The content vault is closing, Fair Use makes a comeback, and delicious dishes too good to be true

Four pillars of publisher SEO, spam update number three, the brands AI remembers, and Total Eclipse of an art - all in this week's Content Aware.

Published: 15:15, 20 August 2026

Corbidge comments on... objections to AI are subjective
Book publishing is having a tough time with AI. Authors increasingly need to explain why their work tests positive to AI detection tools, launches have been abandoned, deals torn up, and AI-generated quotes have been attributed to real people. However, as a researcher points out, some suspected AI works are proving to be smash hits. Our mighty penman Rob brings a cold sword to the debate and asks: if you like a book - or anything else - does it really matter if it is a bit AI?
Read

Publishing & Media

Scarcity as a strategy
According to Cloudflare, default blocking of AI crawlers is working and creating "reliable scarcity" for content, allowing publishers to negotiate better licensing deals. Cloudflare is also evolving its model from pay-per-crawl to pay-per-use, where publishers get compensation when content turns up in AI answers. From September, mixed-purpose crawlers which blend search and training will get the axe by default, unless the companies behind them separate their bots. If smaller publishers are asking whether blocking is worth the hassle, the answer from those who already do is that deals only seemed to start coming after free access was ended.
Read

No clean line, no clean answer
The US federal courts are divided when it comes to whether training AI on copyrighted works is Fair Use or not. In Delaware, a judge ruled against an AI legal research tool which trained on Westlaw headnotes, ruling it a market substitute more than transformative use, while in California, two judges reached the opposing conclusion in cases against Anthropic and Meta, reasoning that deriving statistical patterns from books was Fair Use. All eyes are now on The Third Circuit - the first appeals court to weigh in - and an appeal docket listing more than 90 AI copyright cases.
Read

Food with extra hallucination
Google recently tested AI-generated images inside AI Overviews, replacing photography with generations. Recipe publishers were amongst the first to spot it after Google decided to serve up interpretations of dishes versus what they actually were. The backlash was swift and loud enough to make Google pull the experiment. The intent is clear though, and if your content relies on visual search traffic, prepare for its return.
Read

2026 publisher SEO audit guide
Out of a trillion clicks that Google sends to the open web, up to 86% are discovery-led. While discoverability is shifting toward AI answers and feeds, a publisher's website is still the home base. According to Harry Clarkson-Bennett, auditing publisher SEO in 2026 rests on four pillars: technical SEO, content and on-page, E-E-A-T, and entity consistency, with AI readiness built into each. After all, your site still needs to be crawled, rendered, indexed, and cited.
Read

Domain authority = model memory 
According to a Fractl study, brands with strong SEO footprints are barely surfacing in AI Answers, while others with a smaller search presence show up repeatedly. Every category has a short list of "default" brands which the AI models recall, regardless of how you phrase a prompt, and they're not always the biggest or best known. In this world, ranking well in Google may not be enough to get you into the consideration set.
Read

From inside Glide

The content gap nobody sees
At the top of the Premier League, content departments look like mid-sized publishers, where editorial teams produce the same output across the same platforms on the same day. In this guest post, First Team's Mark Gilbert shares his thoughts on why leaner football clubs can't expect to win on volume, but where optimisation trumps sheer output and what the smart clubs are doing differently ahead of the upcoming Premier League season. (Have you done your FPL team yet?)
Read

Hello!'s un-Discover miracle 
When Google Discover wiped out nearly a quarter of Hello!'s UK website audience in 2025, the brand didn't chase the algorithm back, and instead leaned into syndication across Apple News, MSN, Yahoo, and Newsbreak, and making content was optimised for each platform from a single Glide CMS workspace. It saw page views growing between 33% and 102% depending on the channel. Revenue from syndication not only plugged the gap, it also freed resources to launch Hello! VIP, a paid subscription tier sitting on top of a million free registrations. Now total reach is higher than before, and the audience is owned rather than wholly shared. Listen to the full podcast and read more about it here.
Read

Big Tech

Addiction by design, on trial
Twenty-nine US states have begun a landmark trial against Meta, claiming the company is hiding internal research that shows Instagram was addictive for teenagers while at the same time designing features that fuel addiction. One of the internal documents cited in the opening statements said that Meta itself wrote "Instagram is addictive", while another stated "the young ones are the best ones". If found liable, the damages could pass $200 billion, or the company's entire 2025 revenue.
Read

Google's August cleanup 
Google has started rolling out its August 2026 spam update, the third this year, and it applies globally across all languages. There have been no new spam policy types announced, although the June update was followed by a clarification that spam rules also cover attempts to manipulate AI Overviews and AI Mode. The previous two were rolled out fast, with one in March done in under 20 hours. In case you notice weird traffic moves, check your Search Console from August 18th onwards.
Read

Advertising your replacement
An iPullRank study of 13.1 billion search events found that ChatGPT holds the sixth place when it comes to most-clicked destinations from Google Search, and it also draws a higher share of paid clicks than any other in the dataset. Meanwhile, Google's rising zero-click rate is damaging organic traffic but also leaving paid clicks untouched, which means that advertisers are still getting their share even as publishers are losing referrals. Only 11% of ChatGPT-bound searches get intercepted by a zero-click Google experience - which is one of the lowest rates measured - meaning that if someone just wants to open ChatGPT with an AI Overview, you can't stop them. The irony is clear as Google's own ad system is helping users find their way to a rival.
Read

AI & Copyright

Twitch joins the dark patterns
Twitch users are kicking off after the platform started feeding streams, VODs, clips, and even chat messages into AI training, with an opt-out toggle buried so deep most won't find it. When its CPO was asked why it wasn't opt-in, he was refreshingly honest: if we asked users, nobody would agree to it. Even if you do manage to dig your way to the opt-out button, if you chat on someone else's un-opted stream your messages will still get used. It's a recurring AI platform pattern: change the defaults and hope nobody notices.
Read

Another AI legal sing-song
Music publisher Round Hill on Monday filed a lawsuit against Suno and Anthropic (PDF) for mass infringement, claiming hundreds of Round Hill-owned songs were scraped to train AI models without permission. Round Hill says it plans to amend the claim to cover potentially ten thousand works and damages, including the iconic Total Eclipse of the Heart, coincidentally propelled back into the charts recently after singer Bonnie Tyler passed away and an actual eclipse too. Claims exceed $1 billion, while both Suno and Anthropic are facing similar suits from other sides.
Read