An “AI tutor” can mean anything from guided practice with feedback to open-ended chat that looks like tutoring but behaves like a general-purpose chatbot. In UK schools, the difference matters. Your procurement lens should start with the reality of day-to-day delivery: limited staff time, mixed device access, safeguarding obligations, and a curriculum that is assessed in specific ways. A platform can be impressive in a demo and still be the wrong fit once you test it against behaviour expectations, SEND needs, and what your DSL and DPO will sign off.
In 2025, what schools actually need is not a tool that talks well, but a system that is predictably safe and educationally purposeful. That usually means tight scope, clear learning goals, transparent feedback, and adult oversight. If you are still shaping your baseline expectations for safe use, it helps to refresh the boundaries you want across tools, not just tutors, using an AI acceptable use policy checklist. It also helps to avoid three common procurement traps: buying “autonomy” when you need “support”, assuming curriculum alignment because the marketing says “UK-ready”, and underestimating implementation workload (accounts, rostering, training, monitoring, parent comms).
What to avoid is equally clear. Be cautious of open chat experiences that allow students to ask anything, especially where the platform cannot show you how it moderates, logs, and escalates safeguarding concerns. Be cautious of “personalisation” that is really just endless practice without diagnostic insight. And be cautious of any vendor that cannot give you evidence, in writing, of how they handle data, content filtering, and age-appropriate design.
A single rubric makes your “bake-off” fair, evidence-based, and easier to defend to governors and auditors. The goal is not to find a perfect platform; it is to find the safest, most educationally valuable fit for your scenario.
Start with the learning model. Does the tutor use retrieval practice, worked examples, and feedback that explains errors? Can it prompt metacognition (“Why did you choose that method?”) rather than just giving answers? Ask for examples of how it responds to common misconceptions in maths and writing. If you want a tighter protocol for evaluating claims, borrow the habit of testing with a structured classroom-style script, similar to a rapid evaluation protocol, but applied to tutoring tasks.
“UK curriculum aligned” should mean more than spelling “colour” correctly. Ask what content map exists, what year-group statements are covered, and how gaps are handled. For secondary, ask how it supports exam-style questions and mark schemes without drifting into “here’s the answer”. For primary, ask how it supports language development and concrete representations, not just abstract explanations. If you need a compliance-minded way to check alignment, pair your rubric with a curriculum implementation checklist such as the National Curriculum AI implementation pack.
This is where many platforms look good in a mainstream demo and fail in reality. Test with pupils who use text-to-speech, need simplified language, require reduced cognitive load, or benefit from structured prompts. Check readability controls, dyslexia-friendly presentation, keyboard navigation, captions, and whether the tool can consistently provide chunked instructions. If you are building a coherent approach, it is worth comparing against your wider inclusion toolkit, such as the minimum viable inclusion stack.
Safeguarding is not just “does it block swear words”. You need to know how the platform detects self-harm, sexual content, radicalisation cues, and coercion, and what happens next. Does it block, redirect, notify staff, or simply warn the student? Are the thresholds configurable? Are logs accessible to DSLs, and are they usable in real time? Also ask how it prevents students from using the tutor to generate harmful content for others.
Your DPO will want clarity on roles (controller/processor), sub-processors, data locations, retention, and how training data is handled. “We don’t train on your data” needs to be in the contract, not a blog post. You also need to know what happens to chat logs, whether you can delete them, and how subject access requests are supported.
Procurement is smoother when the platform supports SSO, group-based permissions, and simple rostering. In practice, you need controls that match school reality: restricting features by year group, disabling free chat, limiting topics, setting time windows, and producing reports that staff can interpret quickly.
Licence cost is only one line. Total cost of ownership (TCO) includes devices, bandwidth, staff training, monitoring time, parent communications, and ongoing support. A “cheaper” platform can become costly if it increases admin burden or creates safeguarding workload.
Finally, ask what it takes to get from contract to classroom use. Who trains staff? What does a 30-day roll-out look like? What evidence can you gather for impact and safety? If you want a template for a tightly managed trial, adapt a 30-day pilot with guardrails to tutoring-specific measures.
Khanmigo is positioned as an AI-supported learning companion connected to Khan Academy-style content and practice. Its strengths are typically clearest where structured resources, step-by-step guidance, and constrained tasks matter. In a school setting, that can reduce the risk of the tool “wandering” into unsafe or irrelevant territory, because the learning context is more bounded than an open chatbot.
The limits, from a procurement lens, often show up in curriculum specificity and implementation detail. Even when content is strong, you still need to test whether it matches your sequencing, your methods, and the language you expect pupils to use. You also need to validate the safeguarding model for your use case: what happens when a pupil tries to steer it off-task, or discloses something concerning? The practical question is whether your staff can see enough, quickly enough, to intervene.
Best-fit scenarios tend to include structured intervention, homework support with clear boundaries, and small-group practice where staff want predictable prompts rather than open-ended tutoring. It can also fit well where you want to standardise practice routines across classes, provided you can evidence alignment with your schemes of work.
Synthesis is often associated with collaborative problem-solving and discussion-led learning experiences. From a pedagogy perspective, its strengths can sit in reasoning, talk-rich tasks, and structured debate or group challenge formats that encourage pupils to justify their thinking. That can be attractive where you want more than drill-and-practice and you want pupils to articulate strategies.
The limits are usually about fit and logistics. Collaborative models can be powerful, but they can be harder to timetable, harder to scale across cohorts, and more sensitive to group dynamics and behaviour. For safeguarding and oversight, you will want to understand how sessions are moderated, what staff visibility looks like, and how interactions are logged. You will also want to check whether the platform’s learning goals map cleanly onto your curriculum and assessment expectations, rather than feeling like an enrichment add-on.
Best-fit scenarios often include enrichment for high-attaining pupils, structured reasoning interventions, and programmes where oracy and problem-solving are explicit priorities. It may be less suitable as a universal “tutor replacement” model, and more suitable as a targeted programme with clear selection criteria.
A sensible bake-off includes at least one “platform tutor”, one “content-first practice system”, and one “school-controlled AI approach” that can be constrained by your own policies. In practice, schools often compare against tools such as subject-specific practice platforms with analytics, writing feedback tools with strong teacher controls, and (for MATs or larger schools) potentially self-hosted or tightly governed models where data protection is paramount. If you are considering the self-hosting route, the trade-offs are significant and worth reviewing through a decision pack like Meta Llama for education.
When you shortlist, prioritise vendors who will support a proper pilot: written answers to safeguarding and GDPR questions, a clear admin console demo, and a way to export evidence. Avoid vendors who insist on a “trust us” approach, or who cannot explain how their model behaves under pressure tests.
For many UK schools, the “yes/no” decision comes down to a small set of questions. Ask who the data controller is for pupil interactions, and whether the vendor is acting as a processor under your instructions. Ask for the list of sub-processors and where data is stored and processed. Ask whether any pupil content is used to train models, and ensure the contract states the answer clearly. Ask about retention defaults for chat logs and whether you can set deletion schedules. Ask how they support subject access requests and deletion requests without excessive manual work.
On safeguarding, ask how harmful content is detected, what the escalation pathway is, and what evidence is produced. If a pupil discloses self-harm ideation, what exactly happens on screen, and what exactly is sent to staff? If the answer is vague, treat that as a risk. Also ask how the platform prevents prompt-injection attempts that try to bypass safety rules, and how often safety systems are updated.
Ready to Revolutionise Your Teaching Experience?
Discover the power of Automated Education by joining out community of educators who are reclaiming their time whilst enriching their classrooms. With our intuitive platform, you can automate administrative tasks, personalise student learning, and engage with your class like never before.
Don’t let administrative tasks overshadow your passion for teaching. Sign up today and transform your educational environment with Automated Education.
When you compare pricing, insist on a whole-year view. Include licence costs, but also estimate device access (do pupils need 1:1 devices, headphones, or webcams?), connectivity, and staff time. Admin time is often the hidden cost: setting up accounts, managing groups, resetting passwords, reviewing dashboards, and responding to incidents. Training is another: a one-hour launch is rarely enough. You will need short, repeatable routines that staff can apply consistently, especially in primary. If you want a practical model for safe, small routines, adapt ideas from a teacher-in-the-loop primary playbook even if your context is mixed-phase.
Also consider support expectations. Does the vendor provide responsive safeguarding support? Is there an SLA? Can they support your procurement timeline and evidence needs?
Primary implementation often succeeds when the AI tutor is tightly constrained: short sessions, clear prompts, and adult circulation. You are looking for predictable behaviour, strong accessibility features, and minimal temptation to “chat for fun”. Secondary implementation can handle more independence, but the stakes around academic integrity rise. If the tutor supports writing or extended responses, you need clear boundaries and scripts for staff and pupils so it does not become an answer generator. Even if your focus is tutoring, it is worth aligning with wider integrity guidance such as exam-season AI boundaries.
For SEND cohorts, the best fit is usually the platform that offers consistent structure, adjustable language complexity, and clear multimodal supports without overwhelming the pupil. For intervention models, decide whether the tutor is for pre-teaching, same-day catch-up, or consolidation. For home use, tighten safeguards further: clarify parent visibility, time windows, and what happens outside school hours.
A strong pilot is short, bounded, and evidence-driven. In week one, run a staff briefing, configure controls, and test safeguarding with scripted scenarios before any pupil use. In week two, start with a small cohort and a single subject area, keeping sessions brief and supervised. In week three, expand cautiously to a second cohort or subject, and begin comparing outcomes against your usual intervention approach. In week four, consolidate: gather staff feedback, pupil voice, incident logs, and learning data, then make a stop/go decision with your senior team, DSL and DPO.
Success measures should include learning indicators (for example, improved accuracy on a defined question set, or fewer repeated misconceptions), operational indicators (time spent on admin and monitoring), and safety indicators (number and severity of flagged interactions, and how quickly staff could respond). Stop/go criteria should be explicit in advance: any unmanageable safeguarding risk, inability to meet data protection requirements, or unacceptable workload increase should trigger a stop, even if attainment looks promising.
Best-fit recommendations are easiest when tied to scenarios. If you need a structured practice tutor for intervention with clear boundaries, prioritise platforms that keep the interaction tightly linked to curriculum tasks and provide usable teacher reporting. If you want reasoning, discussion and enrichment, prioritise platforms designed for collaborative problem-solving with strong moderation and clear learning objectives. If your primary driver is data control and you have capacity, a constrained, school-governed approach can be viable, but only if you can resource implementation and oversight.
For procurement, keep your checklist practical. Confirm safeguarding workflows, logging, and escalation routes. Confirm UK GDPR roles, retention, deletion, and sub-processors. Confirm admin controls, rostering, and reporting. Confirm accessibility features with real pupils, not just vendor claims. Confirm TCO with staff time included. Finally, confirm implementation support: training, documentation, and what happens when something goes wrong.
May your next procurement decision be calm, evidence-led, and genuinely safer for pupils.
The Automated Education Team
Black Friday can feel like a rare chance to “save” on AI subscriptions, but …
Three years after ChatGPT’s release, schools have enough experience to …
December in schools brings a familiar spike: cover changes, heightened …
Microsoft Ignite can feel like a firehose of AI updates, but schools need a …
Report writing in 2025 is less about “which chatbot is best” and more about …
Three years into LGR22, many schools report real gains in clarity and …
Anti-Bullying Week works best when it moves beyond awareness and into …
Remembrance teaching asks for careful language, accurate sources, and …
Mock season often fails for predictable reasons: revision plans are …
Ühendkuningriigi koolidele pakutakse kiiresti „AI juhendajaid“, kuid 2025. aasta hankepädevad …
Brittiska skolor erbjuds “AI-handledare” i snabb takt, men upphandlingsbeslut 2025 kräver mer än …
Ison-Britannian kouluille tarjotaan “AI-tuutoreita” kovaa vauhtia, mutta vuoden 2025 …