Getting started with unsupervised activities

Step-by-step setup

1. Make sure to communicate expectations to your students well in advance (consult your institution's policies for deadlines and other constraints). MATCHA is not going to deter AI misuse at-home all by itself; it works together with appropriate course and grading policies. Some key points that should be part of your policies:
   
   • All work should be done in MATCHA, including reading and note-taking. Submitting a project with adequate evidence of work at all stages should be required for full marks (students are rewarded for showing their work). The at-home essay syllabus preset suggests giving 30% of marks for the work process.
   • Students should be required to add all documents they read to their library (using the Add to Library button) and to cite those specific library entries using the built-in citation tool ( ). This will allow you to more clearly correlate reading and citing. In addition, they should be asked to attach relevant passages by copying them from the clipboard when inserting a citation. This is easy for an honest student to do, but hard for someone transcribing an essay. Ideally, full marks should require proper use of these citation tools.
   • An in-class follow-up quiz (facilitated by MATCHA) should be used in courses in which AI misuse is more likely, such as first and second year courses. Full marks should require passing the quiz. See below for setup instructions.
2. Create an account in the MATCHA Dashboard and create a course.
3. Create a browser configuration to allow your students access to all relevant resources.
   • The easiest way to create a new browser configuration from scratch is to use the Create allowlist option in the Instructor Menu of MATCHA Writer.
   • You will choose a name for your configuration then browse the web to the locations that you want your students to be able to access.
   • All visited domains and their dependencies will be added to the allowlist.
   • Make sure to access the required resources through your library.
   Once you have done this, view and adjust your browser configuration in Activity Settings / Browser . Make sure to select some Featured Links for students to use as starting points. Browser configurations can be reused between courses and activities, so this is mostly a one-time setup. Note that the default browser allowlist includes over a thousand publisher websites and other academic resources. But you may need to add your university's LMS. Tip: most textbooks are available as online ebooks on platforms such as Google Play, which students can access within MATCHA's browser.
4. Create an activity and enable the built-in browser. Select the browser configuration you created in step 3. See below for suggested activity settings.
5. Decide whether students may use MATCHA's built-in AI assistant, and if so at what level. For unsupervised activities, an instructor-configured assistant can give students a permitted source of help inside MATCHA instead of pushing them toward outside AI tools.
6. Ideally, pair the activity with a Follow-up quiz . If you choose to do so, you will need to design a reusable quiz in Activity Settings. We provide suggested questions to make this easy. Plan to review and edit AI-generated questions as part of the grading workflow (if grading using MATCHA, each student's quiz will appear as an additional tab in their project file). Decide in advance how the quiz will contribute to the rubric, focusing on students’ understanding, decisions, and reflection. Review AI-generated questions and preliminary evaluations before using them in grading.

Recommended activity settings for unproctored activities

• Proctoring: off. This is the normal choice for take-home or remote work.
• Availability, start and end: set dates and times. The availability times in MATCHA set hard limits for when students can work on an activity (or generate a submission document). The end time is not their deadline. It is the time past which you definitely do not need to accept more submissions. This is often set to a date after the course ends. We ask for an end time for housekeeping purposes, not to police deadlines. Deadlines are communicated using a separate, optional setting.
• Writing time : Set your activity's deadline here. This is for information only as deadlines are not currently enforced by MATCHA.
• Word limit : If desired, specify a word limit with possible enforcement. Note that MATCHA has no word limit in the Notes tab.
• Notes: enabled. Encourage students to use MATCHA notes rather than external tools.
• Browser: enabled. Normally, students need course readings, library pages, and your LMS. By using the built-in browser, they can create a record of their work process.
• Anonymous submission: optional. Use it only if anonymous grading is part of your process.
• External-Paste Block: enabled. This is one of the most important deterrents for unsupervised work because it prevents direct transfer from AI tools or other external sources. If you turn it off, externally pasted text is marked (it can be revealed using an option in the View menu and is reported in aggregate in various reports).
• Complete Work History: enabled. This captures detailed browsing history, enabling detailed reports of how much each cited (or uncited) work has been read. We consider this feature to be crucial for unsupervised work. If enabled, be sure to warn your students in advance and point them to our Student's Guide for information, because this feature has privacy implications. Make sure to include a reference to our Student's Guide in your syllabus.
• AI Assistant: optional but recommended. To help discourage disallowed uses of AI, we recommend enabling MATCHA's AI assistant at a level that fits the assignment: grammar only, grammar and style, or full composition. If you want to help your students learn writing mechanics, turn on pedagogical mode. This makes the assistant act as a tutor rather than auto-corrector. AI-made changes are tracked and reported in the Writing Analysis, so permitted AI use remains visible in the work record. Be explicit in your syllabus and activity instructions about what level of AI help is allowed.
• Follow-up quiz: enabled .
• Submission mode: MATCHA Cloud (.mcha) . To facilitate the evaluation of submissions, use MATCHA Cloud submissions paired with our in-app grading tool, which facilitates the analysis and review of students' work. If you have to, you can use LMS-based submissions, but this will add steps for both you and students. We offer privacy options that keep submissions fully secure on MATCHA Cloud (see our Data Privacy page ).

How to evaluate submissions

We advise instructors to give substantial weight to students' work records, including reading and editing. Do not think of MATCHA as an AI-detection tool, because drawing a bright line between suspect and non-suspect submissions is impossible. Some students whose minimal histories suggest that they might have used AI may, in fact, simply be doing the minimum amount of work. It will be difficult to distinguish between these two kinds of low-effort submissions. Instead, think of MATCHA as providing a complete record of work that you can grade. Students who do minimal work will be penalized alongside students who misuse AI, but that is acceptable because neither deserves a high grade.

Ideally, use MATCHA Cloud submissions so that you can benefit from the Grading Tool in the Instructor menu, which can automatically flag submissions with low work metrics. If you are not using MATCHA Cloud submissions, open the projects you receive using the Open project for grading tool in the Instructor menu. When opened in this mode, a project displays the history analysis as the first tab you see. You can also open the same analysis using History > Analysis in any project. MATCHA's DOCX exports contain roughly the same information in an added preamble.

The most important metrics are the following:

Reading time per word written

This is the number of seconds the student spent reading in the browser per word written. A typical essay assignment requires a minimum of two hours of reading per 1,000 words written, or about seven seconds per word. Expected values will, of course, depend on the specifics of your assignment, but this figure is a good reference point.

Time fragmentation

This measures how much editing was done on the document. We start by calculating the number of characters that are followed by characters typed before them (for example, “Ab” when “b” was typed before “A”). This indicates an edit made after the initial typing. We express this as a percentage of total characters, cap it at 10%, and divide it by 10% (10% is about as high a raw value as we expect in practice). The resulting value expresses how close the document's time fragmentation is to the expected maximum. Normally, a well-edited document should score at least 10–30%, depending on the nature of the activity. MATCHA's time/location graph also gives you a visual, intuitive sense of time fragmentation but does not take into account text in the Notes tab.

Character churn

This is simply the percentage of typed characters that did not make it into the final copy. Well-edited essays normally have at least 30–40% character churn.

We suggest focusing on these metrics and rewarding students for meeting the expected thresholds. The specific weight given to each metric and the thresholds used should be adapted to your activity and students, but it is possible to devise a simple work-grading rubric that reliably rewards good work and penalizes low effort or cheating.