The (old) fact boxes, the AI theft pipeline, and fragile facts

Europe's all-you-can-eat giant grows, the model too dangerous to release, your impressions were lying, and the AI content cliff - all in this week's Content Aware.

Published: 15:27, 09 April 2026

Corbidge comments on… less is more in media again
The Times cut its daily output by 25% and recorded three consecutive months of record audience growth. In today's world, where AI has made quantity more or less infinite, competing on volume is a losing game. When science literature can be polluted by hallucinated citations, critics say the science isn't good enough. Does the same go for journalism? If it can be polluted easily, maybe it wasn't good enough to begin with. Our man Rob, stout and true, discusses why quality is still the thing people come back for.
Read

Publishing & Media

Cash in the attic
Most newsrooms treat their archives like a loft - full of things that might be useful one day, and rarely visited. However, some publishers are doing something a little bit different: turning old content into fresh money. Back catalogues are feeding into subscriber products, via both serious and fun features like interactive timelines, reader quizzes; connecting historical coverage to current events, and through editorial partnerships. Digitising decades of journalism to make it searchable for both journalists and readers is paying off with engagement. Thinking about what your archives could be doing? Take a few notes from these examples.
Read

Cafeyn hit Readly road
All-you-can-eat reading platforms Cafeyn and Readly have merged, with Cafeyn becoming the consumer and Readly the consumed. French Cafeyn now owns Readly's operations in the UK, Germany, Switzerland, and Australia, adding 350,000 users to now reach a total of 2.5 million subscribers who can access 5,200 publications. The Readly branding stays and no redundancies are due. The pitch to publishers is pretty straightforward: more scale equals more revenue distributed, while AI features and video are coming, including a daily digest newsfeed, but with a gradual rollout. For publishers on the platform, the message is that nothing really changes, except the cheques are getting bigger. The group is also eyeing large-scale mobile phone provider partnerships, and white-label app deals with publishers.
Read

Liar, liar, impressions on fire
Google has confirmed a logging bug which has been over-reporting page impressions in Search Console for nearly a year. A fix is rolling out over the next few weeks to cure the ill which dates back to May 13th 2025, so if impression counts fall bear it in mind. Clicks and other metrics were not affected.
Read

The Overview overview:  
The New York Times has done a detailed investigation into the accuracy of Google's AI Overviews, and the findings are worth reading in full. They found roughly 10% of AI Overviews contains false information or shaky sources, and while better than before, at the scale of someone like Google that's still tens of millions of fact manglings every hour. Basic acccuracy has improved - up from 85% with Gemini 2, to 91% with Gemini 3 - but in exchange, sourcing problems have worsened. More than half of correct answers now link to websites which don't actually support the information provided, and then we wonder aloud why people don't trust "the media". NYT brought receipts, and the piece is full of specific examples such as incorrect dates, made up assertions, and the journalist who faked his own hot dog notoriety.
Read

Your content in the pipeline, you are not
We all know the refrain that AIs are stealing your content. But how? Matthew Scott Goldstein bring a useful breakdown of the four-tier ecosystem systematically sucking up publisher content to feed AI systems. At the top are the bulk harvesters who run continuous crawls of publisher sites and flog on your wares, then there are the answer engines such as Perplexity which pull live content from publishers in real time and steal your clicks, then come search and extraction hybrids which still send some referral traffic, and at the bottom of the AI food chain are journey initiators which make all the others possible. What's missing of course is a licensing layer.
Read

Raise prices or hold the line?
As inflation inflates, ad markets are softening and production costs are rising while end users have less to spare. It's no surprise the question of whether to raise subscription prices is never far away. When looking back to 2022, price-change-related churn didn't actually increase as subscribers accepted the increases, partly because news doesn't have real substitutes like streaming does. Now it's a different story. Traffic is down, AI is disrupting discovery, and for publishers with weakening acquisition funnels, aggressive pricing is unlikely to compensate for fewer new subscribers coming in. INMA sheds light on the data and case studies to help publishers weigh options.
Read

One tab to rule them all
Blatant plug time - but this is a banger. Every publishing team knows the problem of too many systems and too many tabs. You're in the CMS pulling the strings of the orchestra, while simultaneously in the content calendar to match output with planning, in the analytics dashboard in another, or an SEO toolset in a fourth. It's work overload, and it costs more than it should and errors creep in. Glide External Apps nicely fixes it by allowing you to bring third-party tool interfaces directly into the CMS interface, so editors can check performance data, review schedules, browse assets, all without leaving the CMS. Any web-based tool that supports iframe embedding can live inside Glide, with role-based access controlling who sees what - which is great for security. Your custom internal tools can authenticate automatically using existing Glide credentials, so no need for a separate login. The time savings per action rapidly stack and across a large team of editors they compound into less time wasted and better insight where you need it.
Read

Big Tech

Your browser, LinkedIn's data
Every time you open LinkedIn, in the background a script triggers and silently scans your browser for the presence of more than 6,000 specific Chrome extensions, creating a detailed hardware fingerprint, encrypting it, and sending it to LinkedIn servers to run their beady eye over you. An investigation by a European association of LinkedIn users and independently verified by Bleeping Computer, has been tagged "BrowserGate". While LinkedIn disputes the characterisation, it doesn't seem to be arguing the technical facts. Among the extensions there are more than 200 direct competitors to LinkedIn's own sales tools, as well as extensions connected to job hunting, neurodivergent conditions, and political interests - all things that fall under sensitive personal data to most people. The scanning started in 2017, for 38 extensions, and by February 2026, the number jumped to 6,167. No opt-out, no disclosure.
Read

AIs "under lock and key"
Anthropic's new model Mythos is reported to be finding thousands of security vulnerabilities across every major operating system and browser, write its own exploits, and chain Linux kernel flaws together like a hacker's dream. OpenAI too alleges its latest new one is so good it should come with a health warning. Some say it's FOMO marketing by AI again in the race to grasp attention, but nonetheless Mythos is being released only to a handpicked group of 40-plus organisations, including Google, Apple, Microsoft, Amazon, and Nvidia. This initiative is called Project Glasswing, with a goal to use the model defensively - scanning and securing code rather than attacking it.
Read

AI & Copyright

How AI picks its sources
In AI search pipelines, retrieval happens before ranking, and if your content doesn't make it into the spot decided at embedding stage, it can't be cited. Perplexity, Google, and OpenAI all approach citing of pages differently; Perplexity uses bidirectional context tuned for real-world queries, Google combines lexical and dense retrievals with passage-level evaluation, while OpenAI uses Matryoshka Representation Learning. The bottom line for content teams isn't to write for machines, but to write clearly, structure your content explicitly, and make sure the info can be verified. Wordlift shares more info.
Read

Greece tells under-15s: log off
Greece has decided it will ban social media for children under 15 from January 2027, following Australia's lead and joining a growing list of countries who are thinking about the same thing. According to the Greek prime minister, anxiety, sleep problems, and addictive platform design are the driving factors, with 80% of those surveyed in a recent poll backing the ban. The platforms that fail to restrict underage users will get slapped with a fine up to 6% of global turnover under the EU's Digital Services Act. Greece is calling for a similar ban to be EU-wide, arguing that these measures won't be enough without a unified enforcement framework across the EU bloc.
Read

When seeing is no longer believing 
Argentina's leading fact-checking organisation Chequeado has published a breakdown of how AI is reshaping the information landscape, both in Latin America and beyond.  According to them, there are three distinct disruptions: AI being used deliberately to disinform, generative chatbots being used as news sources despite their unpredictable error rates, as well as the erosion of what counts as evidence at all. The last one is the one that takes the cake, because when synthetic content becomes convincing enough, people don't care if they trust the source or not. This piece also shines a light on what technology can and can't do, such as the fact that no reliable automated fact-checking model has been created, and that human judgment is still the thing technology can't replicate.
Read

The surge, the drop, and the lesson
When looking at the Grokipedia situation, you are looking at what happens when a site scales heavily with AI-generated content. It surged in Google rankings and then dropped sharply as Google's systems adjusted, and the recent March 2026 Core update seemed to accelerate the decline. The same thing happened with other AI Search tools: AI Overviews, AI Mode, and ChatGPT, they all surged and then dropped. For anyone who still thinks that pure AI content volume is a good growth strategy, take this as a cautionary tale.
Read

Writers settle, actors don't
The Writers Guild of America has agreed to a tentative four-year agreement with the studios and streamers, which covers minimum pay rates, residuals, health fund contributions, and the most important thing, guardrails around the use of Gen AI to train models on members' scripts. On the other hand SAG-AFTRA, the performers' union, is a different thing; they failed to reach a deal with the studios in March, and heads back to the table this spring.
Read