AI for training providers and tutoring businesses: what to automate first in 2026

The "what stays human" column follows Ofqual, Ofsted and DfE guidance and the ICO's Children's Code (sources 1 to 9).

AI saves the most tutor time on drafting resources, and it can check work against a mark scheme, but a person decides every mark.

Lesson and resource drafts

In a trial for the Education Endowment Foundation (EEF), evaluated by NFER and reported by Schools Week on 12 December 2024, 259 teachers in 68 secondary schools planned Year 7 and 8 science lessons. Teachers who used ChatGPT with a guide spent 56.2 minutes a week on planning, against 81.5 minutes for those asked not to use generative AI, a cut of 31%. Reviewers found no noticeable difference in quality, though Schools Week says that finding "should be treated with caution". DfE guidance says "Any content produced requires critical judgement to check for appropriateness and accuracy", so a tutor checks every resource before a learner sees it.

Marking support with a person grading

Ofqual's policy paper, updated 16 July 2026, says "AI may not be used as a sole marker in any assessments that are part of regulated qualifications". It recognises "the potential for AI to support marking of lower-stakes and classroom-based assessment", and says marking decisions must be "accountable, capable of being explained and, where necessary, challenged". Ofqual's January 2026 working paper adds that "Current evidence does not support an overall or general case for the validity of AI marking in high stakes qualifications".

A marking support build compares each piece of work with the mark scheme, quotes the passage that meets or misses each criterion and drafts feedback. The assessor reads the work, sets the mark and edits the feedback. For regulated qualifications, follow the awarding organisation's rules first, because Ofqual expects awarding organisations to show that any use of AI "delivers valid, reliable and fair outcomes".

Learner use of AI

Ofqual's March 2026 briefing for schools says "AI-generated work submitted as the student's own is malpractice". It tells teachers who suspect AI-generated content to talk to the student and then follow the malpractice procedure, and points to JCQ's guidance on AI use in assessments for detail. A build can flag work for a tutor to look at. It never accuses a learner, and a person makes every malpractice decision.

Safeguarding stays with named people. An AI can route a concern to the right person faster, but it never judges one.

The DfE's safeguarding guidance for out-of-school settings, last updated on 9 February 2026, covers private tuition, music lessons and sports training. It sets out policies for safeguarding and child protection, safe and suitable people, and DBS checks for self-employed providers and volunteers. An AI build fits inside those policies.

• Any message that mentions harm, abuse, self-harm or a child in distress goes straight to the safeguarding lead. The AI sends only a short holding reply.
• The AI never holds private one-to-one conversations with a child outside what the safeguarding policy allows.
• The build logs every conversation, so the safeguarding lead can review it.

DfE guidance says "It is recommended that personal data is not used in generative AI tools", and warns that many free tools use inputs to train their models. Use paid tiers whose terms keep data out of training. For learner-facing tools, the DfE's "Generative AI: product safety standards", updated 19 January 2026, are for edtech developers and suppliers to schools and colleges, and they make a useful checklist for any provider building its own tool.

What to keep human

• Every mark and grade, and every malpractice decision
• Every safeguarding concern, and any contact with a child outside the safeguarding policy
• Advice on which course or level suits a learner
• Refunds, withdrawals and extensions
• Every submission to an awarding body or funder
• Messages to parents about a child's progress or welfare

Yarigai is a done-for-you AI agency in London. It builds lead capture and booking flows, support agents that answer from a client's own documents, inbox triage, internal tools that replace spreadsheets and iOS, Android and web apps. For a training provider, it builds them inside the provider's own website, email, booking system and learning platform. The provider owns the code, the data and the accounts from day one.

A project starts with a free 30-minute scope call on Google Meet, and the provider keeps the notes whether it hires Yarigai or not. A blueprint of about a week, credited to the build, sets out a written plan, a clickable prototype, the monthly running cost, a fixed price and the payback maths. The build takes 2 to 6 weeks with a demo of working software every week. A single workflow, such as enquiry replies and booking, can go live in about two weeks.

A person approves anything that sends money, emails a customer or deletes data. AI and hosting bills are paid by the provider at cost. Yarigai's guarantee is one sentence: "If the build isn't profitable for you, we won't build it."

Yarigai has no published training provider case study. Its lead capture and booking work and support agents are described on the services pages.