---
schema: 1
type: term
title: _AI tells_
description: >-
  _AI tells_ are recognizable, repeatable signatures in AI-generated output —
  phrasing habits, formatting quirks, or stylistic tics that make text feel
  machine-produced. In this channel's hands-on model evaluations, spotting tells
  is one of the informal checks used to judge how a new model's writing actually
  reads.
canonical: 'https://liquidgold.cc/channel/everyinc/explore/term/ai-tells-1c71cab23aa0'
markdown: 'https://liquidgold.cc/channel/everyinc/explore/term/ai-tells-1c71cab23aa0.md'
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \_AI tells\_

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/ai-tells-1c71cab23aa0

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

_AI tells_ are recognizable, repeatable signatures in AI-generated output — phrasing habits, formatting quirks, or stylistic tics that make text feel machine-produced. In this channel's hands-on model evaluations, spotting tells is one of the informal checks used to judge how a new model's writing actually reads.

In this channel's material, **_AI tells_** shows up in the context of hands-on model testing — the vibe checks the Every team runs whenever a new release lands. A *tell* is a small, repeatable signature in a model's output: a phrasing habit, a formatting quirk, or a stylistic tic that makes text feel machine-produced rather than human-written. When the panel judges a new model, they read its outputs with an eye for exactly these signatures, because tells are one of the fastest informal signals of where a model's writing quality actually stands — the kind of signal a spec sheet or leaderboard can't give you.Across the team's vibe-check sessions — day-zero tests of releases like **Opus 4.8** and panel evaluations of models like **GPT 6.1 Soul** — judging outputs for tells sits alongside the more structured checks: coding tasks, writing samples, and knowledge work. The purpose isn't to catch a model cheating; it's to *understand* how it writes and whether that writing holds up under real use. A model that produces polished prose laced with noticeable tells reads differently from one whose output feels natural, and that difference feeds directly into the team's verdicts and comparisons.- An **AI tell** is a repeatable, machine-sounding signature in a model's output — a phrase, format, or stylistic tic.- The Every team looks for tells during hands-on **vibe checks** of newly released models.- Tells are an informal but fast quality signal, complementing structured tests like coding and knowledge-work benchmarks.