Legacy still leads, a transatlantic dressing down, and Google starts paying (quietly)

What counts as news now, AI answers hit France, OpenAI goes to Canada, and backwards logic on the AI frontier - all in this week's Content Aware.

by Emina Zulic

Published: 15:26, 17 September 2026

Corbidge comments on... Google's latest  scheme
Perhaps with judges and regulators in mind, Google is tempting small and mid-sized publishers into a scheme to pay for the use of their work in AI. Its "AI contribution pilot" will release payments when content makes a "significant" contribution to AI-generated responses, but under who's measurement? Well, Google's of course. Our resident Crumb Counter Rob sees two possibilities: the move heads off legislation by letting Google claim it already pays publishers, or it builds a new system to insulate against future content scarcity as other firms also pay and free stuff no longer has any value to AI's biggest brainiacs.
Read

Publishing & Media

Trust still favours the pros
UK media regulator and competition authority Ofcom released its latest national news consumption report - the largest in the country and the one government and marketing budget holders would look at when considering things like platform, AI, and social media usage. The report shows for the first time that online has overtaken TV as the primary way news is accessed, 82% versus 79%, however that online access includes vast volumes of stuff made by the TV broadcasters for online channels so perhaps is more an advertising concern than a "TV news is dead" story. The rise of news intermediaries, aggregators, and AI news is stark given how little trust there is in them in the population at large. 73% get news from middlemen, and 21% have used AI for news, however for trust both rate lower than dedicated news sites and providers. The headline of 21% "using AI for news" is somewhat broad, as that includes summaries and AI-assisted search, rather than getting news content generated in a chatbot which sits far lower, and is trusted even less in the UK.
Read

So what is "news" anyway?
Meanwhile a Reuters Institute study looked at how people now think of "news" when it's lumped in the same feed with entertainment, influencers, and everything else a feed might have bracketed under what a platform says is news for you. The definitions are like an upside down tree: massive agreement at the top that core topics like politics, economics, and conflict reporting by established outlets and journalists are "news", and beneath that a growing subset varying by age and platform who regard softer topics made by creators and on social media count just as much as "news". Not really a shock - one person's meh has always been another person's must-read - but the report starts to rate and name the topics and channels. Of note, agreeing that something is "news" is no guarantee people will look at it: serious topics can bore by ubiquity and emotional drain, where lighter lower-stakes creator-led formats seem to avoid the tedium of reality more effectively while still keeping their audiences in touch, even if recognised as being less deep or trustworthy.
Read

Google to pay for AI content! 
Time to dust down your copy of "Hoop jumping for amateur acrobats" and watch the bullet-dodging scene in The Matrix in reverse: Google is running a small pilot to pay publishers who "contribute significantly" to AI-generate responses in AI Overviews, AI Mode, and Gemini. The AI Contribution Pilot will show participating publishers an earnings widget in the Search Console, which will show what meagre or massive value Google places on content it judges it to be of a significant contribution, or what significant actually means in Googlespeak. Dozens of smaller and mid-sized publishers have been approached, and cautious reactions are in the vein of "better to be inside the tent". Is it a real shift in philosophy, or a sop to regulatory scrutiny and copyright cases and forcing cash-strapped OpenAI to burn some more funds in response? The measure we all need to know is what secret rules Google uses to judge "significant", and what you gave up to be included.
Read

Condé Nast's five habit loops
Condé Nast's Global Revenue boss reveals 5 habit loops proving effective at transforming occasional visitors into loyal audiences, tested and replicated across a number of its titles. Holly Johnson talks of the daily ritual loop that puts you into someone's everyday routine, the progression loop encouraging return visits through things like streaks and milestones, a personalisation loop that grows content relevance over time, a curation loop that cuts builds on that, and finally a community loop which really makes people feel part of the offering. The underlying theme of course is turning generic journeys into very personal experiences.
Read

France gets AI Overviewed 
Google has launched its AI Overviews and AI Mode in France a month ago, and Similarweb data now shows clear patterns in its occurrence to readers. All 10 leading news and media sites saw referrals decline, with losses between 11.3% and 29.6%. Ecommerce and real estate fared better, with most sites actually growing. While there is no definitive proof AI Overviews caused the decline, versus says an algo shift, the consistent scale of drop for news sites specifically is a hard to ignore, and the pattern matches what other territories have seen when AI "answers". SEO strategist Aleyda Solís investigates.
Read

International SEO seeds
While the standard advice in international SEO is still "don't translate keywords, research them natively", it's important to remember the very first seed phrase can be linguistically accurate while completely missing the vital shorthand term a market actually uses. A bad starting phrase doesn't merely cause weaker results, it can fully undercut your position in the new territory. Examples show how using translated natural language can leave out important key terms which search engines locally have learned is key, such as a local regulator term or acronym. A good obvious example would be talking about "retirement planning" to a US audience, but not using "401(k)".
Read

Train now, pay maybe
Leaked Australian government documents reveal a proposal to allow AI companies to train on online content without paying creators directly, as long as they manage to strike deals with enough rightsholders. Options that are under consideration would make inclusion in AI training opt-out by default, shifting the burden to Australians who will have to actively protect content they created. While one route funnels payment through a central body, the other simply requires companies to hit a quota of deals, with "no additional $ payable to creators beyond deals negotiated to achieve quota". Critics already decry this protecting AI companies instead of Aussies. 
Read

Google's effort score
According to a leak, Google has a metric called "contentEffort", which is an LLM-generated score which estimates how much work went into a page, sitting right next to "siteAuthority" and "lowQuality". It's not something new, and Google's own Quality Rater guidelines mention effort multiple times and scores pages on it alongside originality, skill, and accuracy. The catch is that Google can't see how long you spend on your content, and the only evidence they have is what ends up on the page: original data, firsthand experience, visible methodology, real screenshots, genuine opinion, and no filler.
Read

From inside Glide

Stop guessing, start asking
Most publishers personalise content based on what they infer from clicks, dwell time, and device patterns, which essentially comes down to guesswork that's getting harder to make as cookies disappear and privacy regulations tighten. There is a simpler model, and that is to simply ask readers what they actually like. Topic follows, saved articles, newsletter selections, and poll responses are all declared preferences that don't depend on tracking, don't decay, and can't be erased with a browser update. The commercial value is higher too, because a segment built on what people actually told you is more useful to advertisers than one built on probabilistic inference. The key is to ask at the right moment, and always give something back, such as a better feed, relevant newsletter, or a reading list worth returning to.
Read

Big Tech

OpenAI lands in Canadian news
Village Media, one of Canada's largest digital-only local news network, struck a partnership with OpenAI to launch Open Door, an AI-powered tool that connects residents to community support like food relief and safe housing. It draws on a curated directory of more than 200 local organisations and launches first in Sault Ste.Marie, Ontario. Village Media is the owner of the product, while OpenAI will provide funding, API credits, and technical support, and in return, they will get Village Media's content into ChatGPT. According to the publisher, this is "fair". Is it, though?
Read

The 48-hour coup
The Automattic/WordPress legal saga resolved itself quickly after an OpenAi-esque turnaround which saw the ousted boss return and board members depart. After Matt Mullenweg was put on paid leave, he's back in the seat with "full support of the board" - those who remain that is. Meanwhile, the backdrop WP Engine vs WordPress lawsuit rumbles on.
Read

Amazon's play for Gen Z
Amazon has announced that they're adding short-form local and national news clips to Prime Video, in a clear play for the Gen Z audience to jump ship from TikTok, Instagram, and YouTube. The clips will surface on living room devices first, with web and mobile to follow. Amazon doesn't have a social platform of its own, so Prime Video is getting the short-form video, but other streamers Peacock, HBO Max, Netflix, and Disney+ have all added short clips this year. One thing to keep in mind though, young viewers like short news because its feels organic, and not just short, which is harder to automate. 
Read

Europe's child safety push
In a union-wide address, EU Commission President Ursula von der Leyen proposed banning under-13s from social media, online games, and chatbots, with supervised "mini accounts" for 13- to 15-year-olds. The proposal now goes to the EU's 27 member states for a vote, and could reignite tensions with US platforms and leadership critical of Europe's tech rules. The speech comes on the heels of a wider global reckoning on child safety online, from Australia's under-16 ban to Meta's recent settlement in the US.
Read

AI & Copyright

Distillers distilled 
While previously US firms such as OpenAI and Anthropic have accused Chinese competitors of illicitly "distilling" their models, now the tables have turned and we're seeing a Chinese robotics startup, JoyIn, accusing OpenAI of distilling their Aether model and copying its cosmic branding for the GPT-6 Astra launch. This counter-accusal lands the same week as US authorities publicly named DeepSeek, Alibaba, and others for "industrial-scale" distillation campaigns. The irony writes itself, probably using your words 
Read

Project Lily spying unmasked
A new reports has revealed how OpenAI's human review of ChatGPT conversations actually works, via a programmed called Project Lily. Human prompt reviewers read anonymised real-world chats and grade the quality of responses, check for AI-speak, sycophancy, and false claims of personal experience. OpenAI admits the anonymisation can let personal data slip through, and also that the version the reviewers read can include a summary of the user's interests, context, and even location. This is rather uncomfortable, as many people treat ChatGPT as a therapist or confidant, while the "use my chats to improve the product" toggle is on by default in most consumer plans and doesn't work retroactively. 
Read

Pacing the frontier or hitting the brakes?
After a week of turbulence in the AI world over safety, which includes a safety researcher's resignation from Anthropic, CEO Dario Amodei has published a 3,800-word letter, backed by the heads of OpenAI, xAI, Google DeepMind, and Microsoft, asking for a slow in progress to boost safety research and let it catch up. But is that their motivation? Guardian's Stuart Russell argues that you don't set a slower speed and hope safety just catches up. Instead, you set the safety requirements first and only progress when they're met. 
Read