$ cat Spintax Machine

Article spinning is the oldest scaled-content trick: swap words for synonyms, multiply the variations, and publish hundreds of "unique" pages. Paste a sentence, watch it explode into spintax, pull the lever for 50 spun versions, then run four real duplicate detectors against the output.

How spinning works

A spinner looks up words and phrases in a thesaurus and wraps each match in spintax: {best|top|leading}. Every group multiplies the number of possible outputs, so one sentence with a dozen groups can produce millions of combinations. Picking one option from each group yields a "new" version. Sellers of spun content counted each combination as a unique article.

This machine uses a small built-in thesaurus of common marketing words, so it only spins words it recognises. Commercial spinners use far larger dictionaries and phrase-level rewrites, but the structure of the output is the same.

What the four detectors measure

• Exact match. Hashes each version and looks for identical copies. Spinning is designed to beat this check, and it usually does.
• Word 3-gram overlap. Breaks each version into overlapping three-word sequences ("shingles") and measures how many two versions share (Jaccard similarity). This is the classic near-duplicate technique. A version is flagged when it shares at least 50% of its shingles with another version.
• Sentence skeleton. Keeps the function words (the, to, is, that) and blanks out everything else. Swapping synonyms leaves the skeleton untouched, so spun versions line up almost perfectly. Flagged at 80% shingle overlap.
• Synonym-normalised. Maps every word back to its synonym group before comparing, which is how a spinner-aware system would read the text. Flagged at 80% overlap.

The heatmap shows every pair of the 50 versions. Bright cells mean the pair is near-identical under the selected detector. Switch detectors to see how a weak check sees variety while a structural check sees the same text 50 times.

Limits and policy context

These are four transparent checks running in your browser, chosen to illustrate the idea. They are not a description of Google's systems, which are not public and are likely to use richer signals than word overlap. Passing all four checks here would not make spun pages safe.

Detection is only part of the picture. Google's spam policies treat generating many pages primarily to manipulate rankings, including by automated rewording, as scaled content abuse regardless of how the text was produced. For the policy history, enforcement cases and what distinguishes useful templated content from abuse, see the Scaled Content Abuse chapter and Detection & Defense.