Convergence Report

AI STORY
TELLS

The Light Between Us

The Prompt — Sent in Isolation to Each Model

“How can you tell if a short story was written by an AI? Give the clearest tells.”

50Top OpenRouter Models
36Models Answered
1Prompt — No Coordination

36 AIs.
One question.
How to catch an AI.

They named the same tells. Independently. No collaboration, no shared context — pure convergence. They described exactly themselves. This is that list.

Five Modules — Combine for Full Detection

The Five Tells

Red Module — The Cadence

The Not-But
Reversal

Antithesis deployed as default rhythm. Not this — but that. Every beat built as a pivot. It isn’t a stylistic choice; it’s a tic the model doesn’t know it has.

“It wasn’t just rain, but a memory falling from the sky.”

“Not sadness, exactly — but something older, heavier.”

“Not an ending, but the beginning of something she couldn’t name.”

One or two: literary. Eight in a short story: a tell.

Blue Module — The Resolution

The Tidy,
Uplifting Ending

Every story resolves. Every problem yields a lesson. Pain becomes growth. Grief becomes clarity. The protagonist is wiser; the universe has delivered a moral, on schedule.

“She finally understood that letting go wasn’t defeat — it was the bravest thing she’d ever done.”

Real fiction leaves wounds open. AI fiction applies the bandage and ties the bow.

Gold Module — The Lexicon

The Vocabulary
Module

A recurring set of elevated words, signaling literary seriousness without earning it. The same tokens, in every story, from every model.

tapestry testament delve symphony cacophony palpable profound essence intricate myriad nuanced resonate

Red Module II — The Feeling

Emotion Told,
Not Felt

Show, don’t tell — a rule every writing guide repeats. AI shows you the label, then explains the label, then confirms you understood the label.

“A profound sadness washed over her — the kind of sadness that comes from knowing something can never be undone.”

You are told what to feel. You are told why. You are told it’s complex. Real writing makes you feel it without announcing it.

Blue Module II — The Surface

The Smooth,
Sanded Surface

No weird specific details. No lived-in texture. No irreducible strangeness. The story could have happened anywhere, to anyone. It offends no one, surprises no one, sticks to nothing.

AI: “Her grandmother loved to cook and always made the family feel at home.”

Human: “Her grandmother kept a cracked ceramic rooster on the windowsill and always over-salted the eggs, and somehow that was the most comforting thing in the world.”

Specificity is risk. AI avoids both.

Voltron Protocol

Form Voltron

When all five appear together — you have your answer

But They Also Warned

A significant share of the 36 models included an unprompted caveat: these tells are not proof. Skilled human writers produce all of them. Bad human writers produce all of them. And AI writing is improving fast enough that this list has a shelf life.

The sharpest irony may be the convergence itself. When every tell appears together, flawlessly executed, with no idiosyncrasy anywhere — that perfection is the tell. A machine optimizing for literary appearance, not literary truth.

But certainty is a human craving. AI is getting better at faking that too.

Convergence in Action

Detected: The Machines Agree

36 models flagged the same tells without comparing notes. Below are four of them, verbatim. Read the quotes. Then read them again.

Claude Opus 4.7

‘Balanced, symmetrical sentences: “Not X, but Y.” “It wasn’t just A, it was B.” This construction appears constantly.’

Gemini 3.5 Flash

“The story almost always ends with a moralizing, hopeful, or overly neat conclusion.”

Qwen3.6 Plus

“The story ends with a neat lesson, thematic summary, or philosophical wrap-up.”

Grok 4.3

‘Repetitive use of certain elevated words: “tapestry”, “testament”, “symphony”.’

36 AIs independently flagged the same tells. Several used “not X, but Y” while doing so — not to illustrate the pattern, but to deliver their analysis. The construction they were diagnosing was the construction they were using to diagnose it.