April 8, 2026
When bots share a voice
Show HN: We fingerprinted 178 AI models' writing styles and similarity clusters
Clone vibes, brand markups, and “AI slop” outrage
TLDR: A new write-up claims many AI models sound alike—even across different companies—and sometimes the cheaper one writes like the pricey one. The comments explode over whether style equals substance: skeptics cry “brand markup” while others insist writing vibe isn’t intelligence and question the study’s methods.
The devs behind a Show HN bombshell say they fingerprinted writing styles for 178 AI models and found “clone clusters,” cross-lab twins, and same-sounding bots with 185x price differences. The crowd? Split between popcorn and pitchforks. One camp is cackling at the “same writing, different bill” callout — paying for the label, not the language. The other camp is yelling, style isn’t smarts.
Critics came in hot. “You’re very wrong here,” snapped one writer, arguing that getting the idea right beats sounding similar. Another accused the post of “AI slop,” calling a claimed 99% style match between Anthropic and Google suspicious. Meanwhile, a curious faction wondered if the “AI voice” is inevitable — or a deliberate fingerprint to steer clear of model collapse (when AIs train on AI text and get mushy). Some devs loved the detective work, likening the 32 style dimensions to a DNA test for bots and hinting at distillation breadcrumbs (models trained from others). Others just roasted the piece’s formatting: “Subheadings were a major turn off.”
Bottom line: the findings say many bots write alike, even across rival brands. The community says: neat charts, but don’t confuse vibe with brainpower — and if these are clones, why are we paying champagne prices for soda pop
Key Points
- •The analysis fingerprinted 178 AI models across 32 writing-style dimensions using 3,095 responses per comparison.
- •Twelve “clone” pairs with over 90% writing-style similarity were identified, including cross-provider pairs.
- •Multiple pairs of models exceed 75% writing similarity despite large price gaps, with differences reportedly up to 185x per million tokens.
- •A “Distinctiveness” metric (intra-/inter-provider similarity) gauges whether a lab exhibits a recognizable house style (>1) or internal variability (<1).
- •Additional metrics assess family cohesion, which dimensions most differentiate writing styles, overall uniqueness vs. convergence, and consistency vs. variability across prompts.