Lídia
All posts

2 Oct 2026 · 7 min read

Writing got cheap, being read got expensive

Writing became almost free and being found got much harder. What lost its value, what became scarce, and what I do differently now.

A few years ago, I felt like my job was to read minds. The game went like this: figure out what someone is typing into Google, understand what they actually meant, and write the content that answered it better than anyone else. Win that, and they’ll land on your website. They will meet your brand. Sometimes, they’ll even buy it.

It was a good game. It had rules and challenges, of course, but you knew efforts would pay off. But in the space of a few years, two things happened that broke it.

Thing one: writing became (almost) free

Once upon a time, a blog post would cost somebody an afternoon. Researching keywords and competitors, drafting an outline, going through a few rounds of writing and editing. Now, it costs a prompt.

The digital marketing agency Graphite has been sampling English-language articles from across the web and running them through AI detectors. Their latest numbers, reported by Axios in May, are hard to ignore: within a year of ChatGPT’s launch, about 36% of new articles were primarily AI-generated. Within two years, 48%. Since early 2025, it has hovered at around half.

This means every other article published on the internet, roughly, was written mostly by a machine.

(AI detectors aren’t perfect, and Graphite says so itself. But even if you knock a few points off, the direction is not in doubt.)

The dictionary noticed too. Merriam-Webster’s 2025 Word of the Year was slop, which it now defines as low-quality digital content produced in quantity by AI. When your industry’s output becomes the dictionary’s example of rubbish, it’s worth pausing for a second.

Thing two: being found got much harder

While everyone was busy producing more, Google started answering questions in a different way. What used to be a full page of possible answers, split into paid and organic results, with the highest-bidding ads and the most relevant organic content at the top, is now a neat AI summary that appears before you see anything else. Convenient for you. Less convenient for whoever wrote the pages it learned from.

Pew Research Center tracked the actual browsing behaviour of 900 US adults across nearly 69,000 Google searches. When an AI summary appeared, people clicked on a regular search result in 8% of visits. Without one, 15%. They clicked on the links inside the summary just 1% of the time. And they were more likely to simply stop browsing altogether: 26% of the time, versus 16%. Ahrefs looked at it from the publisher’s side. In its December 2025 update, the presence of an AI Overview correlated with a 58% lower click-through rate for the page ranked first.

So here’s the situation: there’s more content available than ever, but fewer people are clicking on any of it. You don’t need an economics degree to know what happens to the price of something when supply goes up and demand goes down.

What got cheap, and what got scarce

Let me be specific, because “AI changed content” is the kind of sentence that fills a calendar and says nothing.

What lost its value is the competent explainer. The “What is X?” blog post and the “10 tips for Y” listicle. The content every content writer had in their toolkit and that, well built and with solid keyword research behind it, would bring traffic to the website and keep the analysts happy by hitting the traffic goals.

This kind of content lost value not because it was bad (a lot of it was actually genuinely useful), but because it doesn’t need human intervention anymore: a machine can write it, and another machine can summarise it before anyone visits your site to read it. If your content can be accurately compressed into a 67-word answer box, the answer box will happily do that for you, and keep the reader.

That’s the truth: the content most exposed to AI is exactly the content most of us were trained to make.

What got scarce

When something becomes abundant, the things it can’t produce become valuable, and there’s a lot AI can’t produce.

It can’t produce your point of view. A language model is, very roughly, a machine for predicting the most likely next word. It is built for the average. A brand voice is, by definition, a deliberate refusal to sound average. Most of the hard work on tone of voice isn’t choosing what to say, but how to sound different from anyone else. Machines can’t create an original POV, because they’re trained on the average.

It can’t produce first-hand experience. When I started an AI newsletter at KPMG, the irony wasn’t lost on me. Summarising global AI research is precisely the kind of task a tool could do. What made the newsletter worth opening was everything around the summary: translating global data into what it meant for a Portuguese business, and having local experts put their names to each edition. The research was the raw material, but building relevance was the product.

It can’t produce trust. Audiences have noticed the difference. In a survey of 8,000 consumers across eight countries, Klaviyo and Datalily found that only 7% said visible AI-generated marketing made them trust a brand more, while 31% said it made them trust the brand less (via eMarketer). Meanwhile, when AI answers and brand messaging contradict each other, a Skyword survey found most people don’t believe either: 54% go looking for a third source (via ContentGrip). Trust isn’t only built on your channels anymore, it’s built in the places people go to check you.

The content writer didn’t die, their job just changed

“AI will kill content writers” is kind of slop on its own. Writing still matters, but content now has two readers: the reader, and the machine that decides whether the person reads you. Being cited in an AI answer is the new being ranked first on Google Search. The difference now is that people won’t open one, two, three links just to read about the best cooking pot in 2026: they’ll trust the AI summary that tells them everything they need to know about cooking pots and which to buy, in just a few words.

And just like I mentioned before, yes, most of the time people will not go beyond the AI summary, but those who still click on your website are the ones who’ve nearly made up their minds, so be clever about what you offer them.

This also means you need to check your metrics better. Pageviews were always a proxy. Fewer visits can mean better business.

So what do I actually do differently?

While trying to stay relevant, I’ve built my own set of rules. Five things, in no particular order of virtue:

  1. I stopped writing what the answer box already says. I either go much deeper or I don’t bother.
  2. I start from what only we can say. Our data, our experience, our opinion, our mistakes. If a competitor could publish the same piece by changing the logo, it’s not worth it.
  3. I treat voice as strategy. It’s the one thing that survives summarisation, because a summary of a distinctive voice is still recognisably somebody.
  4. I measure differently. Citations, branded searches, the quality of who shows up, not just how many.
  5. I use AI for the boring bits. Research, first passes, restructuring, catching my typos at midnight. I’m not a purist: I just don’t let it decide what I say.

The calendar was never the point

For years, a lot of content marketing was a volume business. More posts, more keywords, more everything. AI didn’t kill that model, but it made it free, which is worse, because now everyone can do it.

What’s left is the part that was always the job: having something to say, and saying it in a way that sounds like someone.

Good stories sell. Bad content just fills the calendar. And the calendar, it turns out, can now fill itself.

Get The Story Sells in your inbox every week.

Subscribe on Substack