This newsletter is more of a thought piece. I’ve been thinking about the impact of generative AI on content production, and how that could impact the quality of content that people are exposed to.

Since the start of the generative AI hype cycle, much of the discourse has been about how GenAI might replace and/or de-incentivise traditional (human-written) publishing. Those are valid and necessary discussions to have.

Here, I want to add my own two cents these discussions by outlining what I believe GenAI is doing to the quality of content that’s available to readers on the web.

The Content Quality Bell Curve

Before GenAI, the quality of all content available to read for free on the web - every article, review, blog post, explainer, opinion piece, and so on - could be mapped to a standard distribution bell curve.

Pre-GenAI, most of the terrible content on the web was automated spun content, easily identifiable as machine-generated and intended mainly to game search engines. While the number of terrible content pieces was high, most of it was never meant to be read by humans, so its impact on the online information ecosystem was extremely limited.

We can say that the majority of content meant for human consumption would’ve fit in the middle of the bell curve; mediocre content, often produced by low-paid freelancers, with little to no genuine insight or information gain.

Content beyond the mediocre mountain of the bell curve - the good, great, awesome stuff - sits on the right half of the bell curve. Mostly free to read, relatively easy to find, and usually published on high-authority websites.

Generative AI is dramatically changing this distribution. The widespread availability of large language models has enabled the industrial-scale creation of thoroughly mediocre content.

Mediocrity Is Built-In

Mediocrity is inherent in the nature of LLMs. Their training data is the entire corpus of available content on the web; all the terrible, the mediocre, and the awesome content out there and everything in between. The LLMs can’t determine quality and selectively ingest content; they’ll absorb everything on offer, regardless of its merit.

There is probably some measure of training data selection at work; Google doesn’t index all content, the Common Crawl corpus is somewhat moderated, and I imagine (and hope) AI engineers will have some guardrails in place to prevent their LLMs from being trained on pure gibberish.

What we end up with is a set of LLMs that are trained on everything without a qualitative measure of anything.

(This is a bit of a simplification; LLMs do use some qualitative training data that likely skews their models a bit. Nonetheless, due to the necessity of scale, the vast majority of their training data will be quantitative and lacking any qualitative qualifiers.)

This means that the output generated by LLMs is, by definition, mediocre. You can’t ask a GenAI system to create ‘high quality content’ because the system quite literally doesn’t know what ‘high quality’ is.

I’m sure some AI ‘prompt engineers’ can find specific prompts that result in a simulacrum of quality, but that’s exactly what it is. LLMs don’t know anything. They’re highly advanced word predictors, without any semantic understanding of what they regurgitate.

GenAI Content Changes The Curve

This leads back to the concept of information gain. Because LLMs are trained on the available knowledge provided to them, they can’t come up with anything new. They can’t add any new insight or knowledge, because they don’t have any insight or knowledge.

Without branching off into a philosophical tangent about what it means to have knowledge, you can easily see this for yourself when you ask an LLM to create content for you. You get a perfectly digestible but always intensely mediocre product.

The widespread adoption of LLMs by (lazy) content creators on the web means that the content quality bell curve is dramatically shifting.

Bad content is being replaced wholesale by mediocre GenAI content. The ‘mediocre content’ bulge of the bell curve grows and swallows most of the ‘terrible content’ left side of the curve.

We end up with a web where the vast majority of available content is average. Much of it is created with GenAI, sitting alongside mediocre human-produced content.

The effects of this change are subtle, but noticeable.

Mediocrity Is The New Norm

Amidst this flood of mediocrity, publishers will have to work harder to stand out. Pre-GenAI, a publisher could survive just fine sitting in the middle of the bell curve, or even just to its left. The mediocre middle was big but still stood out from bad content. People would still see it as worthwhile to consume.

That is no longer the case. Mediocrity is the new norm. It’s not enough to merely be ‘good enough’. You need to be on the right edge of the bell curve to have any hope of getting people’s attention. The further right on the curve you are - the better your content, the stronger your information gain - the more chance you have of surviving and thriving.

But being awesome is hard. Moreover, it’s expensive. New knowledge doesn’t spring into being on its own (again, philosophers, hush for a second and let me make my point). It takes work and effort to uncover and report new information.

This shift in content quality is happening at the same time where the web’s revenue potential is the lowest it’s ever been. Thanks to monopolising behaviours from the tech giants, it’s harder than ever to make money from publicly available content on the web. And now, thanks to LLMs, it’s harder than ever to create content that gets people’s attention.

GenAI is in the process of shifting the quality distribution of online content. Publishers need to be aware of this, and ensure they’re focusing on the right side of the bell curve.

Being a mediocre, forgettable publication is a death sentence. You need to stand out and produce content worth consuming, worth subscribing to, and worth paying for. Anything less than that, and you will find yourself drowning in an ocean of GenAI mediocrity.

Miscellaneous

Here are some of the latest Google docs, interesting articles, and newest SEO insights from the last few weeks.

Official Google Docs:

• Google News transitions to automatically-generated publication pages

• Simplifying the visible URL element on mobile search results

• Robots Refresher: introducing a new series

Latest in SEO:

• Google Publisher Center Changes: Automatically Generated Publication Pages - SER

• The impact of AI Overviews on SEO; analysis of 19 studies - Growth Memo

• 2025 SEO Benchmarking Report - Similarweb

• Branded search for news SEO - WTF is SEO?

• Controlling Your Brand Position Online With SEO - SEJ

• The Disconnected Entity Hypothesis and Google Spam Updates - Shaun Anderson

Interesting Articles:

• OpenAI Admits That Its New Model Still Hallucinates More Than a Third of the Time - Futurism

• ChatGPT referrals to top publishers up eight times in six months but still negligible - Press Gazette

• Microsoft Study Finds AI Makes Human Cognition “Atrophied and Unprepared” - 404 Media

• There Is No AI Revolution - Ed Zitron

• MAKE IT FAIR: UK publishers unite against government's AI copyright madness - Content Aware

• The Great Media Reset: AI, Consumer Revenue, and the Future of Publishing - Matt Karolian

• US immigration is creating a mirage of mass deportations on Google search - The Guardian

Thanks for reading and subscribing, and I’ll see you at the next one!