AI slop is drowning the web

Two days ago I wrote about design taste being the only durable advantage for AI startups. That post was optimistic. This one is the counter-argument — because if taste is the moat, slop is the flood that erodes it.

Merriam-Webster made "AI slop" their 2025 Word of the Year. Their definition: "digital content of low quality that is produced usually in quantity by means of artificial intelligence."

A year later, the problem is not theoretical. It is the default state of the web.

What is AI slop, exactly?

AI slop is content that looks competent but says nothing. It has the structure of a good article — headings, transitions, bullet points — but no why. No tension. No point of view.

Researchers define three properties of slop: superficial competence (looks right, is hollow), asymmetric effort (cheaper to generate than to review), and mass producibility (one prompt, a hundred articles). The combination creates an information ecosystem where the garbage outpaces the signal.

Why AI slop is worse than bad content

Bad content made by a human at least has a perspective. It might be wrong, but it is someone's wrong take. That makes it useful — you can argue with it, learn from it, or improve on it.

Slop has no perspective. It averages. It hedges. It pleases everyone and convinces no one.

The cost is measurable — employees waste time decoding content that looks substantive but says nothing.

On the web, the scale is staggering. Graphite's longitudinal study found 52% of new web articles are now AI-generated, up from roughly 10% three years ago. AI-generated articles first surpassed human-written ones in November 2024 and have not looked back.

How to spot AI slop

Three telltale signs:

Generic examples. The slop article about productivity will mention "streamlining workflows." The slop article about design will mention "clean layouts." If every example could be swapped into any other article on the same topic, you are reading slop.

No tension. Slop never disagrees with the reader. It hedges every claim with "some might say" or "it depends." Real content takes positions. It makes people uncomfortable. It says something specific enough to be wrong.

Perfect structure, zero soul. Impeccable grammar. Logical flow. Nothing memorable.

When someone detects slop, two things happen fast: they lose trust in the creator, and they feel cheated for giving it their attention. That is not something you recover from with a better hook next time.

How to avoid making slop yourself

The fix is not to stop using AI. The fix is to stop treating AI output as finished work.

Use AI for first drafts. Use it to overcome the blank page. Use it to generate alternatives you would not have thought of. Then edit with the same rigor you would apply to a human writer's draft.

Three rules I follow:

1. Rewrite every introduction. The model's opener is always generic. Replace it with a specific experience, a surprising stat, or a direct opinion.

2. Kill the hedge words. Search for "some might argue," "it is worth noting," "in today's fast-paced" — delete them all. If the sentence is weaker without the hedge, rewrite the sentence.

3. Inject one thing only you know. A personal observation. A specific number from your own data. A direct quote from someone you talked to. If nothing in the article required a human to write it, it should not be published.

The line between using AI and making slop is the same line that separates building from assembling. One requires taste. The other just requires a prompt.