Turnitin removed the purple AI-paraphrased highlight

Turnitin's AI Writing Report no longer prints a purple highlight. On 4 August 2026 the two categories, blue and purple, were folded into a single blue one, so a current report shows one kind of AI highlight where it used to show two. The detection behind purple did not go with it: Turnitin says the model still looks for AI text that was later run through a paraphraser or bypasser, and that text is now counted inside the blue category. Reports generated before that date keep their purple, and an existing submission has to be resubmitted before it gets the new report.

We've updated the AI Writing Report to combine the previous two categories blue and purple into a single blue category. Now, 'AI-generated text that was AI-paraphrased' will be combined with the 'AI-generated only' detection category, so the report will show a single 'AI-generated' category. This means purple highlights previously used for AI-paraphrased text will no longer be shown.

The next paragraph of that release note draws the line that matters most here: "The model will continue to detect likely AI generated content that may have been further modified by AI paraphrasers or bypassers. We will now only show one set of highlights to identify likely AI-generated content." A change to what is drawn on the page, in other words, not a capability being withdrawn.

That cuts in two directions at once. If you are holding a report with purple in it, nothing you were told about that colour has been retracted. If you are holding a new one, the absence of purple is not a finding about your document. Everything quoted below was read from Turnitin's own pages on 18 August 2026, and each quotation names the page it came from, because several of those pages currently describe the report differently from each other.

Purple never meant what it was usually taken to mean, and the merge does not change that. The definition Turnitin published when the two categories arrived sits in a release note dated 2024 July 16, still on its AI writing detection model page:

AI-generated text that was AI-paraphrased - qualifying text that was likely AI-generated and then was likely revised using an AI-paraphrasing tool or AI word spinner, such as Quillbot.

Look at the verb chain. Likely generated, then likely revised. Two model inferences stacked on each other, both drawn from the finished sentences rather than from any record of how they were produced. Quillbot appears there as Turnitin's example of the category, not as a finding about anybody's file, and no version of the report has ever named a tool that was used on a submission.

So an old report is still readable, and the questions worth asking of a purple passage are the ones you would now ask of a blue one: how much qualifying text is involved, where in the document it falls, when the report was generated, and what the department's policy actually says about assistance. Worth knowing that the 2024 note is a dated entry in a release-note archive rather than a description of today's report. A search can drop you into the middle of that page, and nothing around the entry says it has been superseded.

Old reports also do not quietly refresh themselves. The 4 August entry says the update "will apply to newly submitted files" and that "to access the latest AI Writing Report on existing submissions, please resubmit the file to generate the updated report". A July screenshot with purple in it is not somebody being behind on the news.

This is the part with consequences. Merging the categories removed a colour, not a population. The paraphrased text is still counted, and Turnitin's published definitions of the percentage say so on three separate pages.

Neither side of a conversation gets to use the missing colour, then. A writer cannot say the report cleared them of paraphrasing, because the report never had a way of clearing anyone of it. A marker cannot say the blue shows it, because blue now covers both routes and distinguishes them nowhere.

The colour word itself is not settled across Turnitin's own pages, which is worth knowing before anyone argues about it. The report guide, whose Submission Breakdown section now describes a single category called AI-generated, says that text "is highlighted in a blue color in the submission breakdown bar and in the submission". The capabilities FAQ, answering "What do the different colors in the report mean?", says likely AI-generated text "is highlighted in cyan in the submission". One highlight, two words for it, both live on the same day. Copy the legend out of the report in front of you, note its date, and cite that rather than a colour name from a guide.

The product updates page carries the reasoning, and it is unusually blunt about how its own interface was being read.

As large language models and AI writing tools have evolved, the signals that separate text that may be purely AI-generated from AI-generated text that may have been further modified by AI bypassers or paraphrasers have become more difficult to interpret. Using a single highlight color is intended to reduce misinterpretation of tool-specific distinctions and help instructors focus on whether likely AI-generated content is present.

The same entry lists what the release is meant to achieve. One of the three items is reducing "the risk of unwarranted academic misconduct reviews by removing distinctions that could be read to imply more certainty about the specific tool or workflow used to produce a submission". Another is letting institutions work from the overall score "without requiring users to parse distinctions between the specific tools used".

Set that beside the 2024 note, which said the two categories "provide more granularity into what AI capabilities were used and where". The granularity was added, then taken back, and the published reason for taking it back is that readers were getting more out of it than it could carry. Whether the signals genuinely became harder to separate is Turnitin's claim about its own model and no one outside can test it. The shape of the decision is checkable, though, and it is a useful thing to be able to point at: the vendor built a distinction, then removed it because of how people were reading it.

Only new submissions get the new report. A file processed before 4 August 2026 that has not been resubmitted keeps the report it already has, purple included.

Resubmitting changes more than the colours, which is the trap in doing it casually. The report that comes back is produced by whichever model is current, and Turnitin's model release notes log updates on 2026 February 12 and 2026 May 5. Each of those entries says the release "will not retroactively update previously-generated AI writing reports" and that "existing submissions must be resubmitted" to see the new score. So if the percentage moves, the model era moved too, and that has to be ruled out before the difference gets credited to anything you did. The mechanics of resubmitting, including what happens to your earlier draft, are in Will my own earlier draft match my resubmission?.

Before comparing two reports, or quoting one at somebody, write down:

• The generation date of each report, and whether the file was resubmitted in between.
• The category names and colours printed in that report, copied from the report rather than from a guide.
• The submission language. The report guide says only the English detector includes AI paraphrasing and bypasser detection, and the 4 August entry adds that "this change applies to English submissions only", so a Spanish or Japanese report never carried the second category in the first place. Can Turnitin detect AI in non-English submissions? has the rest of that picture.
• Whether both reports are of the same file. One edited sentence can move where a highlight starts and stops, so a like-for-like comparison needs the same document, not a later export of it.

Not everyone crosses over at the same moment, and the entry says who does: the update "will be applied automatically for all customers who currently license Turnitin Originality and have AI writing detection capabilities enabled on their account", and it "will also be available to iThenticate 2.0 customers who currently license AI writing detection and have it enabled on their account". If a colleague's report still shows two colours, the ordinary explanations are the report's date and whether the file has been resubmitted.

Two things come before any writing. If an integrity process is open, keep the submitted file and the original report PDF untouched, because you may need to show what was actually in front of the marker. And if you are answering a flag on work you wrote yourself, the case is built from evidence rather than from prose, which Flagged, but you wrote it yourself: preparing a response goes through step by step.

Where a revision is permitted and the review is closed or was never in dispute, the merged report takes away one small habit: you can no longer sort the highlights by category before deciding where to work. There is one set now. The sorting has to come from you, from where the passages sit, what each one is doing in the argument, and what the report's date says about which model produced it.

That is where our own tool fits, and it is ours, so treat this paragraph accordingly. HumanPen takes the document together with the AI report and rewrites only the paragraphs the report highlights. A paragraph is the smallest unit it will touch, so a part-paragraph selection is expanded to the full paragraph and shown to you before anything runs, and you are billed for the words actually rewritten rather than for the file. Eligible results can continue lowering AI for free. What nobody selling one of these can tell you is what a detector will say afterwards, and we do not.

Purple is a fact about a report's date, not about a document. Before arguing over a highlight, check when the report was generated, whether the file has been resubmitted since, and what the legend inside that report actually says.

My report from June has purple in it. Is it invalid now? No. Turnitin says the update applies to newly submitted files, so a report generated earlier stays exactly as it was, and the definition of purple at the time it was printed still stands. The only way to see the current version is to resubmit the file, which is a separate decision with its own consequences for your submission history.

Does this mean Turnitin can no longer detect paraphrasers, bypassers or humanizers? No, and both pages that announce the change say the opposite in the same breath. The release note: "The model will continue to detect likely AI generated content that may have been further modified by AI paraphrasers or bypassers." The product updates entry: the capabilities "will continue to detect content likely generated by AI, including text that may have been further paraphrased or run through bypassing tools to evade detection".

Is the highlight blue or cyan? Both words appear on Turnitin's pages today: blue in the report guide and in the release note, cyan in the capabilities FAQ. They are describing the same single highlight. Quote the legend from the report you are actually holding.

Will my AI percentage drop now that the categories have merged? The published definitions of that percentage covered both categories before the change and still do after it, and the change is described as one to the highlighting rather than to the number. A resubmitted file can still come back with a different score, because the model may have been updated since the original report and those updates are not applied retroactively.

Does any of this affect Spanish or Japanese submissions? The 4 August entry says the change applies to English submissions only. The report guide says only the English AI detector includes AI paraphrasing and AI bypasser detection, and that the Spanish and Japanese detectors do not, so those reports never had a second category to merge.