A history teacher watches an AI assistant label a source “reliable” before a pupil has explained what it can tell them. The interface works. The educational sequence is wrong.
The teacher's most useful contribution is not another rating of the product. It is a design question: could the tool help pupils examine evidence without supplying the judgment they are learning to make?
Teachers should help shape the AI tools their students use when their expertise can influence the learning purpose, the interaction and the evidence the tool makes available. Meaningful co-design gives that contribution a route into a prototype, a decision and a further examination. An advisory title or a consultation meeting does not establish that it happened.
This does not mean every teacher needs to become a software developer. It means classroom knowledge should help determine what educational software does.
What teacher co-design means
For this article, co-design means teachers participating in the development of a tool while relevant design decisions remain open. They help identify a problem, explore possible responses and examine how a proposed solution behaves.
That differs from choosing among finished products. It also differs from adapting a lesson around an existing application. Both are important professional activities, but neither necessarily changes the product.
A teacher may influence the timing of a hint, the information presented to a learner, the controls available to staff or the circumstances in which the tool should stop and seek human help. Developers contribute technical knowledge about feasibility and failure. Learners contribute experiences adults may overlook.
The useful question is specific: which part of this product could change because these people participated?
What the evidence supports—and what it does not
The OECD's Digital Education Outlook 2026, published on 19 January 2026, recommends educational AI developed with teachers and learners and supported by research on effectiveness. Its public overview places co-creation alongside pedagogical purpose and human agency. That is a direction for development, rather than proof that a co-designed product necessarily improves learning.
England's government made collaboration with teachers part of its 16 April 2026 announcement inviting companies to develop AI tutoring tools for Years 9–10. The announcement described curriculum alignment, safety and testing, with national availability of successful tools intended from 2027. These were programme intentions; the announcement does not establish delivery or results by October 2026.
A smaller empirical study offers a different kind of evidence. Alfredo and colleagues' December 2025 study examined co-design of TeamVision, an AI-powered learning analytics system, with five nursing students and three teachers at an Australian university. Across six sessions, the qualitative analysis identified tensions involving learning goals, privacy and usefulness, and preferences for human or AI guidance.
The study reveals design disagreements worth investigating. Its small, specialised higher-education setting does not establish effects on school pupils or prove that participation causes better outcomes.
Together, these sources support a serious design conversation. They do not provide a universal recipe. The approach below is an original planning proposal.
Start with a difficulty the classroom can make visible
Imagine a fictional secondary history department helping to develop a source-comparison assistant.
Pupils are examining two accounts of the same event. The teaching aim is to explain what each source can support, considering its origin, purpose and relationship to other evidence.
An early prototype offers a quick credibility verdict. A teacher notices that pupils could copy the verdict without explaining the source's usefulness for the historical question.
“Make it more educational” would leave the developer guessing. The teacher instead identifies the missing intellectual move: pupils need to connect a claim with evidence and recognise what the source cannot establish.
That description changes the design problem. The team is no longer simply improving the fluency of an answer. It is deciding how the interaction can make the pupil's reasoning possible and visible.
The scenario is illustrative, not a reported product trial or Learning Coach engagement.
Turn the difficulty into a requirement
A useful requirement explains the desired behaviour clearly enough that another person can examine it.
In the fictional case, the team might propose that the assistant display the selected sources and invite a pupil to identify evidence for a claim before offering an optional question. It should not begin by issuing a global verdict that makes the source judgment for them.
The teacher can explain why: a source may be useful for one question and limited for another. A simple trustworthy/untrustworthy label can conceal that distinction.
Developers can then raise practical questions. What happens when the pupil enters no response? Can a teacher alter the level of support? How will the interface distinguish a quotation from the pupil's explanation? What will it do when the supplied material is insufficient?
The requirement will need revision. That is part of the work.
Avoid turning one suggested sequence into a rigid rule for every learner. Prior teaching, accessibility arrangements and the purpose of the activity may justify different support.
Use a prototype to expose misunderstanding
The first useful prototype need not contain functioning AI.
Paper cards can represent the source display, pupil response and optional support. A teacher and designer can walk through the interaction with prepared fictional material. This makes it easier to change the sequence before investing in a polished interface.
Suppose the designer interprets “evidence” as a sentence the system should retrieve and present. The teacher intended pupils to select and explain it. A paper rehearsal exposes the difference.
The next version might separate the source from the pupil's claim more clearly and ask a bounded question about their connection. Another rehearsal can reveal whether the change addresses the original difficulty or merely adds another screen.
A prototype is an object for discussion. Its existence does not establish safe classroom use, and a successful walkthrough is not evidence that students have learned.
Include students without asking them to do the teacher's job
Teachers understand curricular purpose. Students know what it feels like to encounter the interaction.
A pupil might interpret an optional hint as a required step. Another may think the assistant's tentative suggestion is an authoritative correction. An accessible design may require a different response format or clearer navigation.
Those experiences matter even when pupils cannot formulate a technical requirement. Ask them to show what they believe the tool is asking them to do. Compare that understanding with the intended task.
Participation should be appropriate to age, context and institutional arrangements. Start with representations or suitable fictional material where that can answer the design question. Involvement in a design discussion is not blanket permission to process pupil information or deploy an unfinished system.
Adults remain responsible for the learning purpose and the conditions of participation. Student input should inform those decisions without transferring responsibility to children.
Keep a record of influence
A list of people consulted is a record of participation. A record of influence connects their contribution with what happened next.
For the fictional history tool, that record could contain the initial difficulty, the proposed requirement, the prototype response and what a later examination found. If the team rejects the requirement, it should explain why.
Perhaps the requested behaviour cannot be implemented reliably. Perhaps the change creates an accessibility problem. Perhaps the discussion reveals a different teaching objective. Those are useful answers when the reasoning is visible.
Some matters will remain outside the group's authority. State those boundaries at the outset. Teachers should know whether they can influence a feature, a setting, the learning sequence or only the wording.
Calling a finished decision “co-design” makes participation hard to interpret. Be honest about what is still open.
Give disagreement a useful form
Teachers will not always agree with one another.
One may want earlier support for a pupil who cannot begin. Another may worry that the same support reveals too much. Both may be reasoning carefully from different learners or objectives.
Bring the disagreement back to a concrete task. What knowledge is assumed? Which difficulty is the pupil experiencing? What should the support make possible? What should remain theirs to decide?
A configurable option may be appropriate, but settings also create work. Someone must understand what each option changes and choose it responsibly. More controls do not automatically produce more meaningful professional agency.
The design team needs to examine that burden alongside the proposed benefit.
Do not make every teacher build the tool
Co-design needs educational expertise and technical expertise. It should not quietly assign software maintenance, security testing or model evaluation to teachers who were invited to explain classroom needs.
Schools can contribute through a subject team, a network partnership or a representative group with opportunities for wider colleagues to respond. Different subjects, learner needs and levels of experience deserve attention; enthusiasm alone is an insufficient selection criterion.
Provide time for preparation, examination and follow-up. A rushed meeting after a full teaching day may produce comments but little opportunity to explore alternatives.
Explain how contributions will be used and acknowledged. Participation in development should not automatically become a public endorsement. Any later claims about teachers' involvement should accurately describe their contribution.
Evaluate the product as well as the process
A team can run a respectful co-design process and still produce a tool that does not work well enough.
Ask separate questions. Did participants influence the design? Does the resulting interaction behave as intended? Is it suitable for its proposed context? Does its use support the learning claim?
Each needs appropriate evidence. A teacher's satisfaction does not establish reliability. A more usable interface does not establish improved understanding. Students' completed responses do not, by themselves, show what they can do independently later.
In the history example, the team could first examine whether pupils understand the revised sequence. A later, appropriately designed evaluation would be needed to investigate learning. It should allow for disappointing findings and further revision.
Our guide to turning AI policy into classroom practice addresses the broader arrangements a school needs around an approved use.
Make an education event a working conversation
A school-network event can help teachers, leaders and developers practise this translation.
Give participants a fictional learning problem and a simple prototype. Ask them to identify the learner's task, propose one change and explain what observation would help judge the revision. Discuss where their recommendations conflict.
A keynote can establish why classroom expertise belongs upstream in design. A working session can examine requirements and alternatives. Neither format substitutes for product assurance or a credible evaluation.
For organisers, our guide to choosing an AI in education keynote speaker helps connect the event with a clear educational purpose.
The sources reviewed here comprise an OECD publication overview, a government announcement and selected sections of a small qualitative study. They are not a systematic review of co-design effectiveness. The fictional history example and suggested workflow require adaptation and examination.
Learning Coach's position is that teacher participation becomes educationally valuable when it helps preserve the learner's thinking in the product's actual behaviour. The test is concrete: can the team show what changed, why it changed and what still needs investigating?
Planning a school leadership or education innovation event about teachers' role in shaping AI? Explore Learning Coach's international keynotes and request a tailored proposal.
References
OECD, Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education. Published 19 January 2026; official overview and key messages.
UK government, Edtech and AI companies invited to help build safe AI tutoring tools for disadvantaged pupils. Announcement, 16 April 2026; programme intentions, not measured outcomes.
Alfredo, R., and colleagues. Co-designing AI-powered learning analytics: bringing students and teachers together. Published 23 December 2025. Qualitative nursing-education co-design study; context and generalisability limits discussed above. Sources consulted 29 October 2026.
