Credibility on credit, creators feeding the machine, and Claude's open book moment

Guilty until proven human, the DSA data gap, and making up the web as you go - all in this week's Content Aware.

Published: 15:08, 30 July 2026

Richard rambles on... the high price of broken trust 
To media and publishing firms, trust has become as central to their business model as the words and content they create, arguably more given that trust is what makes it worth paying for in a sea of generic alternatives. What happens though when AI bots rely on your trust to craft their occasionally absurd answers? Do readers criticise the bot or you? And when they do trust you, do they click through anyway? Take a guess. Richard looks at the bots putting the rust in trust.
Read

Publishing & Media

Trust is the supply chain
AOP's AI Publisher Impact Study, which we covered last week for its grim traffic numbers, has a new instalment on trust. If a cited source is completely trusted, over 90% of users will trust the AI answer too - but a third of those users just stop there, without clicking through. On the other hand, 37% of UK adults don't even know that AI tools like to make things up. Your brand is propping up the credibility of these AI platforms, whether you like it or not. The question is what the leverage is worth, and whether the industry can act collectively before divide-and-conquer licensing deals settle the matter, one media brand at a time.
Read

Useful beats snackable
According to Nieman Lab's latest quarterly rankings of non-profit news sites, multiple outlets saw traffic double in the election season, with organic search doing most of the work. The lesson is familiar: detailed, structured local information which AI can't easily summarise still drives clicks, subscribers, and donations. In a world where zero-click is becoming the norm, being genuinely useful to your community still pays off.
Read

Visibility is the new traffic
With a little help of Similarweb data, Harry Clarkson-Bennett breaks down what publisher influence actually looks like in the age of AI. He makes the case that direct clicks from chatbots aren't really the metric that matters - it is more about being present in the research phase so users are aware to search for you afterwards. For anyone tempted by quick "GEO hacks" to join the party, proceed with caution: sites chasing AI visibility with low-quality tactics are getting nuked in search.
Read

AI cites what it finds
As an LLM can't have an opinion, when it gets asked a question it needs to borrow the answer from a human that has already shared theirs via reviews, community threads, and creator content. According to Tinuiti's citation trends data from 2026, around 82% of AI citations don't point towards brand-owned sites, but towards earned media, with YouTube appearing much more in AI Overviews. The upshot? If you don't control the opinions about your brand, you risk an AI agent picking them up regardless. Search Engine Land shares more info.
Read

The detector backlash
Substack has recently partnered with an AI detection tool called Pangram, which allows readers to scan any post for AI-generated writing. Some creators aren't happy, and those that lean on AI for editing or structure call it a witch hunt, while those that do not are afraid of false positives. Substack says it's all about transparency and even added an optional "How I make this" statement, but it's clear that while people want less AI slop, figuring out how to label it without collateral damage is hard
Read

Read

From inside Glide

From opinion, to revenue
Most publisher survey tools collect answers that end up somewhere in a spreadsheet, never to be used again. Nexa Polls does it differently: responses are being fed straight into the same first-party identity layer that handles subscriptions, entitlements, and segmentation. So when a reader answers a question, that answer will become a part of their identity in the platform - making it targetable and actionable, and giving amazing insight into what they want from you. Best of all, it's efficient: there's no costly per-response fees and no need for manual stitching between systems or databases.
Read

Big Tech

The queue for compensation
Google's first DMA fine, $1 billion for self-favouring in its app store, is setting off a sea of private lawsuits across Europe. Rivals from price comparison sites to specialised search firms are all lining up for damages, with total claims potentially totting up to $10 billion. One German firm already received €465 million, while a Swedish court ordered Google to pay roughly $2 billion, and there are surely more to come. Google refers to them as payout-seekers, while the claimants say the fines are simply the cost of doing business for Google and are unlikely to change anything. More to come, for sure.
Read

Researching in the dark
The EU's Digital Services Act was supposed to give researchers access to platform data so that they could study disinformation, algorithmic harms, and coordinated manipulation. And after two years, they still don't have it. Applications have rarely been issued, and security requirements impossible to meet, critics say. Platforms interpret the rules as they see fit, with one example being a researcher who was denied TikTok data weeks before a network of 27,000 fake accounts influenced the Romanian presidential election.
Read

AI & Copyright

Slop on demand
A Google DeepMind engineer has created Gem Browser, a mobile browser that uses Gemini to build a new visual layer on top of any webpage, or if the URL doesn't exist, generate a whole new site from scratch. Want Instagram as a punk zine? Now you can. Interested in seeing a site that doesn't even exist? Gem will make one up on the spot. You can see it as an endpoint of the dead internet: slop upon slop, generated in real time. Currently it's in beta, and plenty hope it stays there.
Read

Share button's fine print
Hundreds of Claude conversations are turning up in Google search results, after using the "share" button without realising links might be indexed. The chats include work research, CVs, transcripts, and corporate project details. While Anthropic says users maintain control and these links aren't "guessable", the share feature never makes clear that a search engine could pick them up. Now, indexing has been blocked, but no one knows how many of these chats have been saved and circulated.
Read