Home/Events/Graphite Study Reveals Tells in AI Writing for Opus 5.5 and Astra Models

Graphite Study Reveals Tells in AI Writing for Opus 5.5 and Astra Models

Confirmed

Confidence

80%

Impact: 70%

Updated Oct 1

Consensus Brief

A study by Graphite identified distinct writing habits or 'tells' in AI-generated prose, particularly in models like Opus 5.5 and Astra. The research highlighted that while some older tells have been reduced, new patterns persist, indicating ongoing challenges in achieving human-like writing consistency.

Sourced from

TechCrunchT2

Primary: TechCrunch

What Changed Since Last Update

Oct 1

The study revealed that Opus 5.5 now avoids certain constructions like 'it's not X, it's Y', while still frequently using phrases like 'this matters'.

Claim Ledger

5 claims tracked across sources

Independent Finding

Claude Opus 5.5’s biggest tell is the word 'dependable', which pops up 23 times more often than in human samples.

Independent Finding

Opus loves to tell you why things matter, using the phrase 'this matters' 116 times more often than human writing.

Independent Finding

OpenAI’s Astra has a different set of tip-offs, including the phrase 'another dimension'.

Independent Finding

Graphite found 13,000 phrases that were at least twice as common in the AI content as human content.

Independent Finding

Opus 5.5 used em-dashes 99% less often than previous models.

Role-Based Impact Analysis

Source Timeline

1 source corroborating

Opus 5.5 loves to tell you ‘this matters’ (and other AI writing tells)

TechCrunch·Oct 1