The artificial intelligence applications offered to everyone by Häme University of Applied Sciences are Microsoft 365 Copilot and Google’s AI services with our login.

As a process

1. Prepare student submissions

• Download the files returned by students from the Moodle assignment tool to your own computer.
• Extract the files to your own folder in OneDrive. Remember to name the folder clearly and separate the files by task.

2. Create an evaluation rubric to support feedback generation

• Create an assessment rubric for the assignment using Copilot.
• When creating the rubric, use:
  • Module and course learning objectives
  • Assignment
  • Assessment scale used

3. Enable Copilot’s Researcher Agent

• Open Microsoft 365 Copilot or the Copilot app in Teams.
• On the left side, select Agents → All Agents → Researcher
• Add or open the Researcher agent.

4. Attach student files to the agent and provide instructions to support verbal feedback

• Attach a folder from OneDrive containing student submissions to the Researcher Agent.
• Write a clear prompt (instructions) that tells you (see tips below for more details):
  • What files the folder contains (e.g. “students have made plans for XXX”)
  • Which folder is the data retrieved from?
  • What other sources should not be used so that Copilot does not mix in extra information?
  • Include the evaluation rubric in the prompt.
• Ask the agent to analyze the files one by one and generate verbal feedback based on each piece of work using the evaluation rubric.

5. Answer the agent’s clarifying questions

• The researcher agent can ask clarifying questions. By answering them, you can influence the style, tone, and presentation of the verbal feedback.

6. Review and edit the verbal feedback generated by Copilot

• Review the verbal feedback produced by the Copilot agent and ensure that it matches the student’s work and the objectives of the assignment.
• Edit feedback. The teacher is responsible for the quality, appropriateness, and final form of the feedback given to the student.

7. Transfer feedback to Moodle

• Write your edited feedback text for each student in the Moodle assignment feedback field.

Tips and danger spots

• Openly tell students about utilizing artificial intelligence to support feedback.
• Name the folder clearly – don’t just use “My OneDrive”.
• Tell the agent the exact location and limit the sources to be used.
• Give clear instructions on what to do and in what format.
• An assessment rubric is ​​essential: it brings consistency, alignment, transparency, and pedagogical control to assessment.

Remember: The teacher retains pedagogical control – Copilot supports the creation of verbal feedback, but the teacher is responsible for the content and final use of the feedback.

Instructions for creating an evaluation rubric

An assessment matrix (rubric) is a table that defines the assessment criteria and quality levels. It is recommended to create the rubric as follows:

1. Determine the evaluation criteria

• Base the criteria on the course’s learning objectives and assignment.
• Examples of criteria:
  • Understanding the content
  • Argumentation and justification
  • Structure and clarity
  • Use of sources

2. Define quality levels

• Usually 5-0 levels (e.g. Excellent, Good, Satisfactory, Needs improvement, Poor, Failed).
• For each criterion, what is required at each level is defined.

3. Record the matrix clearly in the table

• Example:

Criteria	Excellent (5)	Good (4)	Satisfactory (3)	Needs improvement (2)	Poor (1)	Failed (0)
Understanding the content	The assignment demonstrates in-depth understanding and application	The task demonstrates good understanding	The task demonstrates basic understanding.	The task shows a lack of understanding.	The task shows a total lack of understanding	Completely missing
Argumentation	The argumentation is coherent and well-founded.	The argumentation is mostly coherent.	The argument is partially coherent.	The argument is unclear.	Your argument is really flawed.	Completely missing
Structure and clarity	The structure is clear and logical.	The structure is mostly clear.	The structure is partially clear.	The structure is unclear.	There is no structure.	Completely missing
Use of sources	Sources used diversely and correctly.	Sources used correctly.	Sources used partially correctly.	Inadequate use of sources.	The use of sources is really lacking.	Completely missing

4. Ensure alignment

• The criteria and levels correspond to the objectives of the course.
• The rubric is ​​transparent and clear even for the student.

Example prompt for the Researcher agent to generate feedback

You can take advantage by copying and modifying this prompt to suit your needs:

Example prompt for the Researcher agent:

“The folder ‘CourseXXXXX Spring 2026’ in OneDrive contains the plans returned by students in DOCX format. Analyze each file separately based on the evaluation matrix below. Do not use any sources other than the files in this folder. The feedback must include:

• Assessment of each criterion in the matrix (understanding of content, argumentation, structure and clarity, use of sources)
• Brief justification for each criterion
• Summary of strengths and areas for development

Assignment: “In the individual assignment, the student’s goal is to familiarize themselves with……”

Evaluation matrix (rubric):

Criteria	Excellent (5)	Good (4)	Satisfactory (3)	Needs improvement (2)	Poor (1)	Failed (0)
Understanding the content	The assignment demonstrates in-depth understanding and application	The task demonstrates good understanding	The task demonstrates basic understanding.	The task shows a lack of understanding.	The task shows a total lack of understanding	Completely missing
Argumentation	The argumentation is coherent and well-founded.	The argumentation is mostly coherent.	The argument is partially coherent.	The argument is unclear.	Your argument is really flawed.	Completely missing
Structure and clarity	The structure is clear and logical.	The structure is mostly clear.	The structure is partially clear.	The structure is unclear.	There is no structure.	Completely missing
Use of sources	Sources used diversely and correctly.	Sources used correctly.	Sources used partially correctly.	Inadequate use of sources.	The use of sources is really lacking.	Completely missing

“Create feedback in a clear and understandable format for the student. If you need clarification, ask me.”

Below is an example of copying and pasting the above prompt into an AI application’s prompt window. The AI ​​application usually doesn’t mind if the table attached to the prompt looks messy to a human, because the AI ​​doesn’t “see” a table the same way a human does. It processes text as structures, words, rows, column separators, and repeating patterns.

The main reason is this: Humans interpret a table visually. Artificial intelligence interprets it as a text structure.

AI doesn’t mind if a table looks messy to a human, as long as the text still has a logical structure. It doesn’t evaluate the beauty of the design, but rather recognizes the headings, rows, criteria, grade levels, and the relationships between them in the text. The risk only arises if the design breaks the semantic relationships so that it is no longer clear which description belongs to which criterion or grade.

The folder ‘CourseXXXXX Spring 2026’ in OneDrive contains the plans returned by students in DOCX format. Analyze each file separately based on the evaluation matrix below. Do not use any sources other than the files in this folder. The feedback should include:

A rating for each criterion in the matrix (understanding of content, argumentation, structure and clarity, use of sources)
A brief justification for each criterion
Summary of strengths and areas for development
Assignment: “In the individual assignment, the student’s goal is to become familiar with……”

Assessment matrix:

Criterion Excellent (5) Good (4) Satisfactory (3) Needs improvement (2) Poor
(1) Failed
(0)
Content understanding The assignment demonstrates in-depth understanding and application The assignment demonstrates good understanding The assignment demonstrates basic understanding The assignment demonstrates inadequate understanding The assignment demonstrates a lack of understanding Completely absent
Argumentation The argumentation is coherent and well-founded The argumentation is mostly coherent The argumentation is partly coherent The argumentation is unclear The argumentation is seriously flawed Completely absent
Structure and clarity The structure is clear and logical The structure is Mostly clear Structure is partly clear Structure is unclear No structure Completely missing
Use of sources Sources used in a variety of ways and correctly Sources used correctly Sources used partially correctly Inadequate use of sources Really deficient use of sources Completely missing
Create feedback in a clear and understandable format for the student. If you need clarification, ask me.

Data protection tips for teachers

1. Minimize the amount of material to be processed

Do not export more student data to Copilot than is required to generate feedback. The instructions advise downloading student responses from the Moodle assignment tool and extracting them into your own OneDrive folder for each assignment, so from a data protection perspective, it is a good idea to only include the files needed to assess that assignment in the folder.

Practical tip:
Delete or avoid files containing unnecessary personal information. If feedback can be generated without the student’s name, student number, group information, or other identifier, consider naming the files, for example, Task1_student01.docx.

2. Use a clear and limited OneDrive folder

The instructions emphasize that the folder should be named clearly and should not be vaguely referred to, for example, as “My OneDrive.” The instructions also recommend separating files by task.

Practical tip:
Create a separate folder for each assessment, for example:

Course_ABC_Assignment2_Spring2026

Avoid folders that contain multiple course, group, or assignment submissions mixed together. This reduces the risk of Copilot accidentally processing the wrong material.

3. Explicitly limit the sources used by Copilot

Tell the agent the exact location or attach a prompt where to get the information, and specify what other sources are not allowed to be used. The example prompt directly says: “Do not use sources other than files in this folder.”

This is perhaps the most important point in terms of data protection.
It is always a good idea to include a clear source delimitation in the prompt:

Only use the OneDrive folder mentioned in this prompt and the student submissions attached to it. Do not retrieve, infer, or merge information from other files, emails, Teams conversations, calendar, or other sources in your organization.

This reduces the risk of feedback being mixed with other organizational data or previous student information.

4. Don’t ask Copilot to make the final evaluation decision

Copilot supports the generation of verbal feedback, but the teacher is responsible for the content and final use of the feedback. This way, the teacher maintains pedagogical control.

Practical tip:
Format the Copilot instruction so that it produces a draft of feedback, not a final grade or decision:

Prepare a feedback draft for the teacher to review. Do not make a final assessment decision or give a final grade without the teacher’s review.

This maintains teacher responsibility and reduces the risks associated with automated decision-making.

5. Always check feedback before uploading it to Moodle

Review and edit the verbal feedback provided by the Copilot agent and ensure that the feedback is consistent with the student’s work and the objectives of the assignment. The teacher is responsible for the quality, appropriateness, and final format of the feedback provided to the student.

Privacy Tip:
When checking, you should pay attention to at least these:

• Is the correct student work mentioned in the feedback?
• Is the feedback specific to that student’s submission only?
• Is there information included that was not in the student’s own return?
• Are there any references to other students in the feedback?
• Is the feedback factual, justified and tied to the evaluation rubric?

6. Make sure the feedback does not reveal other students’ information

Ask the agent to analyze the files one by one and generate verbal feedback based on each piece of work using the evaluation matrix.

Practical tip:
Add a separate privacy statement to the prompt:

Treat each student’s work as a separate entity. Do not compare students with each other. Do not mention other students’ names, feedback, strengths, weaknesses, or grades in the feedback for an individual student.

This is important because feedback should only be focused on the student’s own submission.

7. Communicate openly with students about the use of artificial intelligence

The “Tips and Dangers” section of the guidelines recommends openly telling students about the use of artificial intelligence to support feedback.

Practical tip to communicate to students:

The teacher can use Microsoft 365 Copilot to help draft verbal feedback. Copilot does not make the final assessment decision, but the teacher reviews, edits, and approves the feedback before giving it to the student.

This increases transparency and helps students understand what the role of AI is and what the role of the teacher is.

8. Do not use Copilot to process sensitive data without a specific justification

The guidelines discuss student feedback and its analysis using Copilot, but do not specifically address the handling of sensitive data. Therefore, as a precautionary measure, I recommend that teachers do not enter health information, personal life situations, learning support information, or other materials containing particularly sensitive information into Copilot unless there is a clear, approved basis for the processing and organizational guidelines.

Practical tip:
If student feedback includes personal experiences, diagnoses, health information, or other sensitive content, teachers should evaluate the handling separately before using Copilot.

9. Use the assessment matrix (rubric) to support data protection and fairness

An assessment matrix is ​​essential because it brings consistency, alignment, transparency, and pedagogical control to assessment.

From a data protection perspective, the matrix also helps to guide the Copilot to assess only the things that are relevant to the task.
So, in the prompt, it is worth saying, for example:

Base your feedback only on the criteria in the assessment matrix and the content of the student’s submitted work. Do not make any inferences about the student’s personal characteristics, motivation, life situation, or abilities.

This reduces the risk of unnecessary personal data inferences.

10. Only write verified and necessary feedback in Moodle

Finally, the revised feedback text is written for each student in the Moodle assignment feedback field and edited into its final form.

Practical tip:
Do not copy the entire analysis, intermediate steps, prompts or any internal comments from Copilot into Moodle. Only write the feedback intended for the student, reviewed and edited by you, into Moodle.

Proposal: data protection-secure prompt supplement for teachers

You could include a ready-made add-on like this in the instructions:

Privacy Policy: Use only the OneDrive folder mentioned in this assignment and the student feedback attached to it. Do not use other sources, such as emails, Teams conversations, calendar information, other OneDrive folders, or other documents in your organization. Treat each student’s work as a separate entity. Do not compare students with each other, and do not mention the names or information of other students in an individual student’s feedback. Base feedback only on the assessment matrix, the assignment, and the work submitted by the student in question. Do not draw conclusions about the student’s personal characteristics, life situation, health, motivation, or other matters irrelevant to the assessment. Prepare a draft of the feedback for the teacher to review. The teacher makes the final assessment, checks the accuracy of the feedback, and is responsible for the final content of the feedback given to the student.

A concise checklist for teachers

Data protection principle	Practical activity
Minimization	Only export to Copilot the submissions needed for evaluation.
Source delimitation	Tell us exactly which folder can and cannot be used.
Separate processing	Ask to analyze each student’s work individually.
No comparison	Do not let Copilot compare students with each other.
No sensitive conclusions	Refrain from drawing conclusions about the student’s personal characteristics.
Teacher’s responsibility	Review, edit and approve the feedback before exporting it to Moodle.
Transparency	Tell students that artificial intelligence is used to support feedback drafting.

In this way, Copilot supports the drafting of feedback, where sources are narrowed down, the assessment matrix guides the feedback, and the teacher is responsible for the final feedback.