Context Window 24
This edition covers a correction to last week’s link on LLM task length, the new native AI text-analysis formulas in Google Sheets, the latest Adobe Firefly upgrades, eye-catching data on AI-generated music on Deezer and rising submissions to publishers, Wikipedia’s training-optimised Kaggle dataset, an Anthropic paper on how Claude expresses values, Ethan Mollick on “Jagged AGI”, Tom Goodwin’s AI-flavoured update to the Eisenhower Matrix, Trump’s executive order on AI in schools, the new CLA/PLS/ALCS Generative AI Licence, Ahrefs research on AI overviews suppressing click-through, and a contrasting Washington Post deal and Ziff Davis lawsuit involving OpenAI.
I’ve got completely hooked on the new, native AI text analysis functionality in Google Sheets, a Workspace Labs/Alpha feature (if you’re a Workspace user and don’t see it, ask your admin). It’s already saved me several hours of manual work. Instead of writing a formula, you just type =AI(“prompt for the job to be done”, [optional range]). Witchcraft. The support page has details, caveats and a range of sample use cases, many of which are highly relevant to publishing and marketing tasks.
Adobe Firefly is one of the most ubiquitous AI tools for the publishing industry, and new features announced yesterday should increase its utility for publishers, including a native app, integration of non-Adobe image models, text-to-video models, collaborative moodboarding (cover design discussions?) and Content Authenticity labelling. It’s worth checking your design team is up to date on these developments.
Streaming platform Deezer revealed that 18% of tracks uploaded are generated by AI—more than 20,000 tracks daily. That’s twice the level from four months ago. For publishing, this is a salutary warning about platform saturation: I’d love to know the equivalent figure for KDP (as a proxy data point, my friends at Storywise have publishing clients that have seen a 200%+ increase in submissions this year).
In response to increased bandwidth from AI developers scraping the platform, Wikipedia has released a free dataset through Kaggle, which is optimised for training (it’s formatted as structured JSON rather than unstructured text). It’s an interesting approach, and one that some Open Access publishers might consider, though one wonders whether bad actors that already ignore existing usage parameters will do the right thing in this case.
A new research paper from Anthropic explores how AI models like Claude express and adapt values in real-world use, with the objective of developing a model that is honest, helpful and harmless. Admirable sentiments, though the research also highlights where AI falls short. For publishers integrating AI into content workflows, a key takeaway is that these models often mirror the values of their users and shift tone depending on the task. This highlights both a risk and an opportunity: without careful guidance, AI may reinforce biases or dilute editorial standards, but with thoughtful prompting and oversight, it can support more consistent, context-aware content creation across subjects.
Ethan Mollick’s exploration of “Jagged AGI” highlights the uneven yet impressive capabilities of advanced AI models like OpenAI’s o3 and Google’s Gemini 2.5. These models can perform complex tasks—such as generating marketing plans, logos, and websites from a single prompt—demonstrating potential for automating multifaceted publishing processes. However, their inconsistent performance across different tasks requires human oversight. For publishers, this means strategically integrating AI to handle specific functions like content summarization or initial copy drafting, while ensuring humans remain central to tasks requiring nuanced judgment and creativity.
If you’re trying to put that theory into practice, Tom Goodwin has produced an updated version of the classic Eisenhower Matrix (do/decide/delegate/delete) for AI tools: in the AI update, the options are now advise/augment/automate/agentify. It’s a really helpful framework and one that I’ll be using with a client next week. Where would you classify particular publishing tasks in your business?
A new Executive Order from President Trump aims to embed AI in the K-12 curriculum. Questions remain around curriculum design, teacher training, and procurement. But it could be an opportunity for publishers with classroom-ready AI resources.
Congratulations to Publishers Licensing Services, the Authors Licensing and Collecting Society, and the Copyright Licensing Agency on the development of their new Generative AI Licence.
Research from Ahrefs found that an AI overview in search results is correlated with a 34.5% lower click through rate to webpages: challenging for anyone with a business model based on traffic or referrals.
Win one, lose one for OpenAI as publishers consider their AI strategy: the Washington Post is making its coverage available through AI outputs, and Ziff Davis launched a new copyright infringement action. The split shows that the “licence or litigate” dilemma is still live, and outcomes are far from settled.
This was originally published in my email newsletter. To receive weekly updates on how AI is affecting the publishing industry, sign up here.