ai slop and trust

AI slop is mass-produced, low-quality AI content, and it eats away at the trust customers put in what you publish. Here’s the catch. Research from 2025 and 2026 shows that a plain “made with AI” label can actually lower the credibility of content that’s completely accurate, and yet consumers still ask you to be honest. So how do you square that transparency paradox? You tie the disclosure to the situation. In customer contact you always say a machine is talking. For marketing content you personally stand behind every word. And internal AI use? That’s nobody’s business. What’s left as the real differentiator is being human. Work that feels like machine work loses either way.

why is everything suddenly called ai slop?

Because in 2025 the saturation became official language. Merriam-Webster made “slop” its word of the year 2025, and defines it as low-quality digital content, usually churned out in large quantities with artificial intelligence (Merriam-Webster, 2025). A dictionary only reaches for a word once everybody’s already saying it, and everybody’s saying it because everybody can see the problem. Look at Spotify: it pulled over 75 million AI-generated spam tracks in 2025 (CNBC, 2025). That’s not some linguistic footnote. That’s a market silting up right in front of you.

For you as a business owner, that’s a market signal. Your newsletter, your website, your socials, they all compete now in a space customers have learned to link with slop. So the question’s a simple one. Do you drift along with the current, or do you set yourself apart from it? That most AI projects strand even earlier, somewhere between pilot and production, is a related story but a different one, and I dig into it in the production gap.

do customers still trust ai content?

Less and less. And the Dutch are warier than most. The big global study by the University of Melbourne and KPMG, 48,000 respondents across 47 countries, puts it plainly: 33 percent of Dutch people are willing to trust AI, against 46 percent worldwide (KPMG, 2025). The research agency Norstat found 44 percent of Dutch people disapprove of AI use in advertising, websites and brochures, and that climbs to 52 percent if AI in advertising keeps growing. Among the over-sixties, disapproval sits at 53 percent (Norstat via Emerce, 2025).

The uncomfortable part? That same consumer can barely spot AI content in the first place. Earlier Norstat research shows over half of young people struggle to tell AI-generated content apart (Norstat via Emerce, 2025). So “they’ll never notice” is exactly the wrong thing to reassure yourself with. Disapproval runs on suspicion, and on finding out after the fact, and the suspicion is growing faster than anyone’s ability to actually detect the stuff. Gartner measured in 2025 that 53 percent of consumers have little trust in AI-powered search results and summaries (Gartner, 2025). And it’s not just here: trust in AI-generated content fell from 73 to 55 percent between 2023 and 2025, according to the consultancy Capgemini, and that’s across every age group, Gen Z included (Capgemini via MarTech, 2026).

why doesn’t “just be transparent” simply work?

Because an AI label can drag down credibility even when the content is spot on. Two separate lines of research show it, and they’re measuring different things, which is the part people miss. The first one is about trust in the sender. Across thirteen experiments with over 5,000 participants, Schilke and Reimann found that people consistently place less trust in whoever openly discloses AI use, even when those same people lean on AI all the time themselves (Organizational Behavior and Human Decision Processes, 2025). The second one is about the credibility of the content itself. An experiment with 433 participants, this time around science communication on social media, found an AI label made accurate information less credible and misinformation more credible. They call it the truth-falsity crossover effect (JCOM, 2026). Research from the University of Twente lands in the same place: AI labels on news articles push perceived credibility down, no matter what the text actually says (University of Twente, 2025). One caveat, and it matters. Those last two studies were run on news and science communication, so whether the effect carries over one to one to advertising, honestly, nobody’s proven yet.

And this one chafes, me included. I build with AI for clients every single day, and I also happen to think you should be honest about how you work. But the research is telling me that honesty costs me credibility on every piece of content. Anyone selling “just be transparent” as the easy answer is keeping that price quiet. Trouble is, the opposite advice, keeping it hidden, is a time bomb. According to YouGov, 32 percent of consumers would trust a brand less once its AI use becomes known, against 15 percent who would then trust the brand more (YouGov via MarTech, 2026). So what you gain in credibility per piece of content, you can lose in brand trust the day it comes out.

human-made as a selling point

The countermovement, though? It’s already commercially proven. Polaroid won an Ad Age Creativity Award 2026 with a global campaign that flat-out rejects AI and data centers and puts the analog, tangible photo forward as the answer, right down to a billboard against data centers at Coney Island (Ad Age, 2026). Scale doesn’t set you apart anymore. Anyone can spin up a thousand texts a day. Realness does.

That said, “anti-AI” is too blunt an answer for most companies, and the research shows you why. According to work by the software company Bynder, 82 percent of consumers have no problem at all with brands using AI for copy, as long as the result feels human (Bynder via MarTech, 2026). So the acceptance is there. It just comes with a condition, and the condition is quality and feel. Customers are tuned into the substance, way more than into the label.

my line: three kinds of ai use, three kinds of honesty

For myself and for clients, I sort it into three lanes, each with its own disclosure rule.

Customer contact. If a customer is talking to a machine, they deserve to know. Always. From the very first line. A chatbot that poses as a human is borrowing trust it can never pay back, and when people find out, it hits your whole brand and echoes through every contact after that. Openness pays off here too. The software company Qualtrics, in its own trend research, finds that customers are more willing to share data with organizations that are transparent about how it is used (Qualtrics, 2026). And yes, this openness is turning into a legal requirement in Europe as well, which I cover on its own in the ai act without panic. The reassuring bit? You can do it without losing trust. The Dutch National Voice Monitor 2025 sees trust in AI-supported customer contact rising, younger people especially, while 78 percent of Dutch consumers have now dealt with a chatbot at some point, up from 70 percent a year earlier (Frankwatching, 2025). And research by Y.digital and Markteffect saw negative attitudes toward AI chatbots drop from 52 to 32 percent (Consultancy.nl, 2026). Trust follows proven performance. Simple as that.

Marketing content. Here, AI is a tool to me, the same way the spell checker and photo editing have been for years. I feel no duty to list under every blog post which tools I reached for. The photographer doesn’t credit his lights either. Two hard limits do apply, though. One: every number, every claim, every experience in the text has to be real and checked by a human, with a name underneath that stands for it. Letting AI write up an experience you never actually had is deception, label or no label. Two: synthetic media that suggests something real, an AI photo of your “team”, a generated product shot that isn’t your product, a made-up review, that’s always deception. No disclosure fixes it. You just leave it alone.

Internal. Analyses, drafts, summaries, code. Here you owe nobody a disclosure. The client is buying the result, and your responsibility for it. What your kitchen looks like is your own business, as long as what leaves the kitchen meets your standard.

And the line running underneath all three lanes is always the same. Deception starts the moment the receiver concludes something about the sender that isn’t true. That they’re talking to a human. That you tested the product yourself. That the review came from a real customer. As long as those conclusions hold up, AI assistance is just honest work.

checklist: how to stay out of the slop

Here’s the routine I actually run, in the order I run it.

First, I sort every bit of AI use into those three lanes: customer dialogue, public content, internal. Edge cases just follow the strictest lane, no debating it. Then, every bot has to introduce itself as a digital assistant in its first line, and it always offers a route to a human. Non-negotiable. Next comes the one publishing rule I never bend: nothing goes live without a human doing the final edit and a name that signs for it. After that, I check every number and every claim against the original source, never against the AI’s summary of it, because that’s exactly where the quiet errors hide. Some things I ban outright, full stop: fabricated reviews, fabricated experiences, and generated images that suggest something real. Then the read-aloud test, and this one I swear by: read the text out loud, and cut anything you’d never actually say to a customer that way. And last, every quarter, I reread the existing content with those same eyes. Slop sneaks in through the back door of time pressure, every time you’re not looking.

Now, what you don’t need to do: run AI detectors over your own texts, or slap a “100 percent human-made” badge on every page. Detectors aren’t reliable, and a badge like that answers a question your customer never asked in the first place. Your customer is asking something else. Is the work right? Does it feel like you?

So, where I land. I use AI every day and I make no secret of it, and at the same time nothing that leaves my workshop is allowed to feel like machine work. Those two convictions get along just fine, as long as you keep the lanes apart. Machines say what they are in dialogue. Humans sign for content. And internally, you do whatever works. The label was always the side issue. Being answerable for the substance, that’s the main thing.