A hot new contender to YouTube pits loyalties against royalties

Living in an age of media multiplicity is, by and large, a wonderful thing. Even given the tumult our own industry seems to have been locked in for the past few decades, the thing we all produce - media - is available in great quantity and in many different formats. This is truly an era of stimulus abundance.

Yet against that staggering variety, one platform dominates video: YouTube. Others exist, but, similarly, so do other search engines, and we know that only one can move industries at a whim. If you're the only one with beans to hand out, you decide how many beans to give. And it's not that many beans, if you ask the average creator.

This past week though, a new entrant signalled a serious commercial assault on YouTube's territory. BiliBili, well established in China with around 380 million monthly active users, is making a deliberate push into Western markets. MrBeast already has 8.8 million followers there, and increasing numbers of big names who built their profile on YouTube have started posting on BiliBili too.

Comments connected to identity
Most commenting tools look at readers as strangers, disconnected from everything else a publisher knows about them. Nexa’s comment system works differently, and every comment ties to a registered identity in the same first-party data layer that handles subscriptions, groups, and audience segmentation. That means that a reader who comments on football articles, holds a premium subscription, and belongs to three team groups isn’t just a random username in a thread, it’s a profile that feeds personalisation, moderation, and commercial decisions without any exports or stitching required.
Read

Journalism’s indispensability gap
Many of the things that made the 20th century news essential - listings, classifieds, and weather - now live elsewhere, and the advertising model which financed everything moved with it. With those things gone, journalism has to make the shift from capturing attention to delivering value, built around what an audience actually needs rather than what might once have been assumable by the editors. The evidence is there to back it up as well, as one news service that redesigned around defined user needs, produced 60% fewer stories but tripled its audience. Across Europe, stories that are built around what the user needs attract a third more readers and a higher share of loyal subscribers, instead of asking what happened today, journalists should ask what job does this do for the reader. Poynter knows more.
Read

Creator-led not left behind
DMG New Media, the Daily Mail’s social publishing arm with 150 million followers, changed its strategy from clipped-up news to original creator-led shows, and the revenue soared. The platform prioritises branded content partnerships over simple ad revenue models, with ten-part series commanding six-figure deals. As a lesson for others, it lays out that you don’t need megabudgets, but rather dedicated focus, repeatable formats, and people in front of the camera. News content still has a role, but is less likely to engage the younger audiences who now make up 65-70% of social followings.
Read

Visible but not recommendable
Being mentioned by AI when someone asks an informational question is one thing, being recommended is a whole different one. AI uses varying criteria when shifting from providing information to making a purchase recommendation, and while it can understand how a product works, it can also still omit it because of its limitations. A brand can have great domain authority, technically good pages, and consistent AI citations, and still disappear as soon as a buyer’s prompt surfaces evidence of higher maintenance costs or missing integrations. The fix often isn’t content or SEO, but product, support, or policy, which makes AI visibility a problem that most organisations aren’t set up to solve.
Read

The licensing gap goes south
Argentinian giant Perfil has filed the first lawsuit by a Spanish-language media outlet against OpenAI and Microsoft, alleging scraping of both free and paywalled content for AI training without permission or payment. The suit points to licensing deals with large international outlets such as Prisa Media and the Financial Times, while not compensating Latin American publishers. As a bonus, the lawsuit also includes unfair competition and abuse of a dominant market position to the copyright claims.
Read

Zero-click inbox inbound
Google, Apple, and Microsoft control 95% of email clients, and all three are building AI which can read, summarise, and triage your inbox before you see a thing. For publishers, this brings Google Zero to the last remaining channel assumed to be safe from platform intermediation, and soon will have agents scanning newsletters, briefings, and anything else to fillet. Growing the email list and mailing the list now pull in opposite directions, and the metric that matters could shift from opens anymore but AI choice. INMA’s Greg Piechota shares his thoughts.
Read

Google’s August cleanup
The Google August 2026 Spam Update has finished rolling out, taking three days (August 18 to 21). According to Google, it’s a “normal” spam update targeting sites violating existing spam policies, while also confirming it doesn’t target link spam or the site reputation abuse policy specifically. If you’ve been hit, recovery can take months, so if your traffic moved between the 18th and the 21st, it’s time to check your Google Search Console.
Read

Ad height up, organic clicks down
Between March and June 2026, the average height of ads at the top of US and UK search results nearly doubled. If you’re thinking about a page that has had ads for 25 years, that pace of change is extraordinary. The questions that need to be asked are whether Google is juicing ad space because users are migrating to AI discovery, and whether the organic click decline publishers are blaming on AI is actually being caused by this? Will Critchlow shares his thoughts.
Read

The disappearance of PDFs
Multiple sites across different industries and tech stacks are noticing reduced clicks to PDFs, due to being fully removed from indexing. It’s not limited to thin or low-quality documents, and even PDFs which reach LLM citation data are disappearing from Google’s Index. No explanations as yet, or if it is even intentional.
Read

AI lost its youth advantage
While Gen Z and millennials were the first to adapt to chatbots, their share of usage is shrinking as older groups catch up. Now the split is around 50/50 between under-35s and over-35s, while two years ago younger users dominated. Whether the change is AI usage drop-off, or simply growth across the wider population, it’s clear that AI chatbots aren’t just a “young people” thing anymore.
Read

The verified click era
According to a new report by Omnivery, bot clicks in consumer inboxes have jumped from 2% in 2023 to 16% in 2026, with Gmail going from 2% to over 20% as Google’s link scanning gets more aggressive. It’s even higher in Microsoft Outlook. For publishers selling newsletter ads against click metrics, this is an issue, as an advertiser can see thousands of apparent clicks dropping down to tens once the bots get filtered out. Beehiiv has introduced a “verified click” metric which only counts confirmed human interactions, and advertisers are starting to demand the same from others.
Read

Meta settles, the feed changes
As you will have seen, Meta has settled a 29-state lawsuit over teen addiction, and now has to hand out up to $18 billion, paid over 10 years, and change how Instagram and Facebook work for teenagers in the US, including daily usage limits, no night-time use, no notifications during school hours, and bans on some filters. For anyone keeping score, that is nearly $20 billion in child safety settlements this year alone when you add the New Mexico verdict, and the thousands of individual family suits that haven't even started yet.
Read

TikTok’s privacy leash snaps
TikTok and the US Department of Justice have come to an agreement regarding the case against TikTok over children’s privacy violations. The DOJ will drop the case in exchange for a $400 million payment. The original allegations were far from a joke, covering the collection of data on millions of under-13s without parental consent and ignored deletion requests. Also, a 2019 FTC consent decree governing how TikTok handles children’s data has been vacated, meaning the company now operates without the privacy order that previously kept it in check.
Read

Prompts and serial numbers
Microsoft is adding a unique identifier in every AI-generated image made with Paint or Photos, linked to the prompt that created it and sent to Microsoft's servers for moderation. A developer that reverse-engineered the approach found out that the successive prompts chain together, meaning Microsoft can link an entire session of AI image creation back to a user. It gives more than the EU AI Act asks for, and Meta and OpenAI are building similar tools.
Read

YouTube counts everything now
YouTube recently changed how they count views and starting this week, a view registers the moment playback begins rather than after a retention threshold. It brings the long-form video content in line with YouTube Shorts, which has been on this standard since March 2025. On another note, according to BrightEdge data, Google AI Overviews cite specific timestamps within videos half the time, while ChatGPT cites the whole video. If you're a publisher producing video, it's good to know that chapters and uploaded transcripts now determine whether your content gets cited, because neither of the engines can actually watch your video... yet.
Read

A million flags, and counting
LinkedIn’s AI Slop button has been used over a million times since its launch on July 30, and according to the Chief Product Officer views of content deemed AI-made are down 40%. It is also rolling out notifications to authors whose content has been flagged, and removed a feature that offered to enhance posts with AI. However, the 40% figure measures reduced distribution of posts that are already flagged as AI, not whether there is actually less AI content published to the site. A separate study by Pangram found that 41% of LinkedIn long-form posts are supposedly fully AI-generated.
Read

Honeytrap, LLM-style?
One to get your security teams twitchy. An anonymous coding model appeared on an open developer marketplace this week, offering a million-token per day context window for free - enough for very large data drops - and not much insight into who runs it. Its EULA granted full data read and training rights to the unnamed provider, and unsurprisingly it set off a chain of alarm amongst researchers trying to trace its origins. A trail led back to Chinese AI firm Zhipu AI’s infrastructure, but could not determine who actually created it. Some say it is perhaps a new Microsoft model being tested, or DeepSeek. All we know is, this is exactly what someone would do, if trying to test or a massive new frontier model - or harvest huge amounts of data.
Read

Charts for humans only
Australia has banned AI-generated songs from its music charts, requiring that all releases be “substantially human made” to qualify. The songs need to be written by people, and perform the lead vocal and instruments too. This came after an AI-credited cover of Madonna’s Like a Prayer topped the charts and racked up 48 million Spotify streams. Artists will have to declare AI use on submissions, and if the song turns out to be largely AI-produced, it will be stripped of its chart position and the artist might even be asked to return the awards it got. Sweden did the same thing earlier this year, and the IFPI, the International Federation of the Phonographic Industry, is rolling out similar guidelines across Latin America, the Middle East, Africa, and Southeast Asia.
Read