Answer A
Text has shape · a guessing game
Guess the engine
Same prompt, two answers from two different models. Which one is Claude? Which is GPT, which is Gemini? Guess from the writing. Then see what a ruler would have guessed: a deterministic instrument that only counts words, sentences, commas and paragraphs, and never reads for meaning.
YouRead both answers. Pick the one the question asks for.
The lensMeasures eight structural axes plus length, and guesses from how each engine usually writes. It never sees the round you're playing.
The catchClaude is easy to spot. GPT vs Gemini is not, for you or for the lens.
Round 1 of 10you 0 · lens 0
Prompt
Answer B
Ten rounds
Where the answers come from. Claude, GPT and Gemini each answered the same 400 everyday, science and math prompts under a batch constraint (the EMS study, 1,200 answers). Every answer is shown exactly as returned, including the ones that were cut off.
What the lens is. The open-source Linguistic Kernel (v19) reduces a text to eight coordinates: weight shift, void, packing, cohesion, sentence length, rare-word pull, flow, clause thickness. Here it guesses by comparing each answer with each engine's average profile, built from the other 399 prompts. Across all 1,200 pairings it's right 82% of the time; structure alone gets 67%, length alone 65%.
It isn't built to identify models. It measures structure, and under these prompts the engines happen to write differently enough to tell apart. Different prompts or newer model versions may land differently.