Dolly vs ChatGPT for marking and feedback

ChatGPT can explain a topic, draft a worksheet, adapt a text for a different reading age or suggest feedback on an answer you paste in. For one-off help in any subject it is quick and flexible, and Dolly is not trying to replace it.

To mark with ChatGPT you paste in the question, the mark scheme and each student’s answer, then copy the feedback back out. The result depends on how the prompt is written, and nothing is kept together for the class: there is no record of who handed in, no view of gaps across the group and nothing to build reports from.

A general assistant answers the prompt it is given, and checking the answer is up to you. Dolly runs each task through a workflow built for teaching, with checks on what the model writes before a teacher sees it.

• Planning, then writing: a subject-specific agent plans what each question should assess against the specification, and a separate agent writes it.
• Every extended-response question is checked against a schema for its subject and for things a student could not answer, such as a reference to a diagram that is not there. A question that fails is sent back to be rewritten.
• Every quiz question is checked for its structure, that the marked answer is among the options, its maths, that the explanation matches the answer and that no two options repeat. A quiz with too many errors is rejected rather than set.
• Each mark is checked against the mark scheme’s structure before it is saved, with a confidence score for the teacher.
• Nothing reaches students until the teacher has checked it and published it.

• Topics chosen from your exam board specification, or your own upload, with subject-specific agents researching each one.
• A mark scheme written with every extended-response question, using levels of response or points.
• Students submitting in their own app, and the whole class marked against the same scheme.
• A confidence score on every mark, and nothing sent to students until you publish it.
• Class gaps, personalised flashcards from each student’s gaps, and reports for the whole class.

The Department for Education’s guidance on generative AI in education (updated 12 August 2025) asks schools to check AI output before relying on it, to keep personal data out of generative AI tools where they can, not to let pupils’ original work train AI models without permission, and to follow each tool’s age limits.

Dolly stores your data on AWS in London and complies with UK GDPR. Writing and marking run on Anthropic’s Claude through Amazon Bedrock in the EU, and Bedrock does not use your content to train models or share it with model providers. OpenAI’s terms allow ChatGPT users from 13, with a parent’s or guardian’s permission under 18.

Use ChatGPT for general help with planning and resources. Use Dolly when you want a class’s work set from your specification, marked against a mark scheme, checked by you and turned into follow-up and reports.