FIELD GUIDE · SLOP, SEARCH & THE OPEN WEB

IS AI MAKING THE INTERNET WORSE?

AI can make excellent things. It can also make ten thousand pages before lunch. The internet's problem is not that machines can write — it is that garbage became almost free.

UPDATED 2026-09-21

THE PROBLEM

THE WEB IS EATING ITS OWN HOMEWORK

Write. Copy. Train. Repeat. What could go wrong?

The web has always contained spam, scraped pages, content farms, fake reviews, and recycled junk. Generative AI changes the economics because producing plausible text, images, and variations can cost almost nothing. That does not make AI content bad by definition. It makes filtering and incentives much more important.

01

MAKE

ONE CLICKProduction cost falls through the floor.
02

PUBLISH

TEN THOUSANDScale becomes the default temptation.
03

CRAWL

EAT IT BACKSearch and model datasets meet the output.
04

FIND

GOOD LUCKAttention and provenance become scarcer.

THE FIGHT

IS THE MACHINE THE PROBLEM OR JUST THE CHEAPEST PRINTER EVER?

This is about who can still find something worth reading when production becomes effectively unlimited. Search engines, publishers, model trainers, creators, and readers all benefit from cheap generation in some contexts and suffer when quantity overwhelms signals of quality.

THE ABUNDANCE CASE

AI CAN LOWER THE COST OF USEFUL PUBLISHING TOO

Generative tools can help research, structure, translate, summarize, code, illustrate, and make niche information economically possible. Google's own guidance does not ban AI-assisted publishing; it focuses on whether pages add value.

A cheap printing press can print a great newspaper or 40,000 menus.

THE POLLUTION CASE

WHEN GARBAGE IS FREE, DISCOVERY GETS EXPENSIVE

Search systems explicitly target scaled low-value content, and model-training research shows that indiscriminate reliance on recursively generated data can erase rare information and degrade distributions.

If everybody can dump dirt into the river for free, clean water becomes the product.

THE WEIRD SHIT

THE FEEDBACK LOOP GETS GROSS

GOOGLE TREATS LOW-VALUE SCALE AS THE PROBLEM, NOT AI ITSELF

Google's spam policy defines scaled content abuse as mass content created primarily to manipulate rankings and says the technique can include generative AI, scraping, or other methods.

The policy targets purpose and value rather than a simple AI/not-AI label.

GOOGLE'S AI PUBLISHING GUIDANCE SAYS VALUE STILL COUNTS

Google says generative AI can help research and structure original content, while generating many pages without adding user value may violate scaled-content policies.

Useful AI-assisted publishing and AI slop are not the same category.

RECURSIVE SYNTHETIC TRAINING CAN LOSE THE TAILS

A 2024 Nature paper found that indiscriminately training successive generative models on model-generated data can cause model collapse, with rare parts of the original distribution disappearing first.

Human-origin and carefully curated data become more valuable when synthetic material proliferates.

THE PEOPLE WITH A STAKE

WHO STILL GETS FOUND?

READERS

They spend more attention separating useful material from plausible filler.

HUMAN CREATORS

Original work competes with near-zero-cost derivatives, summaries, and copies while also becoming more valuable as scarce training and reference material.

SEARCH ENGINES

They must identify scaled abuse without punishing genuinely useful AI-assisted work.

AI LABS

They need high-quality data and methods that avoid blindly feeding generated material back into future systems.

THE UNANSWERED QUESTION

WHAT BECOMES SCARCE WHEN CONTENT IS INFINITE?

TAKE THIS TO DINNER: The machine did not invent spam. It demolished the cost of making more of it.

RECEIPTS

The internet had garbage before AI. Now the garbage has a factory.

  1. Spam Policies for Google Web SearchGoogle Search Central
  2. Google Search's guidance on using generative AI content on your websiteGoogle Search Central
  3. AI models collapse when trained on recursively generated dataNature