AI Slop and the Trust Collapse: Why "More Content" Is Now a Losing Strategy
AI content volume has crashed average engagement and trust. Here's why publishing more is now negative-ROI, and what evidence-first content does instead.
By Robin Deane — Founder, RD
Publishing more AI-generated content is now negative-ROI for most teams: readers, search engines, and AI answer engines have all adapted to discount undifferentiated content, so volume without originality gets buried rather than found. The winning strategy is fewer pieces, each carrying a specific, checkable claim that nothing else says.
For two years, the dominant content advice was simple: publish more, publish faster, let AI handle the volume. That advice is now actively working against the teams still following it. The flood it created has changed how every downstream system — reader attention, search ranking, AI citation — treats content that looks like everything else.
What Is "AI Slop"?
Definition: AI slop is content generated with minimal editorial input that recombines existing information without adding a new fact, a new number, or a new point of view. It is technically correct, grammatically clean, and says nothing that a reader — or an AI system — hasn't already seen stated more clearly elsewhere.
The term describes a quality problem, not a tool problem. AI-assisted content produced with genuine expertise, original data, or a specific point of view is not slop. The difference is not which tool wrote the first draft — it's whether the finished piece contains anything a reader couldn't get from three other search results.
Why Has "Publish More" Stopped Working?
The volume strategy relied on three assumptions that have all quietly broken.
First, that more pages meant more chances to rank. That was true when the supply of content was scarce relative to search demand. It stopped being true once every competitor could produce the same volume at the same cost, flattening the advantage to zero.
Second, that readers would find a way to sort good from bad. They haven't — they've simply stopped trusting the category. Survey after survey shows declining trust in online content generally, not declining trust in AI content specifically. The flood degrades the whole channel, including the good pieces sitting inside it.
Third, that algorithmic ranking systems would keep treating each page independently. They no longer do. Search and AI answer systems now evaluate content in the context of the wider pattern a domain or author is producing — rewarding depth and consistency, and quietly demoting sites where volume has visibly outpaced editorial effort.
How Do AI Answer Engines Detect Low-Effort Content?
This is the mechanism most teams still publishing at volume don't understand, and it matters because AI answer engines are becoming a meaningful share of how people find information at all. See our breakdown of the citation selection logic behind AI search for the full picture — the short version is below.
| Signal | What It Measures | Why Volume Content Fails It |
|---|---|---|
| Originality | Does this page state a fact or figure not found on other pages covering the same topic? | Recombined content has nothing unique to surface — it's redundant with sources already indexed |
| Specificity | Are claims concrete and checkable, or vague and general? | Fast-produced content defaults to hedged, general language that isn't citable as a specific answer |
| Author signal | Is there a identifiable, credible source behind the claim? | Volume production rarely carries genuine named expertise through to the page |
| Consistency over time | Does this domain reliably produce accurate, cited-worthy material? | A pattern of thin pages drags down trust in the domain's stronger pages too |
| Structural clarity | Is the answer extractable as a standalone passage? | Content optimised for word count over clarity buries the answer inside padding |
The practical effect: two pages can say almost the same thing, and the one with a specific number, a named source, and a direct answer in the first two sentences gets cited. The other doesn't. Volume doesn't move that needle — specificity does.
What Does Evidence-First Content Actually Look Like?
The alternative to volume isn't "better writing" in the abstract — it's a different editorial standard, applied before publishing rather than after.
Not "AI is transforming marketing" — a number, a benchmark, a named mechanism. If the opening sentence could appear on a competitor's page unchanged, it hasn't done its job.
A data point, a framework, a counterintuitive finding — something that didn't exist in searchable form before this page. If a piece doesn't clear this bar, it's a candidate for merging into something that does, not for publishing anyway.
Fewer pieces, each covering a topic completely enough that a reader doesn't need to leave to get the full picture. This is the opposite of the "cover every keyword variant" strategy volume content optimised for.
A named author, a specific vantage point, evidence of having actually done the thing being described. This is what E-E-A-T (experience, expertise, authoritativeness, trust) signals are built to detect, and it can't be produced by recombination alone.
Does This Mean AI Shouldn't Write Content at All?
No — the tool was never the problem. AI is excellent at structuring an argument, tightening prose, generating variants of a headline, and handling the mechanical parts of production. What it can't do is originate the fact, the number, or the point of view that makes a piece worth citing. That part still requires someone who has actually done the work being written about.
The teams getting this right use AI to compress the time between "we have an original insight" and "it's published well" — not to manufacture the insight itself. The teams getting it wrong use AI to manufacture volume, and are now watching that volume get quietly buried.
What Should a Content Team Do Differently Starting Now?
The immediate, practical shift is an audit, not a new production process. Pull the last twelve months of published content and sort it by one question: does this page contain something a reader can't get from the top three competing results? Anything that fails gets merged, rewritten with an actual original angle, or retired. Publishing cadence should drop as a direct result — that's the point, not a side effect.
This connects directly to how RD approaches content and brand work: fewer pieces, each built around a claim specific enough to be checked, rather than a calendar built around volume.
- "AI slop" describes a quality failure — recombined content with no new fact or point of view — not a tool
- Publish-more strategies have gone negative-ROI as competitors matched the same volume at the same cost
- AI answer engines actively detect and demote low-originality content using signals like specificity and author credibility
- A pattern of thin pages on a domain drags down trust in that domain's stronger pages too
- Evidence-first content requires one genuinely new claim per piece, checked before publishing, not after
- AI is well suited to structuring and tightening content — it cannot originate the insight that makes a piece worth citing
- The right response to slop-era search is fewer, deeper pieces — not better-disguised volume
Frequently Asked Questions
What exactly is "AI slop"?
AI slop is content — written with or without AI assistance — that recombines existing information without adding a new fact, statistic, or point of view. It's typically fluent and factually accurate, which is what makes it hard to filter out manually; the problem is redundancy with material that already exists elsewhere, not incorrectness.
Why would AI-generated content hurt my SEO if it's factually correct?
Search and AI answer systems no longer evaluate pages purely on correctness — they weigh originality, specificity, and domain-level consistency. A factually correct page that says nothing new gets outranked by a page that adds one distinguishing data point, and a pattern of low-originality pages can suppress ranking for a domain's better content too.
How do I know if my content counts as "slop"?
Ask whether the opening claim could appear unchanged on a competitor's page covering the same topic. If yes, the piece likely doesn't clear the bar. A useful test: could an AI answer engine cite this specific page for a specific fact, or would it just as easily cite any of the other ten pages saying the same thing?
Does this mean I should stop using AI for content production?
No. AI is genuinely strong at structuring arguments, generating headline and format variants, and speeding up production once the original insight exists. The failure mode is using it to manufacture the insight itself — that still requires someone with direct expertise or original data.
How much should a content team publish if not "more"?
There's no universal number — the right cadence is whatever your team can sustain while ensuring every piece clears the originality bar. Most teams find this means roughly a third to half of their previous volume, redirected into deeper treatment of fewer topics.
What's the fastest way to fix an existing content library full of thin pages?
Audit against one question per page: does it contain something not available on the top three competing results? Merge or retire anything that fails, and rewrite the strongest candidates around one clear, specific, checkable claim before republishing.
Keep Reading
