Recently I did a number of user interviews for a client, and 90% of my interviewees brought up AI use entirely unprompted.
They were looking at a generated image of a £50 note. I hadn’t originally clocked it as AI when conducting my normal teardown of the site, instead thinking it was probably a stock image (though I accept the two lines are blurred these days as stock sellers generate lots of images to avoid paying royalties).
The sentiment of each user was always negative. They are most used to seeing images used this way in scams or shortcuts. They’re a signal of “lack of investment” which then is seen by that user as something to mistrust.
Since then I’ve been thinking about AI. I wish I could think less about AI really as it’s kinda had a stranglehold on my industry for a few years now as it rapidly becomes the only game in town and where all the money is spent. Nonetheless I needed a newsletter article, so I’ve been thinking about it and reading about it to write this article.
Leaders love AI, workers like it less but tolerate it in places and the public actively dislikes it.
This tends to cut across material lines. Who’s getting paid changes how strongly you feel about it. It’s causing far more annoyance and harm than benefit at a societal level with huge numbers of people losing their job, having their opportunities limited or otherwise being forced into radical change which they see little benefit from in either free time or higher income. You can couple that with the rising tide of low-effort, low-quality, averaged marketing-sounding copy and images creeping into all the places people would originally seek refuge (social media, public spaces, literature and news, entertainment).
Research is struggling to keep up with the rollout of AI in different spaces, so a bunch of the references I give here are going to be from 2025. Since then, Anthropic’s stated revenue has grown 7 times, which we can assume means that more people are using Claude more often. A recent Reuters report has somehow managed to get details which suggest that the company is still wildly unprofitable with seemingly no clear path to profit short of transforming ALL work in a way that no previous change ever has (and they’re including electricity in that), so please understand that I’m only referring to “how many people are using Claude” here, rather than suggesting Claude is really good and profitable for anyone using it.
So whatever is going on in 2026 with AI, we can assume that there’s just a lot more of it, and general sentiment I see amongst consumers is not optimistic or positive about what the world will look like with a lot more LLM usage in all aspects of everything, the way these tech companies are pitching their products.
There is an emerging gulf between perception and reality when it comes to AI. People with email centric jobs have, no-doubt, found that the language model is more than capable of writing an email you didn’t want to write. One of the earliest applications of LLMs and one of its most popular applications still is this. The expertly crafted email that says no, and the email that says “no” are performing similar tasks and getting similar results, but they aren’t necessarily actively improving results. There was an IAB report from Jan 2026 which shows that execs tend to feel positive about AI ads (82%), and only 45% of consumers feel the same way, with younger audiences (Gen Z) feeling the most negative about AI ads. I think this is a case of mismatched incentives, but it will start to really bite when executives are tasked less with constraining costs and more with actual performance, where AI struggles. If the people you’re advertising to hate your ads, that’s a problem.
“During sampling, models predict tokens based on statistical patterns in training data. Safe design choices–those that work universally and offend no one–dominate web training data.” Anthropic, 2025
Then, reader, is the stuff I think you have noticed too. The Claudeification of everything you see. Cream and Terracotta websites with pills and everything way too close to each other, more eyebrows than a Milhouse Van Houten cosplay convention, more copy everywhere taking 500 words to say something that could have been said clearer in 100. Context windows bleeding into documents. Everything being reported as if it’s a game-changing revelation. The words “quietly” “honestly” “carries” “holds” and other weird AI tics that make you talk like a creep. All of it, in sum, makes something we took relatively for granted (reading things on the internet) feel like wading through sludge.
There is a reason we call it slop.
You are mistaken if you don’t realise that everything reliant on people engaging is an attention economy. Convincing people to part with their cash, which they now have less of generally, in particular requires a lot of attention. You can’t really bore somebody into buying your product. You have to win them over. You have to delight them and that’s really, really hard to do if you, the billion dollar SaaS company and the kebab shop in town all have the same tone of voice, message and visual style.
I think there’s a particularly interesting poison pill swallowed by a lot of platforms in that, they’re tech companies that want to convince you that they’re communities. So you see social media platforms with AI strategies, but whose core product is actively being destroyed by AI. People want to use those platforms for human connection. Either with their friends, or with like-minded people. Increasingly they’re finding a generative AI wall between themselves and others, and it’s driving them away from the platform. Fewer users, spending less time on the platform means you show them less ads. That means you make less money. Your ads being less effective, means you lose interest from advertisers and they go where the people are instead. You rot away the platform bit by bit.
Platforms that are not rewarding to be on ironically makes people MORE likely to automate them. See, for example, the proliferation of AI posts on LinkedIn. The push to appear to be some kind of “thought leader” in order to better secure your future career prospects or new sales was always a LITTLE spurious to begin with but it is a narrative LinkedIn themselves actively pushed. Now that platform is estimated to be somewhere between 40 and 80% AI generated posts1 and… well… You don’t need to get those from a random stranger who got them from Claude do you? You can just ask Claude to give you those thoughts and insights directly and cut out the middle man.
As LinkedIn becomes more and more a home for people to talk (via an LLM), but not to meaningfully listen or engage, the more likely it becomes to be a presentee platform. A place where your digital avatar handles your machine-written press releases, to be consumed by nobody at all. A poison pill for the platform and another recursive content repository on a dead internet.
To understand why perception of AI generated content changes the closer you get to senior leadership, you have to understand the material incentives, and that is about lowering headcount in businesses. Jack Dorsey cut 40% of staff at Block and said every CEO should follow suit. Benioff said they were “not hiring more engineers”. I am old enough to remember the halcyon days of 2023 when we were told AI was just coming for all knowledge work, and not the scary new days of 2026 when tech leaders are having loud public debates on whether it’ll simply kill us all instead. Now, how much of the AI glazing around headcount can be tied to genuine credulity amongst leaders that the work that happens at their companies was so Fordian in nature that it could be done by one or many chatbots, or whether they were making layoffs and AI was a way to get investors excited about them instead of having mass layoffs viewed as a sign of lowered future productivity we will never really know, because to do that we’d have to know the minds of these guys. What we can know however is that after making the layoffs and implementing a lot of ‘AI’ that the returns aren’t really there. PwC reported that 56% of CEOs see no financial benefit at all. NBER reported that 89% of execs saw no productivity change and Goldman Sachs couldn’t link AI investment (which has a load of money thrown at it) to anything economy-wide, finding only marginal gains in software and “customer support”.
More recently CEOs have started denying that they were making cuts due to AI. Again, impossible to understand how much of this is PR.
Workers, on the other hand, are the day-to-day users of AI for jobs and it’s making them highly sceptical about what AI can actually do. New terms are emerging to describe how it feels to work with so much AI but so little care. We are hearing about ‘workslop’ and ‘slop grenades’ as a way of delivering work-shaped objects but that do not meaningfully help anyone, and indeed add layers and layers of extra effort to decode and understand because of the particular quirks in how AI writes, communicates and processes information. What might have once been a Slack message saying “We need to look again at our Meta ads, they’re sending us junk leads” is now an 18 part document epic full of dead-ends and false positives. The effect it has is that there’s a lot of documents flying around a company, all of them long and partially informed and none of them read. If an employee themselves were this disruptive and confusing, it’s likely nobody would want to work with them. Claude, however, doesn’t work here formally and he has no objectives or key results, but we’d REALLY like him to work on everything.
In their lives, it’s making it harder to keep on top of everything, and it’s stopping them from getting paid more or moving jobs to somewhere that pays them more. I am not surprised they are not keen on it.
And finally, the public. They actively dislike it and I think this is a decent way to understand sentiment when you remove material incentives. There is, of course, little separation between the public and the workers. Most of that public are workers. 59% of UK adults reported that they use generative AI. 74% worry it’s inaccurate. 71% want it to be legislated. The irony is currently that, AI has such an image problem with the public, that disclosure would actively work against it and generative AI is currently only really able to perform if it is able to replicate human production closely enough as to be received differently.
So what should you do about it? Well I wish I had that answer for you my reader but I am not sure that I do. I can tell you what I am doing about it though. I do use AI in my work, in places. Mostly highly targeted and constrained generative AI. I also cannot trust it to do stuff unattended and I NEVER want it to communicate for me. As a practitioner, I am paid to understand what’s going on and to have control of that. If I cede control and understanding to something that cannot do it, but can simulate it up until it has to deliver, I have destroyed my own business.
When I do have it assembling information or structuring information for me to read and act upon, even then I’m aggressively and ruthlessly editing the output to control for the stuff I know it’s especially weak at. Now… that’s a hard lesson because it requires you to still do a lot of writing and reading and learning. Outsource that to a machine and you’ve lost. I can tell you the advantage of NOT outsourcing it to a machine though and that is that other people will.
Other businesses will actively lower the quality of their work across the board. Because you are still understanding, writing-for and working for humans you will connect where they can not. You will sound different where they do not. You will ship products and they will ship slop. Your work will shine like a beautiful and desirable oasis in a desert of generative sand.
Look after yourselves. Another world is possible.
1 Pinch of salt on this one. The only companies doing this research are ones trying to sell AI detection software so they’ve got an incentive to find lots of it.
This piece first went out to the Another Web is Possible newsletter in September 2026.
Subscribe here to get it in your inbox.
.png)