Canvas has no built-in AI detector. When a student says “Canvas flagged my essay”, what almost always happened is that Turnitin or Copyleaks ran inside the assignment through an integration the institution installed. Below: what Canvas records, what it cannot see, how to check your own course, and what to do if a score is used against you.
Does Canvas detect AI?
No — Canvas has no built-in AI detector. Instructure ships no AI-detection feature in Canvas itself. What students experience as “a Canvas AI check” is almost always Turnitin or Copyleaks running inside a Canvas assignment via LTI, turned on by the instructor or the institution.
Checked against Instructure Canvas documentation Turnitin guides
Canvas does not detect AI writing. Instructure ships no AI-detection model inside Canvas, so submitting an assignment does not run your text through anything that scores it for AI authorship.
What Canvas does is deliver assignments, receive submissions and keep records: when an attempt arrived, what changed between attempts, and for quizzes an activity log if the instructor enabled one. Those are records about submitting, not analysis of your sentences.
The AI score comes from a separate product the institution licenses and plugs in: in practice, Turnitin or Copyleaks. Someone bought it, switched it on and attached it to that assignment. It is not a default.
If you are a student, the practical part is “How to check whether your own course has it turned on”, below. If you teach the course, it names the dropdown that decides it.
No. Canvas cannot detect ChatGPT, Gemini, Claude or Copilot, because Canvas does not analyse submitted text for authorship. A tool attached to Canvas might. Canvas itself does not.
Canvas also cannot see a ChatGPT sidebar, an AI browser, or a second window on another monitor. That is how browsers work: a page is sandboxed from every other tab, extension panel and application on your machine.
Instructure keeps adding AI features to Canvas, but for course building and grading, not authorship scanning. Any change would appear in its release notes first.
The oldest detection method is not software: instructors notice a shift in voice, an argument that matches nothing said in class, sources that do not exist. See how teachers spot AI writing without a detector.
Canvas supports a plagiarism framework that lets an external tool inspect submissions. Turnitin and Copyleaks are the two most often attached to it, and both sell AI-writing detection alongside similarity checking. A Canvas admin installs the tool at account level; the instructor enables it per assignment.
Where the result appears explains most of the confusion. The similarity indicator surfaces in Canvas beside the submission in SpeedGrader. The AI writing indicator usually does not live in Canvas at all: it sits in the vendor’s own report viewer, and students rarely see it. So a student can honestly say “Canvas flagged me for AI” without ever having seen the tool that produced the number.
More on how AI detectors actually work, and on Turnitin vs TextSight for teachers and how Copyleaks compares.
If you are a student, open the assignment before you submit. Three signals mean a plagiarism or AI tool is attached: a licence agreement prompt from Turnitin or Copyleaks before your submission is taken; a similarity indicator or report link next to your submission afterwards; or a submission page that clearly is not Canvas, which means an external tool. Read the assignment description and syllabus too, since institutions using these tools are usually required to disclose it. What you cannot tell from the student side is whether the AI-writing part is enabled, because its output is instructor-only. So ask. A course that uses detection should be able to answer.
If you are the instructor, edit the assignment and look at the submission options. With an online submission type selected, a Plagiarism Review dropdown appears. Whatever is named there is what runs: Turnitin, Copyleaks, another vendor, or None. If it says None, nothing is scanning that assignment. AI detection itself is usually not a Canvas checkbox but part of your institution’s licence, so your Canvas admin or the vendor’s console is where to confirm it.
No. Canvas does not log paste events on assignment text entry. It does not read your clipboard and it does not record your keystrokes.
What Canvas always records is timing. Every submission is timestamped and every attempt is kept in the submission history, so an instructor can see that a finished essay arrived seconds after the assignment page was opened. That is a timing inference, and the difference matters more than anything else on this page: “this was submitted quickly” and “this was pasted” are different claims. Canvas can support the first. It cannot make the second, because it never sees where your text came from.
The clearest evidence is Instructure’s own: blocking copy and paste in text entry is still an open feature request in the Instructure Community, and a capability people are still asking for is one that does not exist. The full breakdown of what is and is not recorded is in what Canvas records.
Quick submission has ordinary explanations: most people draft somewhere with autosave and paste the finished text in, which is normal and, at almost every institution, not misconduct. File uploads involve no text entry, and the file’s own revision history is more informative. A similarity tool such as Turnitin may still match pasted text to its original source — see can Turnitin detect ChatGPT.
For instructors: a short gap between opening and submitting is a reason to ask a question, not to conclude anything. “Walk me through how you wrote this” settles it. A timestamp does not.
Most pages answer this as a yes/no. It is not one. The middle state — recorded only if your instructor switched it on — is where nearly every “Canvas tracks your tabs” post goes wrong.
On every assignment and quiz. This is submission metadata Canvas keeps by default.
What an instructor can infer from this: that a finished essay was submitted seconds after the assignment page was opened. That is a timing inference, and the difference matters: it is not paste detection. Canvas did not see a paste. It saw a clock.
Quizzes only, and only when your instructor has turned on Quiz Log Auditing for that quiz. On assignments, and on quizzes without it, none of this is captured at all.
What this does not tell your instructor: which tab or app you switched to, what was on it, or what you did while you were away. The log records that the quiz page lost focus. Nothing more.
Not by Canvas itself, on any assignment or quiz type, with or without quiz logging.
Screenshots cannot be detected at all. A screenshot is taken by the operating system and fires no event a web page can read, on macOS, Windows, iPhone, iPad and Android alike. No setting changes that.
Split screen is worth getting precise about, because it is the opposite case. In iPadOS Split View the Canvas page stays visible, so no blur or visibility event fires at all. Split View and Slide Over are drawn by the system rather than by the web page, and Canvas gets no view into whether an app was swapped or a divider dragged. Switching away can be logged on a quiz; splitting the screen cannot, because you never left.
A phone pointed at the screen never touches Canvas at all. Where a proctoring tool is in use you were told, because screen recording, webcam recording and device lockdown all require software you install and permissions you grant. There is no silent version. More on how we handle student data.
Canvas has no built-in lockdown browser and does not record your screen, your face or your room. The products that do are bought separately by the institution and each announces itself: you download a browser, install an extension, or grant camera access. Their capabilities are their own, not Canvas’s.
What Canvas offers natively is the quiz log. Classic Quizzes calls it Quiz Log Auditing; New Quizzes has an equivalent view. It timestamps events during an attempt, including that the quiz page stopped being viewed, which is the source of nearly every “Canvas tracks tab switching” post.
Instructure’s own documentation is unusually blunt here: quiz logs are not intended to validate academic integrity or identify cheating for a quiz. That is the platform owner saying its own log is not evidence.
Instructors also have build options such as an access code, IP filtering and a time limit. Those restrict what a student can do; none detect anything. And quiz essay responses are generally not routed through the plagiarism framework, so a quiz essay is usually the least-scanned text in a course.
No, and discussions are the weakest place in Canvas for detection of any kind. The plagiarism framework attaches to assignments, so a discussion reply is normally not scanned at all, even when the discussion is graded. Some institutions do route graded discussions through a tool, so check the specific assignment. Canvas timestamps posts and marks an edited post as edited: metadata about posting, not analysis of writing.
What does happen is manual: an instructor reads a reply that does not sound like the student and pastes it into a detector. That is a person choosing to check, not Canvas doing it.
For instructors, one caution. Discussion replies are short, and short text is where detectors are least reliable. Turnitin does not return an AI score below a minimum amount of prose, for precisely this reason. To find out whether a student understood the reading, ask a follow-up question in the thread.
Every AI detector, ours included, returns a probability: a model’s estimate that text resembles patterns learned from AI-generated writing. It cannot measure what happened at your desk.
False positives are real and documented. A 2023 Stanford study by Liang and colleagues found that popular detectors misclassified a majority of TOEFL essays written by non-native English speakers as AI-generated, while performing far better on native-speaker writing. Plain, formulaic prose scores higher on average, which describes second-language writing, lab reports and the five-paragraph essay. See are AI detectors biased against ESL writers? and AI detector false positives.
We publish the same kind of measurement for our own detector, because a page arguing that scores need scrutiny should not exempt its own. We ran TextSight over 1,180 human-written academic papers published in 2018, years before ChatGPT, so every flag is a false positive by construction. It flagged 5.85 percent. Looking for the second-language penalty the Stanford study found in other tools, we did not find one in ours: the gap was −3.5 percentage points, 95 percent confidence interval −7.8 to +0.7, which contains zero. That is a null result, not a claim of no bias, and the scope is narrow — journal-edited abstracts of 150 to 400 words from one year, scored by one version of our model. It does not transfer to a student essay submitted through Canvas. The dataset and the analysis script are published so you can check it.
Turnitin’s own guidance states that its AI indicator is not a determination of misconduct and needs human review before any allegation. Our position on our own product is the same. More on whether Turnitin’s AI detector is accurate, whether AI detectors can be wrong, and why detectors get it wrong.
Grammarly is the most common source of confusion. A detector sees only the final text, so it cannot distinguish “Grammarly fixed my commas” from “Grammarly’s generative feature wrote this paragraph”. Accepting suggestions that rewrite your sentences moves the text toward the patterns detectors score, though a grammar checker is not misconduct at most institutions. See does Grammarly use AI.
Gemini, Claude and Copilot are covered by the same detectors as ChatGPT: detectors score statistical patterns rather than recognising a product. No detector can name the model that produced a passage.
This section is about process. We will not tell you how to change your writing to avoid a detector, and we do not publish that anywhere on this site. What settles a meeting is your ability to show and discuss your work.
If you wrote it and you have been flagged, do these things, in this order.
More detail: essay flagged as AI but you wrote it, accused of using AI in college, how to prove you didn’t use AI, and how to write an AI detection appeal letter.
For instructors: open with a question, not an allegation. The vendors’ own guidance says a score alone is not sufficient, so acting on one departs from the tool’s documented use. See AI detector for professors and sentence-level AI detection.
Before anything happens: leave version history on. It is the evidence you will wish you had.
TextSight is a text AI detector. It does not proctor, does not watch your screen or camera, and is not connected to your Canvas gradebook. There are two honest uses for it around a Canvas course.
Students can run a draft through it before submitting, to see whether writing they did themselves reads as AI and which sentences carry that risk. That is a self-check, not a guarantee: our score is a probability, and we cannot tell you what Turnitin or Copyleaks will say about the same text.
Instructors can use it as a second opinion on a passage before starting a conversation, with sentence-level highlights showing what drove the score. A second model agreeing is still not proof, only one more input to a judgement that stays yours.
Our detector is English-only, and we would rather say so than let you find out from a bad result. Try it on the AI detector page or in the web app; plans are on the pricing page; more on For students and For teachers.
What works today is manual: copy the submission text out of SpeedGrader, paste it into TextSight, read the score and the highlights, and write your own comment back in Canvas. TextSight never touches your Canvas instance, so there is nothing for an admin to approve.
To be told when a native integration exists, email the address below with your institution, your Canvas environment (Cloud or self-hosted), and roughly how many instructors would use it. It is non-binding on both sides. The same is true of our Blackboard, Moodle and Google Classroom pages: none are shipped either.
Submitting opens your mail client addressed to support@textsight.ai. Please include Canvas environment (Cloud or self-hosted) and rough instructor headcount in the body. No tracking pixel on this form. No automated signup, on purpose.
If your question is not below, email support@textsight.ai. Replies go to a human inside our team, not a routing bot.
No. Canvas has no built-in AI writing detector. Nothing in a standard Canvas assignment scans your text for AI. What students experience as a Canvas AI check is almost always Turnitin or Copyleaks running inside the Canvas assignment through an LTI integration that the institution licensed and switched on separately. Canvas passes the submission to that tool and displays the result it sends back. Canvas itself never reads the text for AI.
No. Instructure has not shipped an AI writing detector inside Canvas, and there is no setting an instructor can switch on to make Canvas scan for AI by itself. Institutions add detection by installing a third-party tool through the Canvas plagiarism framework or the newer LTI asset processor. If your course shows an AI score, an outside vendor produced that number, not Canvas.
Not by itself, and no detector can name ChatGPT specifically. If your institution enabled Turnitin or Copyleaks on the assignment, that tool estimates how much of the submission reads as AI-generated and reports a figure to the instructor. That figure is a statistical likelihood measured across large language models in general. It is not a ChatGPT fingerprint and it does not identify which tool you used.
The instructor sets up an assignment with a plagiarism or AI tool attached over LTI. When a student submits, Canvas hands the text to that vendor, the vendor scores it, and the score returns into SpeedGrader beside the submission. The instructor sees a percentage or an indicator, rarely an explanation of why. Canvas is the pipe and the display surface. The vendor does the analysis and owns the model.
Not on ordinary assignment submissions. Canvas does not log paste events in a text-entry box, and blocking copy and paste in text entry has been raised as a feature request in the Instructure Community rather than shipped as a feature. What Canvas does keep is the submission timestamp and every submission attempt, which an instructor can open and compare. A similarity tool such as Turnitin may still match pasted text to its original source.
Only inside a Canvas quiz, and only when Quiz Log Auditing has been enabled for the course. The quiz log then timestamps a stopped-viewing event when the quiz-taking page loses focus, and timestamps your return. It records that the tab lost focus and for how long. It cannot see which site, app or document you opened. Outside quizzes, Canvas logs nothing of the kind.
No. Canvas runs in an ordinary browser tab and has no access to your other tabs, your browsing history, your clipboard, your screen or your other devices. It cannot detect a screenshot, a second monitor, split screen on an iPad, or a phone lying beside you. Only a lockdown browser or a proctoring product that your institution installs separately can restrict or observe any of that.
Canvas has no webcam recording, no screen recording and no lockdown browser of its own. Institutions add those with separate products such as Respondus LockDown Browser and Monitor, Proctorio or Honorlock. Those tools announce themselves: you have to install or launch something, and the browser asks for camera permission before the attempt begins. If nothing asked for your camera, nothing recorded your face.
Not natively. Canvas quizzes run in your normal browser unless your institution has licensed a separate lockdown product, most commonly Respondus LockDown Browser, and attached it to that specific quiz. When it is in use you cannot start the quiz without launching the special browser first, so it is never a surprise. A Canvas quiz you can open in Chrome as usual is not locked down.
Usually not. The Canvas plagiarism framework attaches to assignments, not to native discussion entries, so Turnitin and Copyleaks do not scan Canvas discussion posts the way they scan an assignment submission. Instructors who want discussions checked normally ask for the same text as a graded assignment as well, or paste posts into a checker by hand. If you are told discussions are scanned automatically, ask which tool is doing it.
There is no Canvas AI detector to rate, so the real question is how accurate Turnitin or Copyleaks is on your particular assignment. Every AI detector produces false positives and no vendor claims none. A score is a probability that text resembles AI-generated writing, never proof that it is. Ask which tool produced the number, what error rate that vendor publishes, and what evidence exists besides the score.
No. Canvas holds circumstantial records: submission timestamps, earlier submission attempts, and on quizzes with logging enabled, when the quiz page lost focus. A detector attached to the assignment adds a probability score. None of that establishes authorship. Institutions generally treat a high score as a reason to open a conversation, and most academic integrity policies expect corroborating evidence before a finding is made.
Primary documentation, not other people’s blog posts. Check them yourself.
Moodle has no native detector either. What runs there, and what the activity log actually shows.
Blackboard, SafeAssign and the difference between similarity checking and AI detection.
Originality reports, what they compare against, and why that is not AI detection.
Who gets falsely flagged, why plain prose scores higher, and what to do about it.
How to use a detector as one input to a judgement rather than a verdict.
What Turnitin publishes about its own error rate, and how to read the number.
Run your own writing through a detector before you submit, or get a second opinion on a passage before you raise it with a student. English-only, and the score is a probability, not a verdict.