The Authors Guild published an update to its guidelines on AI usage: the result is a mixture of good sense, practical advice for authors and campaigning rhetoric—it’s unsurprising to see LLM training described as theft, for example, though I wonder whether authors could be better served by explaining the Fair Use argument even if one might disagree with it.
Last year research by BISG in the US and Canada and the IPG in the UK suggested that fewer than a third of publishers had an equivalent AI policy: I would have hoped that had changed by now, but judging from some of the conversations I’ve had in Portland, I think there is still work to be done on that.
Amazon brought together two of its AI tools, Rufus and Alexa+, into a single new product, Alexa for Shopping. There are few AI tools that could so directly shape the purchase experience for books: the announcement highlights that Rufus was used by over 300 million users last year, against a user base of perhaps 250 million Prime customers.
The strategic question for publishers is shifting from How do readers discover books? to How do AI systems recommend them? Those are not necessarily the same question. New research in HBR suggests that brands may be less prepared than they think: of eight common levers for increasing ecommerce sales, only one was found to be effective when dealing with AI assistants.
Google updated its list of AI case studies, which now stands at more than 1,300 examples from a range of industries. The media and marketing section doesn’t contain anything publishing-specific, but if you have time for a scroll, the list provides some useful inspiration. If you don’t have time, NotebookLM or another AI tool could probably help narrow it down for you.
In economics, Goodhart’s Law states that once a measure becomes a target, it ceases to be a good measure. Case in point: Amazon used AI token consumption as a measure of its targets for staff AI use, so engineers started automating unnecessary tasks to boost their personal ratings. There’s a lesson here for anyone thinking about how to prove ROI, and it is not to follow the Amazon example and build purely quantitative targets.
In 2011, Eli Pariser coined the phrase “filter bubble” to describe how algorithmic media presents us with a restricted view of the world. In a new presentation, Pariser and colleagues explore what might happen to media when we move from a media landscape curated by algorithms to one curated by AI agents.
This is hands down one of the best media strategy decks I have seen in years. The key insight: in a world of algorithms, many consumers are swapping reach for trust. If your marketing team isn’t already thinking through the implications of this for your messaging, you should read the presentation today and send it to them.
It’s an article of faith among many publishers—some of whom I’ve spoken to at this convention—that consumers don’t want to read AI content. Evan Armstrong takes a contrary view on that with an analysis showing that in some content categories, nearly a third of Substacks are AI-generated—and some of those are making seven-figure revenue annually. If true, the uncomfortable implication is that consumer tolerance for AI-generated content may vary far more by category and utility than many creative industries believe. That sits uneasily alongside assumptions about authenticity: readers may care less about who or what created content than whether it reliably solves a problem.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.