Turnitin AI Detection Update (Aug 2026): Why the Purple Highlight Is Gone

On 4 August 2026 Turnitin replaced the two-color AI writing report for English submissions with a single set of blue highlights. The old split — blue for likely AI-generated text, purple for AI-generated text that was then AI-paraphrased — is gone. The blue now covers likely AI-generated text whether it came straight from a chatbot or was run through a paraphraser or bypasser, so the report no longer shows which of those Turnitin thinks it found. Students still never see the AI report. Instructors do. Old submissions are not rescored. The rest of this piece is a dated timeline from February through August 2026, plus a correction: bypasser detection launched on 27 August 2025, English-only, not in 2026.

What actually changed on 4 August

Until 4 August 2026, an instructor opening an AI writing report for an English submission saw two colors. Blue (sometimes described as cyan in older help articles) marked text Turnitin classified as likely AI-generated. Purple marked text it classified as AI-generated and then AI-paraphrased. Two colors implied two claims: "this looks generated" versus "this looks generated and then rewritten."

Turnitin's own write-up of the redesign is How Turnitin is simplifying AI detection for educators and publishers, and its product-updates note for 4 August ("Simplifying the AI report") adds that the change applies to English submissions only. Instead of flagging AI-generated and AI-modified content separately, Turnitin now shows one set of blue highlights covering:

• AI-generated text
• AI-generated text that was further paraphrased by an AI paraphraser
• AI-generated text that was run through a bypasser (humanizer) tool

Turnitin's stated aim is to let instructors focus on whether likely AI-generated content is present, rather than parse which tools were used. Neither the announcement nor the product-update note says anything about how the AI percentage itself is presented, so we won't guess. The interpretation cost is in the highlights: there is no longer a legend that separates "generated" from "generated and then modified." If an instructor treats a highlighted passage as "the student pasted this from ChatGPT," they may be reading a combined signal as if it had a single cause.

The highlight is now a single blue. Screenshots from 2024–July 2026 are out of date.

What did not change on 4 August: the 300-word English minimum, prose-only scoring (lists, tables, and code are excluded), the *% display for 1–19%, instructor-only visibility of the AI score, and the rule that old submissions are not retroactively rescored. Those constraints are still in force.

If you need the other half of the Turnitin screen — the text-matching percentage — that is a different report. We unpacked the two numbers in Turnitin Similarity Score vs AI Score.

Timeline: February through August 2026

The July "single model" note — and the missing day

Turnitin's own AI writing detection FAQ now says: "In July 2026, we updated our model architecture to consolidate a multi-model ensemble into a single model." The same FAQ still says Turnitin aims to keep its false-positive rate under 1% for documents with over 20% AI writing. We have not found an official Turnitin post or help article that pins the change to a calendar day. Treat the month as real and the day as unpublished.

"Single model" means the English detector (or Spanish, Japanese, Arabic) is no longer described as an ensemble of several models in that language glued together. It does not mean one model covers every language. Arabic did not exist as a Turnitin AI product until 18 August. A consolidation can change recall without a UI change: a June 2026 submission was scored by whatever English stack existed then; a new upload in September can differ. The archive copy is not silently rewritten. Independent testing of detectors as a class is less tidy than the under-1% claim; see Are AI Detectors Accurate in 2026? and the false-positive crisis.

Bypasser detection is not an August 2026 feature

A large number of ranking pages still date Turnitin's humanizer / bypasser detection to August 2026. That is wrong.

Turnitin launched bypasser detection on 27 August 2025 (press release). It is English-only. It was designed to flag text that had been processed through paraphrasers and AI humanizers, on top of the older "this looks generated" signal. We covered the feature, including how the then-separate layers interacted, in Turnitin's 2026 Update: AI Bypasser Detection Feature Explained. The "2026" in that slug is the year we published the explainer, not the launch year of the feature.

What did happen in August 2026 is the merge into one highlight. Before 4 August, an instructor could at least see purple as a distinct "this looks AI-paraphrased" claim. After 4 August, AI-modified text sits under the same blue highlight as raw generation. The bypasser capability did not launch in 2026; in 2026 the report stopped showing AI-modified text in its own color.

Two practical consequences:

1. English papers are still the only ones where the bypasser layer is in play. An Arabic submission scored after 18 August 2026 can receive an Arabic-model AI percentage. It does not get the English bypasser detector. Translating Arabic to English, running a humanizer, and translating back is a different risk — we walk through that pipeline in the Arabic detection piece, without inventing Arabic bypass rates we have not measured.
2. Older write-ups that describe two categories or two colors are describing the pre-4-August report. They are not describing what an instructor sees today.

Full disclosure: we build SupWriter, which is an AI humanizer. We are not going to pretend the bypasser layer is irrelevant to us. We have not published live-detector bypass tests — not against Turnitin's redesigned report, not for English, and not for Arabic — and we are not going to make up a number (how we test). No humanizer can guarantee how Turnitin will score a draft.

Students versus instructors

Students never see the AI writing percentage. The AI report is instructor-only. An instructor can export it as a PDF. There is no student-facing toggle equivalent to "show similarity report to students." If your LMS shows a percentage after you submit, that is the text-matching report. Direct version: Can Students See Their Turnitin AI Score?. Admins can download AI Writing Reports from Manage Submissions as of 16 July 2026 — staff workflow, not a student view.

To preview before you submit, check somewhere that is not Turnitin. Our AI detector gives one AI score with sentence highlights; the named labels it shows are a guide mapped from that single score, not results from those third-party detectors. It is not Turnitin. Detectors disagree for structural reasons: why AI detectors give different results.

What still has not changed

Strip away the August redesign and the detector is still the same product with the same constraints:

• 300-word English minimum. Below that, Turnitin does not return an AI score. Short forum posts, captions, and many quiz answers never enter the model.
• Prose only. Bullet lists, tables, and code are excluded from AI scoring. A methods table that looks "AI-ish" to a professor is often not in the percentage at all.
• *% for 1–19%. Turnitin does not show 7% or 14% in that band. It shows an asterisk. That is a confidence display, not a policy. It is also not an "acceptable" range — Turnitin has said there is no official acceptable percentage. Institutions set policy. See What Percentage of AI Is Acceptable on Turnitin?.
• No retroactive rescoring. A submission from March 2026 keeps the report generated at submit time. The 4 August UI and the February English-model update do not crawl backward through the inbox and rewrite old scores. If an instructor opens a pre-redesign paper today, they may still see the archived two-color report for that file. New submissions get the new report.
• Instructor-only AI score, with PDF export.
• Separate language models. English is not Spanish is not Japanese is not Arabic.

The 12 February 2026 English update aimed to improve recall while maintaining low FPR. A file that "passed" in January can score differently on a new upload after February. Spanish on 5 May 2026 is a separate model update, without the English bypasser layer.

How to read the AI score now

The number is a model output over qualifying prose. It is not "how much of the assignment you used AI for." Formulaic literature reviews inflate it; heavy editing can deflate it. ESL writing is a documented false-positive problem — in a 2023 Stanford study, seven GPT detectors flagged human-written TOEFL essays as AI at an average false-positive rate of about 61% (Liang et al.); see AI detection and ESL students. After 4 August you cannot use color to argue "that paragraph is paraphrased, not generated." If a hearing needs that distinction, the screenshot will not supply it.

Turnitin still publishes no official acceptable percentage. Local rules vary; named policies are in What Percentage of AI Is Acceptable on Turnitin?. Some institutions have stopped using AI detection or are phasing Turnitin out. Curtin disabled Turnitin's AI detection from 1 January 2026. SUSS turned off Turnitin's AI writing score from 11 August 2026. NTU will deactivate its institutional AI detector by 31 December 2026. Southampton will not renew its Turnitin contract beyond 2026–27 after Turnitin proposed licence terms that would have let it train AI on student submissions (Turnitin has since postponed that change). List: universities dropping AI detection. Mechanics that predate the redesign: Can Turnitin detect ChatGPT in 2026?, GPTZero vs Turnitin.

If you use a humanizer on English coursework, the English-only bypasser layer sits under that one blue highlight. SupWriter's detector is a pre-check based on one scoring model, not Turnitin; the humanizer is the rewrite; the free humanizer is 300 lifetime words and no card. List prices: Basic $9.99/month (5,000 words), Pro $19.99 (15,000), Ultra $39.99 (30,000), annual roughly half. None of that is a Turnitin setting.

Frequently Asked Questions

Is the purple highlight gone for good?

On current reports for English submissions, yes. As of 4 August 2026 Turnitin uses a single blue highlight. The purple "AI-paraphrased" color is not part of the current UI. Archived submissions from before that date can still show the old two-color report, because Turnitin does not rescore old files.

Does the single blue highlight mean Turnitin thinks more of my paper is AI?

Not necessarily. The redesign merges two highlight categories into one; Turnitin's announcement doesn't say the detector itself became stricter. What changed is that the report no longer separates text flagged as generated from text flagged as generated-then-modified. A score can still move between uploads because the February English model update and the July consolidation (exact day unpublished) changed the model underneath. Compare like with like: same file, same week, same product.

Will Turnitin rescore papers I submitted in the spring?

No. Turnitin does not retroactively rescore old submissions when the model or the report UI changes. The copy in the assignment is frozen. If you upload the same document to a new assignment, that new upload can receive a new score.

Can I see the new AI report as a student?

No. The AI writing report remains instructor-only. Instructors can export a PDF. Admins can download reports from Manage Submissions as of 16 July 2026. Student-visible percentages, when they exist, are similarity scores. Details: Can students see their Turnitin AI score?.

Did Turnitin launch bypasser detection in August 2026?

No. Bypasser / humanizer detection launched on 27 August 2025 and is English-only. August 2026 is when Turnitin stopped showing AI-modified text in its own color and folded it into the single blue highlight.

Does the August update apply to Arabic submissions?

The 4 August redesign is the report UI, and Turnitin says it applies to English submissions only. Arabic AI detection launched later, on 18 August 2026, as a separate language model. Bypasser detection still does not apply to Arabic. See Turnitin's Arabic AI Detection.

---

Before you submit, run the text through SupWriter's AI detector as a pre-check — it is not Turnitin, Turnitin's AI percentage is instructor-only, and the new single highlight will not tell you whether it thinks the prose was generated, paraphrased, or humanized.