How we use AI, and where it falls short
AI does a lot of work on this platform. Pretending otherwise would be dishonest, and pretending it is reliable would be worse. This page sets out exactly where a model is involved and what it gets wrong.
Next review due .
The one-sentence version
A language model drafts most of the study material and answers every tutor and writing-feedback request live; the examination, pricing and policy pages are written by people; and no subject specialist has yet signed off the AI-drafted material, which is why you should check anything that matters against your own syllabus and teacher.
Where AI is used
Practice questions and their explanations
Generated into the platform database with AI assistance, then served as stored. A learner sees exactly the text that was generated.
Skill and subject study guides
Drafted from the skill or subject’s own questions, topics and levels. The model used is recorded against each guide, so any guide can be traced back to how it was made.
Blog articles
Drafted against a planned editorial calendar. Platform figures quoted in an article are read from the database rather than written by the model.
AI tutor
Generates a reply live, in response to what the learner types. Nothing is pre-written and nobody reviews it first.
Writing feedback
Reads a learner’s writing and produces comments and suggestions live. Same conditions as the tutor: generated on demand, unreviewed.
Where AI is not used
- Examination pages. These are written by hand and stored in the code.
- Pricing, privacy, terms and every page in this trust section.
- Curriculum and examination notes. Those come from a fixed mapping keyed on education level — see the curriculum alignment statement.
- Counts and statistics. Those are read from the database at request time.
- Marking of practice answers, which is a straight comparison against the stored answer.
The limitations, stated plainly
These are the failure modes we know about. Listing them is not a disclaimer; it is what you should watch for.
- Confident errors. A model will state something false in the same tone it uses for something true. Fluency is not accuracy.
- Level drift. An explanation may be correct but pitched above or below the level a learner is actually working at, or use a method their class has not been taught.
- Stale curriculum knowledge. A model’s training data has a cut-off. It will not know that a syllabus has been revised.
- Local context. Generic material can miss the terminology, units, currency, place names and conventions used in Ghanaian classrooms.
- Arithmetic and edge cases. Multi-step calculation, unusual wording and questions with more than one defensible answer are where errors cluster.
- Invented specifics. Models fabricate references, dates and citations. This is the reason the editorial policy forbids publishing any named expert, partner or result that has not been verified by a person.
What we do to limit the damage
- Guides are drafted from the platform’s own questions and topics, not from open-ended prompting, so a guide cannot describe content the platform does not have.
- Curriculum labels, counts and examination references are taken out of the model’s hands entirely.
- Every AI-drafted page is labelled as such, on the page, where a reader will see it.
- The model used to draft a guide is recorded, so a systematic fault can be traced and the affected guides re-generated.
- Reader reports are the main way errors are found today, which is why the corrections process is open to anyone and needs no account.
How to use the AI features sensibly
- Use the tutor to get unstuck, then check the method against your exercise book or your teacher.
- If an explanation contradicts what your class was taught, your class is the one to follow.
- Treat writing feedback as a second opinion on structure and clarity, not as a mark.
- Report anything that looks wrong. It takes a minute and it fixes the page for everyone.
Questions people ask about AI here
Is the content on DigiTransact Mastermind written by AI?
Much of it is. Skill guides, subject guides, blog articles and the practice questions with their explanations are drafted with AI assistance from the platform database. Examination pages, pricing and the policy and trust pages are written by hand. Every page states which of the two it is.
Can the AI tutor be wrong?
Yes. The tutor generates its reply live and nobody reads it before the learner does. It can state something false with complete confidence, misread a question, or give a method that is correct in general but wrong for the level being studied. Treat it as a study aid, not as an authority.
Is my work used to train AI models?
What a learner types is sent to the model provider so a reply can be generated, and it is handled under the privacy policy. We do not sell learner work and we do not publish it. Details of how data is handled are on the privacy policy page.