Why Turnitin Flags the Explanation of Your Own Diagram
A diagram, its caption, and the paragraph explaining it are three different pieces of a submission. Turnitin's published guidance describes how its AI percentage applies to qualifying prose, but it cannot reconstruct how you researched, drew, or explained the diagram. That distinction tells you what to check before changing anything.
HumanPen Team
· 7 min read
Why did Turnitin flag the explanation of my own diagram?
Turnitin's AI percentage applies to qualifying prose, not to every object in the submitted file. A diagram is not a prose sentence, but the paragraph explaining it may be. The report can identify prose it classified and show highlights; it cannot reconstruct your research process or prove who wrote the explanation.
If you were shown only a screenshot with `48%`, first confirm that it came from the AI Writing Report. A Similarity Score is a different measurement. Then ask for the complete report so you can see the passages behind the number. A percentage without its highlighted text leaves out the part you can actually review.
The diagram and its explanation are different text types
The word diagram can hide three separate items. Turnitin does not publish a diagram-specific rule, so the safest reading starts with the kind of text actually present in your file.
| Part of the submission | What it usually contains | What the published guidance supports |
|---|---|---|
| The diagram object | Shapes, arrows, lines, plotted values, icons, or an image | These are not prose sentences. Turnitin's page does not say whether every diagram object is included or excluded, so the absence of a highlight is an observation about one report, not a universal rule. |
| The caption | Anything from `Figure 2` to one or more complete sentences | Turnitin does not publish a blanket caption rule. A short label and a grammatical sentence are different structures. Check whether the caption is actually highlighted in the report. |
| The explanatory paragraph | Sentences that interpret a relationship, justify a choice, or connect the figure to the argument | Long-form prose sentences may be qualifying text. This paragraph can therefore appear in the AI Writing Report even when the visual object is not prose. |
Writing the diagram yourself does not create a technical exemption for the paragraph around it. The classifier receives text, not the history of your browser research, notes, calculations, or drawing file. That is why a classification can conflict with a truthful account of authorship without telling you what caused the classification.
Why 48% may not mean 48% of the whole submission
Turnitin defines the denominator around qualifying text, which it describes as prose sentences in a long-form writing format. Its guidance also says mixed writing types can create a gap between the percentage and the amount of highlighted material. The percentage is therefore not necessarily a percentage of the entire submission.
Imagine a report containing a diagram, a table of measurements, short labels, references, and several explanatory paragraphs. The total page area and the qualifying prose are not the same pool. A `48%` AI result does not mean that 48% of the diagram, 48% of the research sources, or 48% of the whole file was proved to come from AI.
Keep the two Turnitin measurements separate:
| Report | What its percentage refers to | What not to import from the other report |
|---|---|---|
| AI Writing Report | Text the model classified within the qualifying prose it evaluated | Do not use source-match exclusions or bibliography-match behavior to explain an AI classification. |
| Similarity Report | Submission text that matches material in Turnitin's comparison collection | A source match is not an AI classification, and a low Similarity Score does not cancel an AI result. |
If the screen really shows a current `48%` AI result, it is above the range where Turnitin replaces an exact score with `*%`. Ask for the corresponding highlights and page locations, not only the number.
What the AI Writing Report can and cannot establish
| The report can show | The report cannot establish by itself |
|---|---|
| An overall AI percentage for qualifying prose | That the same percentage applies to every object or word in the file |
| Which passages the interface highlights | Why one exact sentence received its classification |
| The page and highlighted text selected in the submission breakdown | Whether you drafted the sentence from your own research, notes, or analysis |
| A model output that an instructor can examine with other evidence | Who authored the work or whether academic misconduct occurred |
Turnitin says its model may misidentify human-written, AI-generated, and AI-paraphrased text. It also tells educators not to use the result as the sole basis for adverse action. Those warnings do not automatically clear a disputed paper, but they do define the proper next step: review the passages, the writing process, and the institution's policy together.
Build a passage-to-evidence map before editing
Do not answer a document-level number with a document-level denial. Map each highlighted passage to the work behind it. This gives an instructor something more useful than another detector score.
| Location in submitted file | What this passage does | Supporting material you can produce | Earliest dated record |
|---|---|---|---|
| Page __, paragraph __ | Explains __________ in Figure __ | Source notes, calculation sheet, diagram working file, outline, or version history | __________ |
| Fictional example: page 5, paragraph 2 | Explains why the arrow from survey response to theme coding is one-way | Dated interview-code spreadsheet, notebook entry defining the categories, and the editable diagram file | Version-history entry from 4 June, 14:18 |
The fictional row shows the level of detail to aim for. It is not evidence for anyone else's assignment. For your own map:
- Copy the exact location and function of every disputed passage from the report.
- Name the material that supports the ideas in that passage, not merely a list of websites visited.
- Record the earliest date that can be checked in notes, file history, comments, or the diagram source file.
- Leave a cell blank when no record exists. Do not manufacture a cleaner history after the fact.
A single timestamp does not prove authorship. A coherent sequence of sources, notes, calculations, drafts, and diagram revisions gives a reviewer a process to inspect.
Preserve these records before changing the prose
Save the materials that describe what was submitted and how it was produced:
- the exact submitted DOCX or PDF, kept unchanged;
- the complete AI Writing Report PDF, including its title, date, percentage, and highlights;
- the editable diagram source, such as the original presentation, drawing, or chart file;
- research notes, source annotations, outline versions, and document version history;
- calculations, data extracts, or coding sheets used to create the diagram; and
- the assignment brief plus the institution's current AI-use and appeal policy.
Preserve these before revising. Replacing the submitted file with a cleaner copy or overwriting the diagram source can erase the chronology you are trying to show. If revision is allowed later, make it in a copy and keep the original report beside it.
This order is not a detector-evasion method. It is evidence preservation. Editing only to move a score can make the authorship question harder to resolve, especially when the original wording and its working history are no longer available.
How to ask your instructor for a passage-level review
Start with the evidence gap, not an argument about whether `48%` feels high or low. Ask four concrete questions:
- Is this the AI Writing Report or the Similarity Report?
- Which exact passages are under review?
- What process evidence will be considered alongside the report?
- Which written policy and appeal path apply to this assignment?
A short request can be enough:
I wrote the diagram and the explanation myself. Could you share the complete AI Writing Report and identify the passages you want to review? I can provide the original diagram file, research notes, calculations, and version history for those passages. Please also let me know which course policy and review process apply.
This does not demand that the instructor ignore the report. It asks them to use the report at the level it actually provides evidence and to compare it with records the classifier never saw.
Frequently asked questions
Does Turnitin analyze diagrams? Turnitin's public page defines qualifying text around prose sentences and does not publish a categorical diagram rule. A visual diagram is not a prose sentence, but text around it may be. Check the actual highlights instead of assuming every diagram is processed or excluded.
Can a figure caption be included? The published guidance does not give a blanket answer for captions. `Figure 2` is a short label; a two-sentence caption is prose. The report's highlights are the relevant evidence for your specific file.
Does 48% mean almost half my assignment was written by AI? No. The percentage applies to qualifying prose, not necessarily the entire submission, and it is a classifier result rather than a finding about who wrote the work.
Should I rewrite the explanation immediately? Preserve the submitted file, full report, and process records first. Ask what passages and policy are under review. If revision is permitted, edit for accuracy and clarity in a copy, not merely to make a number disappear.
Will version history prove I wrote it? Not on its own. Version history, source notes, calculations, and the editable diagram file can form a checkable process record, but the institution decides how that evidence is weighed.
Official sources checked
- Turnitin, Using the AI Writing Report, checked 31 August 2026.
- Turnitin, How to access the AI Writing Report, checked 31 August 2026.
- Turnitin, How should I review the AI Writing report?, checked 31 August 2026.
- Turnitin, Understanding the similarity score, checked 31 August 2026.
KEEP READING