The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
In Graphite’s October 1, 2026 update, Claude Opus 5.5 used “this matters” at 116 times the rate found in a comparison set of human articles. The phrase “why _ matters” appeared at 92 times the human rate. Those are corpus-level frequency ratios—not a way to prove that any one passage or writer used AI.
What Graphite means by an AI writing “tell”
Graphite’s analysis looks for words, phrases and reusable frames that occur unusually often in model-written articles compared with human-written articles. A frame is a pattern with a gap of up to three words, such as “why ___ matters.” Graphite counts a feature as a tell when it appears at least twice as often in AI writing as in the human comparison, after normalizing for text length and applying frequency filters.
The October 2026 Opus 5.5 update compares articles on 9,974 aligned topics: one human article and one article from each model for each topic. The human articles predate ChatGPT. The matching topics help hold subject matter relatively steady, but the baseline is not a representative sample of contemporary human writing.
Opus 5.5’s reported phrase and style patterns
The “this matters” result is the most striking, but it is one of several patterns Graphite identifies. The ratios below compare each feature’s frequency in Opus 5.5 with its frequency in the study’s human articles.
#1 Best Overall
| Pattern | Reported rate | What it illustrates |
|---|---|---|
| “this matters” | 116 times the human rate | Importance-signalling phrase |
| “why _ matters” | 92 times the human rate | Importance-signalling frame |
| “rather than simply” | 32 times the human rate | Contrastive phrasing |
| “what comes next” | 24 times the human rate | Transition |
| “dependable” | 23 times the human rate | Evaluative adjective |
These are relative frequencies, not statements that a phrase appears in every Opus 5.5 response or is absent from human writing. A large ratio can flag a recurring preference in this corpus without making the phrase exclusive to a model.
How Opus 5.5 compares with Opus 5
Graphite reports different results depending on what it measures. Its update finds 2,548 qualifying tells for Opus 5.5, compared with 2,666 for Opus 5. That count is not the same as overall word-distribution divergence or a stylistic rating.
| Measure | Opus 5.5 | Opus 5 | How to read it |
|---|---|---|---|
| Qualifying tells | 2,548 | 2,666 | Count of words, phrases and frames meeting Graphite’s criteria |
| Word-distribution divergence | 19% lower than Opus 5 | Reference point | Summary of how word-use distribution differs from the human comparison; not a tell count |
| Mannered-prose score | 10.57 | 16.75 | Rated on a 0–100 scale by Opus 5; this measure used a 1,000-topic subsample |
| Em dashes per 1,000 words | 0.015 | 2.92 | Graphite reports Opus 5.5 used them 99% less often than Opus 5 |
Graphite says the other update comparisons used 9,974 aligned topics; mannered prose was evaluated on a 1,000-topic subsample. The lower tell count, lower word-distribution divergence and lower mannered-prose score describe separate aspects of the comparison. They should not be collapsed into one score for how “human” a model sounds.
The em-dash result is a useful reminder that familiar clues can change sharply between versions. An editor who treats one punctuation habit as a permanent AI signature can miss those changes.
Rank #3
Why there is no universal checklist of AI phrases
Graphite’s original study covered 10,000 human articles and 90,000 AI-generated articles across nine models. It identified nearly 13,000 tells, with 65% unique to a single model family. Its examples varied across systems: corrective phrasing for GPT-6 Astra, superlative or contrast patterns for Claude Opus 5, and formal transitions for Gemini 3.1 Pro.
In an interview with TechCrunch, Graphite Chief AI Officer Gregory Druck said Claude models were “getting closer to the human word distribution over time,” while GPT models were “getting further away.” That is Druck’s characterization of Graphite’s comparisons, not an independently established rule about every model or future version. He also described a pattern in which familiar tells fade and other tendencies emerge: “every model version has its own.”
Rank #4
Can “this matters” tell you whether a passage was written by AI?
No. Graphite’s results show recurring differences across a corpus, not a validated authorship test for an individual passage. A human writer can use “this matters,” and a model-generated passage can avoid it. The study does not establish that any single phrase, punctuation mark or writing habit proves AI authorship.
For an editor, these patterns are more useful as prompts for close reading than as verdicts. Repeated importance-signalling or formulaic transitions may be worth revising when they make a particular piece sound generic, regardless of who wrote it. But authorship judgments require evidence beyond matching a phrase on a list.
What the findings can—and cannot—show
- They can show: which words, phrases and frames Graphite found disproportionately common in its matched model-and-human corpus.
- They can show: that the measured patterns differ by model family and can change between versions.
- They cannot show: that the human comparison represents writing produced today; Graphite’s human articles predate ChatGPT.
- They cannot show: that an individual writer or passage is AI-generated because it uses one of the reported expressions.
- They do not establish: that Graphite’s findings have been independently replicated. Graphite published the study, and TechCrunch describes the company as a growth marketing agency.
For the underlying figures and methods, see Graphite’s Opus 5.5 update, Graphite’s original study, and TechCrunch’s interview with Gregory Druck.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
