Moodle AI Detector Workflow

Moodle's assignment activity accepts student work in two main shapes — online text typed straight into the editor, and uploaded files — and treats both the same way it treats everything: neutrally. There is no built-in judgement about how the text was produced. Moodle does have a long-standing plagiarism-plugin framework through which institutions connect third-party integrity tools, but whether anything is plugged into yours is an administrative decision made far above the grading screen, and on a great many Moodle sites the answer is: nothing is.

So let's be precise about what this page offers. TheChecker.AI has no Moodle plugin. It does not appear in your activity settings and never will by installation. What it is instead is a paste-based second-opinion tool that runs in a browser tab next to your grading view — which cuts both ways: no integration, but also no waiting on your Moodle administrator, no version-compatibility worries, and nothing new touching your site.

The grading-view workflow, step by step

1. Open the assignment and view the submissions. From the assignment activity, open the list of submissions and click into the attempt you want to review, the same way you would to grade it.
2. Get the text. For online-text submissions, the writing is displayed right in the grading view — select and copy it. For file uploads, download the document, open it, and copy the text from there.
3. Paste into the free demo. No account, nothing stored, results in seconds. Paste the full piece where you can: sentence-level scoring is most informative with complete context, and partially generated essays only show their structure when the detector sees all of it.
4. Read the result — the sentence map and the named model, covered below — and then grade in Moodle as usual, with one more piece of evidence than you had.

Two shortcuts exist for people who do this often. The Chrome extension lets you check text in the browser where you're already reading it, trimming the tab-switch out of step three. And because so many Moodle sites are self-hosted by institutions with their own developers, the REST API is unusually relevant here: your team can build submission checking into whatever internal tooling you run, on your terms. That is an engineering project on your side — we provide the detection endpoint, not the plugin.

Reading the result

TheChecker.AI's output is built around two things a single percentage can't tell you.

Sentence-level scores. Every sentence gets its own score, so the overall number resolves into a map: is the flagged text spread thin across the whole essay, or concentrated in two suspiciously fluent paragraphs? Those are different situations calling for different conversations, and the map shows you which one you have. Hybrid submissions — a student's own draft with generated sections stitched in — are exactly what document-level scores blur and sentence-level maps reveal.

The likely model, named. The text is compared against more than 40 models — GPT-4 and GPT-5, Claude 3 and 3.5, Gemini Ultra, Llama 3, Mistral and others — and the result names the closest match. That specificity changes the follow-up: "this section reads like ChatGPT output" is an observation a student can respond to; a bare "AI: 71%" is just a number to argue about.

In our own benchmark the detector reaches 93% accuracy. The accuracy page explains how that figure is measured and what it does and doesn't claim — read it before you lean on the number, because knowing a tool's limits is part of using it fairly.

Fairness: signal, not verdict

The uncomfortable truth about every AI detector — ours included — is false positives. Formulaic structure, non-native English, and heavily polished prose all raise scores on every tool on the market. A flag on a real student's work therefore obligates you to three things:

• Look at why. Read the sentence map before the total. A score you can't explain shouldn't drive a decision.
• Ask about the process. Drafts, notes, and earlier versions tell you more than any percentage. If the work was drafted in Google Docs before landing in Moodle, its revision history is often the most decisive evidence available — the Google Docs workflow covers how to read it.
• Talk before you conclude. A five-minute conversation about how the argument was built resolves most flags, in whichever direction the truth lies. The detector's job is to give that conversation specific passages to start from.

The teachers page lays out this whole approach as a classroom workflow — calibration on known text first, sentence maps over single scores, conversation over accusation. It is how more than 10,000 educators across 500+ institutions use the tool, and it works the same whether your LMS is Moodle, Canvas or anything else. If checking becomes routine, a paid plan removes the demo's limits.

When your institution already has a detector in Moodle

If your site's plagiarism framework is wired to Turnitin or a similar service, integrity reports may already appear in your grading flow. That doesn't make a second opinion redundant — it makes one possible. A single detector's flag is one statistical model's view; a second tool that reasons differently and shows different evidence tells you whether that view holds up.

The protocol is short. When the integrated tool flags a submission, run the same text through the free demo before you do anything else. Three outcomes, three meanings. Both tools flag the same passages: your footing is much stronger, and the sentence map now tells you which sentences to raise with the student. The second opinion comes back clean: proceed with real caution — one of the two tools is wrong, and with false positives endemic to the field, the flag may be the error. Partial overlap: the intersection is where to focus, and the disagreement is worth mentioning honestly in any report you write. What the protocol never does is average the numbers into a verdict; two scores are two pieces of evidence for a human judgement, not a vote. The Turnitin comparison maps out in detail where each tool's approach starts and stops.

Checking a whole class efficiently

Whatever tooling you have, resist screening every submission by hand. Pasting an entire class through a detector burns hours, frames every student as a suspect, and — because false positives are a statistical certainty at volume — manufactures difficult conversations about work that was honest all along.

Triage is the workable discipline. Grade as usual and let your reading flag the anomalies: the essay whose register doesn't match the student's forum posts, the argument shaped like nothing you taught, the suspiciously frictionless paragraphing. Those few submissions — not the whole gradebook — get the paste-and-check treatment, and a handful of checks fits inside a normal grading session. Instructors running large courses will find this triage approach developed further on the professors page.

Moodle adds one genuinely distinctive option here. Because so many sites are self-hosted, "check systematically" can be a real engineering decision rather than a wish: your developers can build against the REST API to check submissions on whatever schedule and criteria your integrity policy defines, with results flowing into your own tooling. That keeps the judgement where it belongs — in your institution's process — while removing the copy-paste labour.

Edge cases: uploads, forums and scanned work

Paste-based checking needs selectable text, and Moodle's flexibility produces a few shapes that test that requirement.

• File uploads are the everyday case: download, open, copy. With PDFs, glance at your pasted text before running the check — exports from unusual tools sometimes scramble line breaks, and a detector scoring mangled text tells you nothing.
• Scanned and photographed documents — a handwritten exam scanned to PDF — contain no text layer. Nothing can be pasted, so nothing can be checked; that is a hard limit of any text-based detector, worth knowing rather than working around.
• Forum posts, workshop submissions and essay-question answers all work fine. If you can select the text, the workflow applies unchanged — useful in courses where the writing that matters happens outside the assignment activity.
• Work drafted elsewhere carries evidence Moodle never sees. An essay written in Google Docs before upload has a revision history that can corroborate or complicate a flag in either direction; if a flagged student says "I drafted it in Docs," that history is usually worth more than a second score. The same applies on other platforms — the Blackboard workflow faces identical constraints with file-heavy submissions.

Frequently asked questions

Does Moodle have built-in AI detection?

No. Moodle itself does not detect AI-generated writing. Its plagiarism-plugin framework lets institutions connect third-party integrity services, but what is available depends entirely on what your institution has installed and licensed.

Is there a TheChecker.AI plugin for Moodle?

No. TheChecker.AI works alongside Moodle rather than inside it: copy the submission text from the grading view and paste it into the free demo. Institutions that host their own Moodle and want automated checking can have their developers build against our REST API — that is a project on their side, not a plugin we ship.

How do I check a Moodle assignment for AI?

Open the assignment's submission list, open the attempt you want to review, and copy the student's text — directly from an online-text submission, or from the downloaded file for uploads. Paste it into TheChecker.AI's free demo for a sentence-level result in seconds. Nothing you paste is stored.

Can the detector be wrong about a student?

Yes. Every AI detector produces false positives, ours included — formulaic writing, non-native phrasing and heavily edited prose all raise scores. Use the sentence-level result as the start of a conversation about the student's drafting process, never as a verdict on its own.

Does Turnitin work with Moodle?

It can — Turnitin is one of the third-party services institutions commonly connect through Moodle's plagiarism-plugin framework, and where it is licensed and enabled its reports appear in the grading flow. Whether your site has it is your administrator's decision. Even where it is present, an independent second opinion that reasons differently — per-sentence scores, a named model — tells you whether a flag is corroborated or shaky.

Can I check Moodle forum posts or quiz essay answers?

Yes. The workflow only needs text you can select and copy, so it applies to anything in Moodle that displays writing — forum posts, essay-type quiz answers, workshop submissions, online-text assignments. Copy the text, paste it into the free demo, and read the sentence map the same way you would for an essay.

Do I need my Moodle administrator's permission to use this?

No, because nothing is installed. TheChecker.AI runs in a separate browser tab and never touches your Moodle site, so there is no plugin to approve, no compatibility to test, and no data flowing between the systems except the text you choose to paste — which is not stored.