GPTZero 4o Gray Highlights: What Excluded Text Means and Whether It Counts as AI

Gray text in a GPTZero report marks content that was excluded from evaluation by the AI detector. It does not count as AI-generated. However, under GPTZero's published documentation, excluded text is not the same as text verified as human-written: it simply means the model masked those sections and did not consider them as features during analysis. The primary example documented by the vendor is section headers. Furthermore, this highlighting scheme is specific to GPTZero's own interface and model; GPTZero's documentation does not make statements about how other detection tools such as Turnitin handle headers.

When writers submit an essay, report, or article to GPTZero and notice sentences or titles highlighted in gray, the immediate confusion is understandable. Earlier versions of the detector primarily relied on yellow or orange highlights for text suspected of being AI-generated or mixed, and green for human-written prose. Both header masking and the third highlighting color—gray—were introduced on 1 August 2026 with Model 4.8b and detailed in GPTZero's 10 August blog post, before being carried forward into Model 4.10b (GPTZero 4o) in September and Model 4.11b (4.1o) in October. (Note: GPTZero also uses gray in its separate citation and source scanning tool to mark uncited claims, but in the AI detection interface, gray indicates excluded text.)

Below is exactly what GPTZero's published records say about this color, what it does and does not mean for your document, and why you should not confuse it with the scoring logic of other academic integrity systems.

Visible gray highlights and header masking were first introduced on 1 August 2026 alongside GPTZero's Model 4.8b release. In an official blog post dated 10 August 2026 (A New Dawn of More Meaningful AI Text Detection), the vendor announced:

"GPTZero's AI Detector now masks headers as of our August 1, 2026 model release 4.8b … We've also changed how we communicate these results to users by introducing a third highlighting color: gray. Gray spans in our dashboard or chrome extension indicate pieces of text which are ignored by our detector."

In that same August announcement, GPTZero defined what types of text qualify for header masking:

"Markdown indicators, common header phrases such as "Conclusion", and even short lines starting with an emoji are now considered to be headers, and are excluded from AI detection as indicated by the gray highlighting"

The post also clarified that intentional non-headers—such as list prefixes, code block descriptors, hashtags, or email greetings—are deliberately not treated as headers.

As detailed in our analysis of AI detector model updates in September 2026, vendor updates often roll out in stages across release notes and user interfaces. When GPTZero released Model 4.10b (GPTZero 4o) in September 2026, it carried forward this gray highlight layer and documented it across two channels:

1. Model and API Release Notes (Notion): Under the entry for `September 13, 2026 (2026-09-13-base, Model 4.10b aka 4o)`, the log notes that the update was `released on September 18, 2026`. (The release notes have since added an entry for `October 7, 2026 (2026-10-07-base, Model 4.11b aka 4.1o)`.)
2. GPTZero News Blog: In the announcement article Introducing GPTZero 4o, the vendor stated: > "As of 20 September 2026, the updated model is available to all users and is the default model offered on GPTZero." (Note: While originally published in late September 2026, the live blog page shows a revised date of October 18, 2026).

In the Introducing GPTZero 4o article, GPTZero reiterated its header masking and gray highlight rules. The post first explained header masking:

"we currently mask headers, meaning our detector does not consider headings as a feature, helping prevent changes to those headings from affecting the result"

and immediately added:

"We’ve also introduced a new highlighting color: gray. Gray spans show which pieces of text are excluded by our AI detector, which means that you can expect even more consistent results."

Rather than an unannounced change exclusive to 4o, gray highlighting is GPTZero's established visual indicator for masked text, active across dashboard scans and extension reports since August 2026.

When a scan returns gray spans, a frequent assumption among students and writers is that the gray sections received a "clean pass" or were confirmed as human writing.

That is a misunderstanding of how machine-learning masking works.

When a classifier marks text as "Human," it evaluates stylistic signals, statistical perplexity, and linguistic features across sentences, concluding with high confidence that the passage resembles human authorship.

Gray spans never undergo that evaluation. Masking means the detector removes the text before the model extracts linguistic features. The vendor's phrasing is exact: the detector "does not consider headings as a feature."

This distinction carries two important implications:

• Gray does not certify human authorship: If a heading was generated by a large language model or suggested by an AI assistant, GPTZero's gray mask still excludes it. The gray highlight is not an endorsement that the phrasing is human.
• Gray does not penalize human text: Conversely, if you spent hours perfecting a heading by hand, it is masked solely because of its structural role as a header, not because of suspicion or scrutiny.

Headings in academic papers, technical documentation, and essays frequently follow rigid, predictable conventions (such as "Literature Review," "Methodology," or "Data Analysis and Discussion"). In natural language processing, short, formulaic headings can distort stylistic models if evaluated alongside continuous narrative prose. By masking headers in gray, GPTZero prevents these formulaic labels from skewing the evaluation of the body text.

A natural follow-up question is whether having extensive gray-highlighted headings alters your overall AI percentage or changes how the rest of the paper is scored.

Here is what GPTZero has explicitly published about score impact:

• In its August 2026 announcement, the vendor stated that documents are "no longer subject to having different AI scores when headers are added or removed."
• In the 4o announcement, the vendor reiterated that masking headings helps "prevent changes to those headings from affecting the result," adding that users "can expect even more consistent results."
• Text elements recognized as headers (such as Markdown indicators, phrases like "Conclusion", or emoji-prefixed short lines) are masked and excluded from feature consideration.

Here is what remains unpublished in vendor documentation:

• GPTZero has not published the mathematical formula or calculation rules showing whether excluded words are subtracted from the denominator of total word counts when computing document-level aggregate probabilities.
• The vendor does not publish formulas showing whether masking a header alters the transition probabilities or confidence scores of adjacent body paragraphs.
• There is no documented threshold for how many gray spans can appear in a document before overall report metrics behave differently.

In short, the vendor has stated the intended effect—preventing header edits from fluctuating the document score—without disclosing the underlying mathematical aggregation formulas. You know with certainty that gray words are excluded from feature extraction, but you cannot infer undocumented mathematical adjustments beyond what the vendor has stated.

Students frequently see a gray highlight on GPTZero and wonder whether Turnitin or other institutional systems will display a similar gray marking on their official submission.

It is critical to distinguish between different detection platforms. Gray spans are a proprietary interface feature of GPTZero. They are not an industry standard, and GPTZero's documentation does not make statements about how other detection tools such as Turnitin handle headings.

Each vendor trains separate models, maintains separate parsing rules, and uses its own interface conventions. A text element excluded or highlighted in gray by GPTZero cannot be assumed to be treated the same way by another detector.

If you receive a GPTZero report containing gray highlights, here is how to read the output objectively:

1. Verify the structural role of the highlighted text: Check whether the gray spans correspond to section titles, Markdown headers, common phrases like "Conclusion", or emoji-prefixed short lines. This aligns with GPTZero's documented header masking rules. (Also ensure you are viewing the AI detector rather than GPTZero's citation tool, where gray indicates uncited claims.)
2. Focus attention on the analyzed body prose: The colored highlights and class confidence ratings (AI, Mixed, Human) reflect the unmasked body sentences. Those are the sections the detector actually evaluated.
3. Do not alter document formatting to manipulate masking: Attempting to disguise body sentences as headers, or removing legitimate headers to alter the detector's view, disrupts the academic structure of your work without any documented predictability.
4. Treat gray as unassessed, not validated: When reviewing your draft or discussing submission records, remember that gray highlights indicate excluded text. They do not constitute evidence of AI generation, nor do they serve as proof of human writing. If you need to keep records of a scan, save the original report with its timestamp and model version, noting the vendor's documentation on header masking. Whether you revise your draft or consult writing tools depends on your institution's specific academic policies.

• GPTZero News: A New Dawn of More Meaningful AI Text Detection (published August 10, 2026; announcing Model 4.8b header masking and gray highlights as of August 1, 2026): https://gptzero.me/news/meaningful-ai-text-detection
• GPTZero Model and API Release Notes (Notion page, entries for Model 4.8b released August 1, 2026; Model 4.10b / 4o released September 18, 2026; and Model 4.11b / 4.1o released October 7, 2026; retrieved October 2026): https://gptzero.notion.site/GPTZero-Release-Notes-Model-and-API-6f58686f6381498baef35212463b7da6
• GPTZero News: Introducing GPTZero 4o (originally published late September 2026; default model availability as of September 20, 2026; live page updated October 18, 2026): https://gptzero.me/news/introducing-gptzero-4o/