The feeling is specific and most people who read for a living recognise it immediately. You open something. Within two sentences you know it will not answer your question. You open the next result. Same. Somewhere around the fourth attempt, a small calculation happens, and the calculation is about whether the search was worth starting.
What is actually being spent
The standard framing of slop is that it wastes time. The more precise version is that it adds a step to everything.
Before, reading carried an implicit assumption that the writer had done some work. That assumption was never fully reliable, and it was reliable enough. Now it has to be tested on every page, in a context where the test itself is unreliable, because you cannot tell what wrote a text by reading it.
So every document gets two jobs: work out whether it is worth reading, then read it. The first job used to be nearly free.
Research framing on this subject treats the result as a system-level cost rather than a personal complaint. A Columbia University report on AI slop and the information ecosystem lists discernment fatigue among the concerns driving research interest, alongside saturation and aesthetic degradation.
Why it is rational, not fragile
There is a version of this discussion that treats fatigue as a failure of resilience. That reading gets the economics backwards.
When the cost of verification rises and the expected payoff falls, checking less is the correct response. Falling back on sources that have not burned you yet is also correct. Abandoning a channel that has become unusable is arguably correct too. Each of these is what a sensible person does with a budget.
The evidence that people behave this way is easiest to see where the cost is itemised. The 2025 BetterUp and Stanford study of 1,150 US desk workers found that 40% had received AI-generated work content that looked finished but was not, and put the average time lost per incident at 1 hour 56 minutes. In that setting, the fatigue has a timesheet.
What it looks like
Four patterns, in rough order of how healthy they are.
Colder reading. You skim for the specific fact and leave. This is efficient and it works, at the cost of never being surprised by something.
Narrower sources. You find four or five places that have not wasted your time and you stop looking further. This is the most common adaptation and probably the best one available.
Channel abandonment. You stop using a search engine for a category, or stop trusting retail listings, or give up on a forum. The cost is real: the replacement is not better, it is just less tiring.
Generalized suspicion. Nothing is checked because everything is suspect. This is the endpoint, and it is not skepticism. It is the absence of a method, and it hands the advantage to whoever is most fluent in producing plausible material.
The last one is the outcome worth watching for, because it feels like wisdom from the inside. A reader who trusts nothing has stopped doing the work of evaluating, which is the same failure as a reader who trusts everything.
Who feels it first
Professionals whose job already includes checking: editors, teachers, recruiters, researchers, lawyers, translators, illustrators, and anyone who reviews incoming work.
For these people the fatigue is not an occasional irritation. It is a change in job content. The volume of material requiring scrutiny has risen sharply while the time allocated to scrutiny has not, and the failure mode is not that they get fooled. It is that they stop reading closely, because close reading at this volume is not sustainable.
Teaching is the clearest case, and it is where the consequences are least abstract. A teacher assessing written work faces a judgment that cannot be reliably made and cannot be abandoned, with a student's outcome attached to getting it right.
What actually helps
Nothing on this list solves the problem. All of it reduces the cost.
Prefer accountability over authorship. Ask who stands behind a page, not what wrote it. A named editor, a corrections process and a stated source are durable signals. Authorship is not, because it cannot be verified. Use specificity as a filter. Content full of confident adjectives and empty of names, dates and numbers can be discarded in seconds. That single test removes most of what wastes your time. Narrow deliberately. Curated sources are a form of insurance. A small trusted set works because those sources can be held to it when they are wrong, rather than because they are always right. Accept the residual uncertainty. You will occasionally rely on something generated. The goal is not a perfect filter, which does not exist. It is to spend your checking budget where being wrong would cost you the most.
The part nobody has solved
Three honest caveats, because the topic invites confident prescriptions that do not hold up.
There is no reliable detector, so any advice that amounts to checking authorship is advice you cannot follow. There is no measurement of how much harm the aggregate does, only documented mechanisms and a workplace figure that happens to be unusually easy to count. And disengagement is a rational individual response that is a bad collective outcome: if the people who care most about quality stop reading widely, the remaining audience is less equipped to notice.
That last point is the reason this site bothers to cite anything. A page that names its sources costs the reader one glance and settles the question of whether to keep reading. It is a small thing, and multiplied across enough pages it is the only mechanism anyone has proposed that does not depend on a detector being invented or a platform changing its incentives.