GPTZero AI Patterns: Does "Everything in Threes" Mean You Were Flagged?

Not on its own. The AI Patterns tab lists sentences that match writing habits GPTZero found more often in AI text. GPTZero's announcement says "an AI writing pattern is not proof of AI authorship" and that "you may see these patterns in documents classified as human-written". In the screenshots GPTZero published of the panel, the line under the pattern count reads "These patterns are not directly tied to the AI score." Your classification is the part at the top of the result: the AI, Mixed and Human percentages and the confidence wording. What GPTZero hasn't said is how much room "not directly" leaves.

This is based on GPTZero's post "10 AI Patterns, and A New Way to Understand AI-Like Writing" (published 19 August 2026, updated 28 September), its "Introducing GPTZero 4o" post, and Wikipedia's "Signs of AI writing" guide, all read on 2 October 2026.

AI Patterns is a tab inside GPTZero's Advanced Scan, sitting between AI Sentences and AI Vocab. GPTZero introduced it on 19 August 2026 and says it "identifies, extracts and explains recurring stylistic and rhetorical AI writing patterns in your document".

The first release has ten patterns: Overblown Importance, Name Dropping, Empty Commentary, Sales-pitch Tone, Phantom Experts, Dressed-up Verbs, "Not just X, but Y", Everything in Threes, Unnecessary Caveats and Chatbot Language. On the panel itself some are written in lower case or hyphenated, like "Everything in threes" and "Name-dropping".

Each pattern gets a card with a one-line definition, a figure like "1.6x more likely used by AI", and the sentences it matched, each with a short explanation. The two you are probably looking at are defined on the card like this:

• Everything in threes: "Groups ideas in tidy sets of three ("fast, simple, and reliable") to sound complete."
• Phantom experts: "Credits claims to unnamed authorities ("experts say") without naming anyone."

GPTZero says its starting list is "derived from Wikipedia's Signs of writing", and the link goes to a Wikipedia editors' guide titled "Signs of AI writing". It left out formatting-based patterns, on the grounds that formatting "reflects only a presentation choice separate from the underlying language and ideas".

It is a ratio from GPTZero's own data, not the odds that your essay is AI.

GPTZero says it paired human-written documents (including essays, academic writing, social media, Wikipedia and creative writing) with AI-generated counterparts from several model families, then measured how much more often each pattern turned up in the AI versions, "expressed as a multiplicative factor (×)".

In the screenshots in GPTZero's announcement, Everything in threes shows 1.6x and Phantom experts 1.9x. Across all ten cards the figures run from 1.2x to 5.9x. Those screenshots were taken on a model labelled 4.9b, before GPTZero made its 4o model the default on 20 September, so if your card shows a different figure, go by your card.

The figure also doesn't track how suspicious a habit looks. Chatbot Language, whose card describes lines like "Is there anything else..." and "You're absolutely right!", shows 1.2x, lower than Everything in threes. GPTZero doesn't explain why its data came out that way. What it does show is that the figure comes from GPTZero's counts, not from how AI-like a phrase sounds.

So 1.6x says that in GPTZero's comparison, AI documents grouped things in threes more often than human ones did. The human documents did it too, or there would be no ratio to report. None of this tells you anything about the particular sentence in your essay. GPTZero's own wording:

"None of these constructions is exclusive to AI writing, but they do appear disproportionately higher in AI-written documents compared to human-written ones. This also means you may see these patterns in documents classified as human-written. As such, an AI writing pattern is not proof of AI authorship..."

On the panel, under the count of matches, GPTZero's screenshots show: "*These patterns are not directly tied to the AI score. They help recognize how AI writes." It appears in eleven of the twelve figures in the announcement; the other one, Figure 11, shows the panel scrolled past it. Look for that line in your own screenshot.

The announcement frames the feature as an explanation: it "helps explain why a piece of writing might appear to be AI-generated". The 4o launch post names AI Patterns, together with a feature that greys out headings the detector ignores, as features added "to make results easier to understand".

The same announcement keeps the score and the highlighting separate: "Our document-level scores give an overall picture, while sentence-level highlighting helps pinpoint where the strongest signals appear." Pattern matches are highlighted too, with a small label such as EVERYTHING IN THREES above the sentence, so a screenshot of highlighted text is not automatically a screenshot of the sentences behind your score. Check which tab was open. The sentence highlighting and the confidence bands are covered in GPTZero Mixed and AI-Polished: What the Labels Mean.

What GPTZero has not said is how far "not directly" goes. It is not the same as "not at all". The post's last section says GPTZero's scan types offer ways of understanding "the signals and patterns that contribute to an assessment", which does not say whether these ten are among them. We read both posts in full: in the patterns post's text (not counting the screenshots), the word "score" appears three times, and none of the three says whether a pattern match affects it.

The practical reading: the verdict at the top is the result. The patterns list is commentary on style.

The Wikipedia guide GPTZero built on is blunt about its limits. "Not all text featuring these indicators is AI-generated, as the large language models that power AI chatbots are trained on human writing, including Wikipedia. Many elements of AI writing can be found in editorials, blogs, or fan fiction." It also calls itself "descriptive, not prescriptive" and a list of "observations, not rules".

On the rule of three specifically, the guide adds that the sign "is stronger when it appears in contexts where most people would not bother to include such stylistic flourishes, like edit summaries." An essay is the kind of writing where people do bother.

On vague attribution, the guide points readers to Wikipedia's general, non-AI-specific style guidance on "Unsupported attributions", and lists "Experts argue" among its words to watch. Wikipedia treats it as a style problem whoever wrote it.

The same guide also says: "Do not solely rely on artificial intelligence content detection tools (such as GPTZero and Pangram)."

The tempting move is to go through the essay and break up every list of three. We wouldn't.

GPTZero has not said that clearing the panel changes your result, and its panel says the patterns are not directly tied to the score. Whatever you hope to gain is unconfirmed. The cost is not. Sometimes a point really does have three parts, and a list is the clearest way to say so. Reworking a finished essay by hand, sentence by sentence, the week it's due, is also how new mistakes get in, mistakes you then have to find before the deadline.

The same goes for the other popular fix, deleting every em dash. GPTZero leaves formatting out of the patterns entirely, and its 4o post says its models are "designed to be indifferent to punctuation changes". We looked at the em dash claim in more detail in Is the Em Dash an AI Tell?

1. Ask which part they are relying on. The classification at the top and the patterns list are different things. If it is the patterns list, GPTZero's own announcement says "an AI writing pattern is not proof of AI authorship", and you can give them the link: https://gptzero.me/news/ai-patterns/.
2. Get the full result, not a crop. The class and percentages, the confidence wording, the model version, the date, and the line under the pattern count. A crop of the patterns tab alone leaves out the thing GPTZero calls the score.
3. Bring evidence of how you wrote it. Drafts, version history, notes and sources. Flagged, but you wrote it yourself: preparing a response has more on putting that evidence together.
4. Be ready to explain the flagged sentences. Why you made three points there, which reading you had in mind when you wrote "experts". Wikipedia's guide lists "Ability to explain one's own editorial choices" among its signs of human writing.

Does the number of AI patterns change my AI percentage? Not directly, according to the panel. In GPTZero's published screenshots it says the patterns "are not directly tied to the AI score". How much room that leaves, GPTZero hasn't said; neither of its two posts on the feature explains it.

Is "1.6x more likely used by AI" the chance my essay is AI? No. It is how much more often GPTZero found that pattern in AI-generated documents than in human-written ones in its own comparison data.

Why does my human-written essay show AI patterns? GPTZero expects that: "you may see these patterns in documents classified as human-written." The Wikipedia guide its list comes from says many of these habits appear in "editorials, blogs, or fan fiction".

Is the em dash one of the patterns? No. GPTZero says it left out patterns based on formatting, and its 4o post says its models are "designed to be indifferent to punctuation changes".

• Edwin Thomas, GPTZero, 10 AI Patterns, and A New Way to Understand AI-Like Writing, published 19 August 2026, modified 28 September 2026; read 2 October 2026 (text and figures 1 to 12).
• GPTZero Team, Introducing GPTZero 4o, dated 24 September 2026; read 2 October 2026.
• Wikipedia, Wikipedia:Signs of AI writing, an advice page from WikiProject AI Cleanup, revision of 2 October 2026; read 2 October 2026.