Virtual Mentor

The method reduces the need for lecturers to check multiple drafts, while giving students constant access to feedback. The lecturer adds course materials, assessment criteria, examples of good work, checklists, and requirements to the tool. As a result, feedback is available immediately and helps students improve their work as they go. The tool acts as a quality filter: it identifies inconsistencies, gaps in argumentation, structural problems, and failure to meet the requirements.

The method can be used in several variants. In the simplest version, students use the Virtual Mentor only once, just before submitting their work, to check compliance with the requirements and identify formal or logical errors.

A more developmental variant is based on iterative work. Students prepare successive versions of the text, for example version 1, 2, and 3. After each version, they receive feedback, make deliberate improvements, and then briefly explain what they changed and why.

The tool can also be used as a quality checklist rather than a source of ready-made suggestions. In this version, students receive guiding questions that help them assess the coherence of their arguments, the structure of their work, and its alignment with the aim of the task.

In another variant, students compare their work with a well-assessed sample text, while the mentor points out differences in quality, logic, and structure. This helps students understand the expected standards.

The mentor can also be used in examiner mode, where the tool assesses the work according to the criteria and explains its decision. This allows students to see which elements affect the final grade and why they gained or lost marks.

Course: Engineering Design
Topic: Preparing a Research Report
The lecturer creates a Custom GPT called “Laboratory X Report Reviewer”. The bot’s knowledge base includes a model report, a list of the ten most common mistakes made by students in previous years, and a detailed assessment rubric.

Warm-Up (10 min)
The lecturer shares a link to the Custom GPT. As a backup, it is also worth sharing the instruction text, or system prompt, as a PDF file, so that students who reach the free usage limit can paste it into a standard chat.

Diagnostic Session (10–15 min)
Students upload the first draft of their report to GPT and give the following instruction: Analyse my report in terms of the coherence between the thesis and the conclusions, and check whether I have described the methodology correctly in line with the syllabus guidelines.

Iteration and Revision (20 min)

Students do not copy the bot’s response, but revise their own text files based on its suggestions. For example, if the bot writes, “Your Conclusions section is too general and does not refer to Table 2,” students must return to the data and independently add the missing explanations.

Final Review (10–15 min)
Students ask the bot to carry out a final review using the grading scale, for example from 2 to 5. They bring the revised version of the report to class, together with a short list of the errors the bot asked them to correct and an explanation of how they addressed them.

• For projects with a rigid structure and numerous formal requirements (e.g. business plans and technical reports).
• For situations in which you want to avoid assessing work that is clearly inconsistent with the assignment instructions.

• For creative or artistic tasks, where a template imposed by the bot could undermine originality.
• When working with sensitive data.

• The limits on free accounts may result in students losing access to a Custom GPT after only a few queries. As a backup plan, provide a text-based instruction, or System Prompt, which students can paste into a free chat.
• The bot may fail to notice factual errors hidden in the text. Emphasise that approval from the bot does not guarantee a high grade from the lecturer.

Easier → The bot only checks a checklist, for example whether all sections are included and whether the reference list is correctly formatted.
More Challenging → The bot acts as a sceptical reviewer: it identifies weaknesses in the student’s argument and requires them to defend their thesis.

• In the Custom GPT configuration, make this clear: You are an adviser, not a writer. Do not correct the text yourself, but point out errors and suggest how the student can fix them.
• Create a section where you can upload PDFs containing the syllabus and examples of good work.
• Encourage students to work in short sessions, for example: students submit one chapter, make corrections based on the feedback, and only then move on to the next one.

Formative Assessment
Ask students to submit a short change log, for example: What did the bot point out to me? What did I disagree with? What did I improve? This makes it possible to assess whether they are critically analysing the feedback.

Summative Assessment
Assess the final report. You can add bonus marks for reflection, such as a brief justification of why they decided to reject one of the AI’s suggestions. This demonstrates a deeper understanding of the topic and greater autonomy.