What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
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
Graphite identified 12,877 words, phrases and recurring word patterns that appeared at least twice as often in AI-generated articles as in its human comparison set. In a later update, it found that Claude Opus 5.5 used “this matters” 116 times as often as the comparison set. Those are corpus-level patterns—not proof that a particular sentence or article was written by AI.
What are the new AI-writing tells?
In a report dated September 16, 2026, Graphite counted 12,877 distinct “tells” across nine AI models. It defined a tell as a word, phrase or recurring word-pattern frame that met its frequency and prevalence thresholds and occurred at least twice as often in AI-generated text as in the human comparison. The total is specific to Graphite’s sample and method, not a universal inventory of AI language. Graphite’s original report describes the study.
Graphite’s October 2026 update focused on Claude Opus 5.5. It counted 2,548 tells for that model and found “this matters” at 116 times the human comparison rate; the pattern “why _ matters,” where a less-common word can fill the gap, appeared at 92 times that rate. The update says Opus 5.5 had four percent fewer total tells than Opus 5. The Opus 5.5 update reports these results.
“This matters” is a striking example, but the finding is about how often a phrase appeared across the tested texts. It does not mean that using the phrase identifies a writer or establishes that a specific passage was generated by AI.
#1 Best Overall
How did Graphite identify them?
Graphite began with 10,000 web articles collected from Common Crawl and published before ChatGPT launched on November 30, 2022. It summarized each article with GPT-4.1, then had nine models write an article from the corresponding summary. The original comparison used 9,984 matched topics, each with a human article and AI-generated article.
The models in the original study were GPT-4.1, GPT-5, GPT-5.6 Sol, GPT-6 Astra, Claude Opus 4, Claude Opus 4.6, Claude Opus 5, Gemini 2.5 Pro and Gemini 3.1 Pro. Graphite counted individual words (unigrams), two- and three-word phrases, and frames: word patterns that allow a gap of up to three less-common words. It normalized counts for text length and required features to occur in a minimum number of articles as well as pass its twice-the-human-rate threshold.
Graphite said it chose this interpretable approach to find recurring patterns, rather than to build the most accurate detector of AI authorship. That distinction matters: the output is a catalog of relative-frequency differences in a particular experiment, not a tool that classifies an individual passage.
Why do the tells vary between models?
AI-writing patterns are not one fixed checklist. In Graphite’s original report, 65% of tells were unique to one model family. The report counted between 2,355 and 3,746 tells per model; its combined total for GPT-6 Astra, Claude Opus 5 and Gemini 3.1 Pro included 7,043 tells. Those totals describe the study’s feature counts, not the number of reliable ways to detect any model in arbitrary writing.
Rank #3
![MixPad Free Multitrack Recording Studio and Music Mixing Software [Download]](https://m.media-amazon.com/images/I/71ltIxIuz1L._SL160_.jpg)
- Create a mix using audio, music and voice tracks and recordings.
- Customize your tracks with amazing effects and helpful editing tools.
- Use tools like the Beat Maker and Midi Creator.
- Work efficiently by using Bookmarks and tools like Effect Chain, which allow you to apply multiple effects at a time
- Use one of the many other NCH multimedia applications that are integrated with MixPad.
Patterns also change as models are updated. In the versions Graphite tested, total tell counts fell 29% for Claude and 32% for Gemini, while GPT’s rose 48%. The combined frequency of nine well-known features fell between the earliest and latest tested models by 41% to 86%, depending on model family. A lower tell count does not, by itself, mean a model’s overall writing has become more like human writing.
Graphite measured overall word-distribution similarity separately from its tell counts. It reported that Claude’s distribution became more similar to the human comparison over the tested versions, while GPT’s and Gemini’s diverged. For Opus 5.5, Graphite said the overall word distribution was closer to the human comparison than Opus 5’s, even though its total tell count was only four percent lower. Its separate measure of well-known tells was six percent lower than Opus 5 and 53% lower than Opus 4. The measures capture different things and should not be treated as interchangeable. Graphite chief AI officer Greg Druck summarized the trend to TechCrunch: “It turns out that Claude models are actually getting closer to the human word distribution over time,” while “for the GPT models, it’s getting further away.”
Rank #4
![Express Schedule Free Employee Scheduling Software [PC/Mac Download]](https://m.media-amazon.com/images/I/41yvuCFIVfS._SL160_.jpg)
- Simple shift planning via an easy drag & drop interface
- Add time-off, sick leave, break entries and holidays
- Email schedules directly to your employees
What the findings can—and cannot—tell you
Graphite’s report can show that selected words and constructions were more common in its AI-generated sample than in its human baseline. It cannot establish authorship from a matching phrase. Druck told TechCrunch: “It’s not like the tells are decreasing. They are managing to remove the most well-known tells, but other ones pop up. And every model version has its own.” That cautions against treating any list as a permanent or comprehensive signature.
Free tools Windows power users keep installed
One-click scans. No signup required.
Other research makes a similar distinction between linguistic traces and proof of AI use. A 2026 study by Botes, Teuber, Vitali, Dewaele and Colling examined 512,970 psychology abstracts across 975 journals and estimated that at least 16% of 2025 abstracts showed linguistic traces consistent with LLM editing. That is a counterfactual-modeling result, not a direct count of confirmed AI-assisted abstracts; the authors describe such markers as indirect rather than definitive evidence. The study concerns psychology abstracts and should not be read as a finding from Graphite or generalized to all writing.
Best Value
![WavePad Audio Editing Software - Professional Audio and Music Editor for Anyone [Download]](https://m.media-amazon.com/images/I/B1fcLEGCs6S.png._SL160_.png)
- Full-featured professional audio and music editor that lets you record and edit music, voice and other audio recordings
- Add effects like echo, amplification, noise reduction, normalize, equalizer, envelope, reverb, echo, reverse and more
- Supports all popular audio formats including, wav, mp3, vox, gsm, wma, real audio, au, aif, flac, ogg and more
- Sound editing functions include cut, copy, paste, delete, insert, silence, auto-trim and more
- Integrated VST plugin support gives professionals access to thousands of additional tools and effects
Authorship attribution is also distinct from spotting a surface pattern or deciding that a text seems machine-written. Penn State’s account of authorship-attribution research distinguishes determining whether text is machine- or human-written from identifying which generator produced it. Phrase frequency, overall word distributions, surface traits such as sentence-length variation, and attribution to a particular model answer different questions.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why the study’s scope matters
Graphite compared older web articles with newer model outputs generated through a fixed workflow: each AI article was prompted from a GPT-4.1 summary of its human counterpart. The human baseline predates ChatGPT, so differences can reflect publication era, source material, prompt design or remaining boilerplate—not only AI authorship. Different prompts could produce different patterns, and results from general web articles do not necessarily apply to every genre or writing task.
For readers assessing a particular passage, a phrase such as “this matters” is therefore a reason to examine context, not a verdict. Graphite’s measurements describe aggregate rates in the study’s corpora; they do not establish who wrote an individual text or whether a human used AI assistance.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchQuick 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.
