Submitting Through Canvas SpeedGrader: How Instructors Read the Turnitin AI Flag

You hit submit on a Canvas assignment, and somewhere on the other side a small colored number appears next to your name. Most students never see it. But your instructor, working inside SpeedGrader, sees two scores side by side: the familiar similarity percentage and, increasingly, a separate Turnitin AI writing flag. Knowing what that flag actually looks like from the grading chair helps you write with less anxiety and respond with more clarity if it ever comes up.

Key takeaways

• Inside Canvas SpeedGrader, the Turnitin AI flag is a separate number from the similarity score, and it isn’t shown to students by default.
• The percentage estimates how much of the submission Turnitin’s model thinks was AI-generated. It’s a probability, not proof.
• Instructors usually see a small colored indicator next to the similarity score, then click through to a separate AI writing report.
• A flag is a prompt for a conversation, not a verdict. Many schools forbid using it as standalone evidence.
• Understanding *why* detectors misfire helps you write naturally and stay calm if you’re flagged.

Where the flag lives inside SpeedGrader

Canvas SpeedGrader is the panel instructors use to grade one student at a time. When a Canvas assignment is connected to Turnitin, SpeedGrader pulls in Turnitin’s report instead of, or alongside, the plain Canvas submission view.

What an instructor sees there is usually two distinct markers:

• The similarity score, the long-standing number that flags overlap with existing sources.
• The AI writing indicator, a newer, separate percentage that estimates how much of the text appears machine-generated.

These are not the same thing, and that distinction matters. A paper can come back nearly clean on similarity and still surface a high AI flag, or the reverse. They measure different problems. Similarity is about copying; the AI flag is about *how the text was produced*.

The flag typically appears as a small colored badge or percentage. Clicking it opens Turnitin’s AI writing report in a separate view, where the instructor can see which passages the model considered likely AI-generated, usually highlighted in the document.

A quick mini-scenario

Picture an instructor named Dana grading 38 lab reports. In SpeedGrader, most submissions show a quiet, low AI indicator. One report shows a higher number. Dana clicks through to the AI report and sees the middle three paragraphs highlighted while the intro and methods sections aren’t.

Here’s the thing: that doesn’t tell Dana the student cheated. It tells Dana the model found those paragraphs statistically “smooth” in a way it associates with generated text. Dana still has to decide what to do with that, and a careful instructor treats it as a question, not an answer.

What the number actually means

Turnitin’s AI indicator is the output of a classifier that scores text on statistical patterns. Two of the big ones are perplexity (how predictable each word is) and burstiness (how much sentence length and rhythm vary). Generated text tends to be low-perplexity and low-burstiness: smooth, even, a little too consistent. If you want the deeper version, we walk through the statistics behind AI detectors in plain English.

The percentage you might hear about (say, “this came back 85% AI”) is the model’s estimate of how much of the document fits that pattern. It is emphatically not a measurement of intent, and it is not a certified fact. Detectors output probabilities. They guess, with confidence levels, and they’re wrong often enough that the whole field stays contested.

A useful reality check: OpenAI shut down its own AI text classifier in July 2023, citing a low rate of accuracy. The company that builds these models couldn’t make a reliable detector for them. That should temper how much weight anyone puts on a single percentage.

Why instructors are told to read it carefully

Most institutions that turn on the AI indicator pair it with guidance, because the flag has a real false-positive problem. A 2023 study by Stanford-affiliated researchers (Liang and colleagues) ran 91 TOEFL essays, all written by humans, through seven detectors. More than 60% were flagged as AI-generated by the detectors on average, and 97% were flagged by at least one of them. The same detectors were near-perfect on essays by U.S.-born eighth-graders. The takeaway: non-native English writing gets caught in these nets far more often.

So when a thoughtful instructor sees a high flag in SpeedGrader, the responsible read is something like:

> “This submission has statistical patterns my detector associates with AI. That’s a reason to look closer, not a conclusion.”

Looking closer might mean comparing the flagged paper to the student’s earlier work, checking the draft history, or simply asking the student to talk through their process. The flag is the start of a question. It is not the end of one.

This is also why the indicator is hidden from students by default. A raw probability score with no context invites panic and bad decisions on both sides of the gradebook.

What this means if you’re the one submitting

You can’t control Turnitin’s model, but you can control how you write and how you respond.

Write in a way that’s naturally varied. The patterns detectors latch onto, uniform sentence length, relentlessly clean transitions, no rhythm, are also the patterns that make writing dull. Mix short sentences with long ones. Use contractions. Add a specific detail only you would know. Good human writing and “low flag” writing tend to overlap.

Keep your trail. Draft in a tool that saves version history. Outlines, notes, messy early drafts, and timestamps are the most persuasive evidence you have if a flag ever gets questioned, far more convincing than arguing about a percentage.

Don’t assume a flag is an accusation. If an instructor raises one, it usually means they’re inviting you to explain, not announcing a verdict. Stay calm, ask what specifically was flagged, and offer your process. Many schools’ policies explicitly forbid using the AI score as standalone evidence in an integrity case.

Sanity-check your own drafts before you submit. If you’ve written something genuinely yourself but worry it reads “stiff,” it can help to see roughly how an automated tool reads your real voice, then revise toward something that sounds more like you.

A second mini-scenario, from the student side

Mara, an international student, writes a polished, grammatically careful essay because that’s how she was trained to write academic English. It gets flagged. Her instructor opens SpeedGrader, sees the indicator, and emails her.

Because Mara wrote in Google Docs, she shares the version history showing weeks of edits. The instructor reads it, recognizes the false-positive pattern the Stanford research describes, and closes the matter. The flag started a conversation; Mara’s evidence ended it. That’s the system working roughly as it should, and it’s why keeping a trail beats arguing about a number.

The similarity score vs. the AI flag, side by side

Because they sit so close together in SpeedGrader, it’s worth being crisp about the difference:

• Similarity report: Compares your text against a database of published work, web pages, and other submissions. High score = overlap with existing sources. This has been around for years and is well understood.
• AI writing report: Runs a classifier over your text looking for generated-text patterns. High score = the model thinks parts read like AI. Newer, noisier, and more contested.

An instructor who conflates the two does students a disservice. A high similarity score points to a fairly checkable claim (here’s the source you matched). A high AI flag points to a probability that can’t be verified the same way. They deserve different levels of trust.

Frequently asked questions

Can students see the Turnitin AI flag in Canvas? By default, no. The AI writing indicator is shown to instructors inside SpeedGrader and the Turnitin report, but it isn’t part of student-facing feedback unless an instructor or institution chooses to share it. Students usually learn about a flag only when an instructor raises it.

Is the Turnitin AI percentage the same as the similarity score? No. They’re two different metrics. Similarity measures overlap with existing sources. The AI percentage estimates how much of the text the model believes was machine-generated. A paper can score high on one and low on the other.

Does a high AI flag mean I’ll automatically fail or be reported? Not on its own. The flag is an estimate, not evidence. Most institutions treat it as a starting point for a conversation, and many bar instructors from using it as the sole basis for an integrity case.

Why did my completely original essay get flagged? Detectors look at statistical patterns like low perplexity and low burstiness, which can appear in clean, formal, or non-native English writing even with zero AI involved. A flag doesn’t prove how you actually wrote the piece.

Can an instructor turn the AI indicator off? It depends on institutional settings. Some schools enable it everywhere, some disable it, and some leave it to the department. If you’re unsure, ask your instructor or check the syllabus.

The short version

The Turnitin AI flag in SpeedGrader is a separate, probabilistic estimate that most instructors are told to read as a prompt for a closer look, not a confession. The best defense against false flags isn’t a trick. It’s writing naturally, keeping your drafts, and understanding the tool well enough to talk about it calmly. If you want to see how your own writing reads before you turn it in, you can check how your draft reads, or browse more guides on the blog for the full picture of how detection works.