Slop until we drop? Not for discerning minds
There's more AI-made content than there is human-made. Is this the death of the web, or a tipping point where the AI-internet eats itself and lets a new better web emerge?
Years back when I worked in a commercial analogue data outlet - a bookshop - it became apparent to me that fully three-quarters of the books on sale could simply be other books. That is not to denigrate the efforts of the authors and editors involved, but simply to state that 75% of the stock was good enough to be stocked, and not much more.
A simple test of quality in this instance would be how long a book stayed on the shelves and in production. If a book, outside of specialist educational textbooks etc, could maintain a shelf slot for five years, it was quite something. If it could maintain 10, then a certain test of cultural curation had been passed. To take this process of cultural curation to the extreme, you will still likely find a copy of Plato’s The Republic sitting somewhere in any large bookshop worthy of the name. It was written around 375 BC. Other classics are still older, and still read, and still stocked.
In our haphazard and very human way, we’re pretty good at this process of cultural curation. Of course some wonderful things get lost, and other wonderful things sometimes pass unnoticed in their own era, to be later “discovered”, yet by and large there is a property of discernment at play and it has yielded decent results, as we’re still here and still reading.
It was a post by the wonderful Barry Adams that led to my line of thinking. He was referencing a prediction that GenAI would alter the content quality bell curve, a prediction made only eight short months ago. Simply put, the sheer volume of content that GenAI systems enable to be produced is swamping human-produced content, meaning the distribution of low quality content is now of such great volume that it has changed the shape of the information landscape.
All available data appears to confirm this prediction, and indeed, the crossover to more than 50 per cent AI content versus human may have occurred as far back as November 2024.
It’s possible to view this as a form of the process of malign industrialisation. We can manufacture intricate things in the millions that only a few hundred years ago would have taken significant time and skilled labour, if they could be done at all. This is what is being offered to us in AI content. It takes less time and effort to produce a thousand words on something than it ever did, and goodness knows producing coherent content is labour intensive work for humans, so the enticement to use Generative AI is clearly there.
The fundamental issue is that, in the context of the written word, GenAI feeds but it does not contribute. That is to say that anything produced by GenAI is simply the sum of its parts, to which it adds nothing but a transformative process of aggregation and weighting. Many would say that’s quite a big nothing, but I would disagree.
Creativity is the foundation of contribution. By way of example, consider one William Tyndale, the devout 16th century Christian, who translated the Bible into English directly from the Greek and Hebrew texts, an incredibly dangerous and controversial thing to do. His efforts, combined with the new availability of the printing press, certainly changed the course of English history, and Tyndale ended up strangled to death in Flanders for his troubles.
Whatever your religious persuasion, or the absence of, if you use spoken English you will use some of the words and phrases Tyndale invented. Such invention was necessary for Tyndale, as he had to find relatable concepts in English to match those of the ancient texts he was working from. Hence he gave us, among others “scapegoat” and “Passover”. He even came up with “Jehovah”. “The salt of the earth” is his, as is “The spirit is willing, but the flesh is weak”, and “Am I my brother’s keeper?”.
There are hundreds of such examples, and it’s safe to say that it was a combination of Tyndale and Shakespeare that shaped much what we would consider modern English.
The important thing to consider here is that Tyndale made words up. Words that are now so familiar and ingrained as they pass without notice or knowledge of origin. By comparison, GenAI offers only creative atrophy by way of contribution, as it can only take from what is, and not make what might be.
This might be a long view, yet, for publishers, it is this quality of human discernment that will be our saviour. As we all are aware, GenAI content may be all-present, but it’s not very good. Nor, despite promises made by the technologically zealous, does it show much sign of improving.
The creativity of GenAI is in the engineers who build the systems, not so much in the end result.
Glide CMS is the headless CMS designed for media, sports, and publishing.
Featuring an intuitive interface, Glide CMS’s comprehensive content toolset, including live reporting, digital asset management, automated SEO, and AI-powered text, image, and audio tools, gives editorial teams the tools they need to improve content quality and speed up production.
Achieve your audience engagement and growth goals faster with Glide CMS.
Get Glide and get going.
Growth beyond Google
No more crying over lost clicks, now it’s all about building audiences that stick around. It’s earning report season, and analysing the statements and assertions of some of the media’s biggest brands reaffirms how much of the industry has pivoted away from faceless traffic towards direct audience connections. While some outlets such as People Inc. and BuzzFeed felt the Google search traffic pinch, others are seeing their digital ad revenue climb: New York Times reports +20%, Ziff Davis +5.9%, USA Today +2.9%. The trick? Direct connections, more video, and - for some - licensing content to AI platforms. Publishers can still survive (and thrive).
Read
Privacy pause?
Brussels seems willing to soften its stance on data protection if it helps AI to progress. The bloc’s upcoming “digital omnibus” will allow AI firms to play around with sensitive data, redefine what actually counts as personal, and (yay) simplify cookie banners. Privacy experts are perturbed, especially around who owns said AI, while pro-business groups think it might be a good thing. More details due next week.
Read
Japanese finance site Glides ahead
A chance to toot our own horn, after Japanese financial information and data brand DealWatch relaunched on Glide Publishing Platform. The service powers terminals and screens for banking and finance industry clients in Japanese, and leans on Glide CMS and Glide’s audience Engagement platform Nexa to manage a complex mix of both traditional content and realtime data and information.
Read
Munich’s historic nein
Regional Court judges in Germany ruled that ChatGPT broke copyright when using lyrics by local artists. Owners OpenAI will appeal the ruling, made after action brought by German IP rights organisation GEMA. The ruling does not have the same authority as a national or EU ruling, but it is giving legal experts a useful steer on sentiment for other cases.
Read
Discover fix on the way
Google Discover is working on ejecting AI spam, it says. Large numbers in the UK read fake articles in the prime Discover spots, despite being from fictional content from recycled and dubious domains. Google says a fix is due, but it is still handing out top spots like candy. Press Gazette knows more. Trusted publisher domains would be an obvious answer.
Read
Archives to algorithm
Time Magazine is jumping into AI, teaming up with Scale AI to breathe new life into their 102-year-old archive with an AI chatbot readers can interact with. Readers can now chat, question, and get short summaries from decades of journalism, in 13 languages. Does this single-source type of system work commercially? Give it Time.
Read
AI citations go rogue
AI answer engines use a process called “query fanout” to bring replies back to users, where a single original question might be broken down into parts, and rephrased with probable supplementary questions to assimilate a much wider set of replies from which to package its eventual single response. Quite how this effects optimising content for AI is hard to pin down - but AI researchers at Profound have had a good go at it. A deep dive into data from 2,867 ChatGPT fanouts shows that Google rank matters only for the first query or two - which matters for giving citations. After that, answers barely link to SERP position at all, and become almost impossible to cite - making optimising for AI hard to do.
Read
Decoding GEO
Grab some popcorn and take some time to watch Lily Ray’s new video on GEO, SEO, LLMO, organic search, and other topics vital to visibility.
Watch
Canadian publishers bite
Another sign that AI companies can’t just bulldoze copyright rules is coming from Canada. A court there says that OpenAI must answer a suit over scraping news content for ChatGPT, and publishers from Postmedia to CBC are rubbing their hands. The court awarded an initial $260,000 in costs to the publishers and argued that OpenAI’s arguments over lack of jurisdiction are irrelevant, and Canadian courts are empowered to try the case.
Read
Surviving the AI murky waters
Is the AI content flood upon us? Is quality a rare commodity? SEO sage Barry Adams discusses hallucination-prone slop, the need for publishers to stand out and whether unique value is optional, or needed for survival.
Read
Where news dries up, so does democracy
Turns out that when local papers vanish, legislative transparency quickly follows suit. Two new studies, aptly named Dark Deserts and Hometown Transparency, which came out of the University of Florida’s Brechner Center led by David Cuillier, show that states which lack newspapers also lack transparency. Those states are more secretive, and less averse to fees, delays, and excuses. Turns out that newspapers aren’t just good at informing - they keep democracy alive.
Read
No PR, just press
Twice a year, the last few of America’s family-owned newspapers retreat behind closed doors to compare notes. The Independent Newspaper Group (ING) has been trading both numbers, war stories, and survival hacks for nearly 40 years. Members are swapping ideas that are actually working, and admitting to those that aren’t.
Read