When Your Audience Names the Problem, That's the Signal

LinkedIn just added a "Seems like AI slop" button.

Think about what that admission actually contains. A platform built on professional self-presentation, owned by Microsoft, deeply invested in the AI moment, looked at its own feed and decided it needed a formal mechanism for users to say this feels fake. Its chief product officer said the company blocks hundreds of thousands of automated slop comments a day and has stopped billions of automation attempts in a matter of months. And in the same breath, LinkedIn quietly killed its "enhance your post" tool and replaced it with something that proofreads without changing your voice.

Read those two moves together. They tell you everything.

The platform is not anti-AI. It is responding to a signal its audience has been sending loudly and creatively for a while now. People coined a term — "slop" — and shoved a whole category of content into it. That is not idle snark. That is a market telling you where the line is.

Suspicion is not a bug. It's a signal.

I made this argument in The Glass Horse, and the LinkedIn story is that argument playing out at scale. Christopher Nolan called AI "a transparent horse, made of glass — everybody can see what's going on." His point was that public skepticism of AI is healthy, and healthy skepticism is how you get the best out of a new technology instead of blind faith that it will all work out.

Most companies treat that skepticism as friction. Resistance to change. A messaging problem. Something to smooth over with a better rollout narrative.

That is exactly backwards. When your audience invents a mocking name for your pattern of AI use, they have already run your risk assessment for free. The question is whether your operating model is built to hear it.

This is where the Human-Driven Operating Model earns its keep. HDOM does not ask only whether a feature works. It asks whether the experience you shipped is the one you actually meant to create — and whether the humans on the receiving end agree.

Three AI UX patterns

Before you can read the signal, you need a shared vocabulary for what the audience is reacting to. So here are three recurring patterns of how people experience AI inside a product or content stream. I'll expand each of these in a follow-up post, but the working definitions are enough to get us moving.

Invisible Assistance. AI improves the experience without demanding attention to itself. It proofreads, organizes, summarizes, cleans up friction, and preserves your voice. You barely notice it's there. LinkedIn's replacement tool — proofreading without changing how you sound — is a deliberate move toward this pattern.

Visible Augmentation. AI is noticeable, and that's fine, because the value is obvious and the human intent still holds. A labeled summary. A clearly marked suggestion. The audience can see the machine, understands why, and trusts the result.

Slop. AI's presence degrades trust, authenticity, or experience quality. It reads as generic, mass-produced, engagement-farmed, or hollow — machine residue where human meaning should have been. Auto-generated comments. Voice-flattened thought leadership everyone can smell from six feet away.

Here's the part that matters most: the pattern is not baked into the technology. It's relational. The same underlying capability can land in any of the three depending on context, stakes, disclosure, quality, frequency, and whether a human voice survives. AI proofreading feels like Invisible Assistance. The identical model writing your condolences on a layoff post feels like something closer to sociopathy.

The alignment problem nobody names

So which pattern are you building?

Ask your Product lead and your UX lead that question separately and you may be unpleasantly surprised. Product often assumes Invisible Assistance — users will love the efficiency. UX often suspects Slop — if we push this, people will read it as fake. Leadership, meanwhile, is watching a dashboard and reading any uptick in activity as proof the thing is working.

That is not alignment. That is three people optimizing three different experiences and calling it one roadmap.

If Product thinks the feature is helpful assistance and UX suspects the audience will read it as slop, you don't have alignment. You have a future metrics argument.

You will have that argument. It will just happen later, after launch, when the numbers are ambiguous enough for everyone to interpret them in their own favor. Better to have it now, on purpose, before you scale.

And once you've aligned, you don't get to stop there. You have to test the assumption with real users. Not a Slack poll with three emoji reactions. Not the enthusiasm in the room where the feature was born. Not the vendor deck. Capability is not the same as desirability, and internal excitement is not evidence of anything except internal excitement.

Mapping the patterns to HDOM

This is where the framework gets practical. Each HDOM stage forces a version of the same question — which pattern, and does the audience agree?

Frame Opportunity. What AI UX pattern does the market actually want here? The failure mode is mistaking capability for demand. The tools can generate comments, so leadership assumes people want more generated content. LinkedIn's data suggests the opposite: users wanted less synthetic noise and more help preserving their own voice. If you frame the opportunity around throughput, you've already missed the signal.

Define Experience. Which pattern are we intentionally designing for? Name it out loud. This is where Product and UX put their assumption on the table in the same words, before a single sprint gets funded.

Commit as Business. Are our incentives and KPIs set up to reward that pattern — or to accidentally manufacture slop? Show me the incentive and I'll show you the behavior. If the operating model rewards volume and top-line activity, AI will happily produce volume and activity, and it will look productive right up until trust collapses.

Build for Impact. Are we measuring whether the audience experiences the intended pattern? This is the discipline most teams skip. You cannot instrument only output, cost per asset, and clickthrough. You have to instrument the human reaction underneath.

Sense & Respond. If the audience experiences Slop where we intended Assistance, what do we change? Narrow it. Add disclosure. Shift from generative to assistive. Put a human back in the loop. Or kill it. LinkedIn's whole sequence — reporting button, killed enhancement tool, stronger verification — is Sense & Respond maturity in public.

Measure trust, not just motion

Here is the measurement trap: usage is not consent. If the message from the top is use these tools or get left behind, people will use them and the dashboard will glow. You'll have learned nothing about whether they trust the output, understand where it breaks, or are quietly cleaning up work they don't believe in.

So track two different things and watch the space between them. Output and activity on one side. Trust and authenticity on the other. Measure the hides, the mutes, the reports, the abandonment, the "this feels fake" sentiment. The gap between what people do and what they trust is your real story.

When that gap opens and you ignore it, you start accruing trust debt — the quiet, compounding cost of shipping experiences your audience tolerates but does not believe. It doesn't show up on the launch dashboard. It shows up in month three, when the "successful" rollout starts falling apart and nobody can quite say why.

And watch for pattern drift. What starts as Invisible Assistance can slide into Slop as novelty wears off, volume climbs, or someone decides to squeeze a little more automation out of a feature that was working precisely because it stayed modest. The pattern you launched is not the pattern you keep. You have to keep sensing.

The real job

None of this is about deciding whether AI is good or bad. That question is boring and beside the point.

The job of a human-driven operating model is not to persuade your audience that their instincts are wrong. It's to figure out whether those instincts are warning you about a real experience failure. The young hire rolling their eyes at "AI slop" and the LinkedIn user reaching for the report button are not obstacles to your strategy. They're your early-warning system, running for free.

The horse is made of glass. Your audience can already see what's inside.

The only real choice you have is whether you ask them what they see — or let the dashboard tell you a story you'd prefer to hear.

This is the first of two posts. In the next, I'll take the three AI UX patterns — Invisible Assistance, Visible Augmentation, and Slop — and expand each one: what it feels like, how teams misclassify it, and the signals that tell you which one you've actually built.

John Oliver covered the dangers of AI on his weekly HBO show, calling it “worryingly corrosive” for society.

On Last Week Tonight, Oliver said that the “spread of AI generation tools has made it very easy to flood social media sites with cheap, professional-looking, often deeply weird content” using the term AI slop to describe it all.