Generative AI Canvas
A thinking tool for turning broad “what if?” ideas into concrete, structured opportunities.
Quick and flexible, it can guide a conversation around a real challenge or opportunity, help you consider whether and where AI might contribute, and move you towards a practical next step. No AI expertise is required.
Choose the version that best suits what you want to explore. Blank Canvases can be completed on screen or printed.
Choose a Canvas
What challenge are you facing, or what opportunity are you exploring? Why does it matter?
Start with the work itself: something that takes too much time or effort, is difficult to do well, or something you would like to make possible.
How does this work today? Where does it work well, and where are the difficulties?
Describe what actually happens now: who is involved, what they do, what information they use, and where the most time, effort or judgement is needed.
What information, data, context or expertise would be needed for AI to help here?
Identify what AI would need to draw on, including people's knowledge, documents, data, systems or other resources. Is it available, reliable, appropriate and accessible in a form you can actually use?
Where could AI help, and what would it need to do?
Focus first on what you need AI to do. Could it analyse, compare, generate, organise, explain, translate or automate something? Then consider which approach or tool might fit, and why.
If this worked well, what would improve?
Look at what would change in the work. Could this improve quality, save time, reduce errors, make information easier to use, or help people make better decisions?
What could go wrong, and what would need particular care?
Consider sensitive information, accuracy, bias, fairness, transparency and relevant professional standards. Be clear about what needs checking, where human judgement matters most, and who is responsible for the final decision or output.
How would you know whether this was genuinely useful?
Decide what you would look for before testing it. This might include quality, accuracy, time, feedback from the people involved, or whether the result is useful enough to keep developing.
What is the smallest useful thing you could do next?
Choose one small action: test an assumption, try a small example, speak to someone, gather the right information or run a low-risk experiment. Aim to learn something useful without trying to solve everything at once.
If this worked, what would you have more time or headspace for? What would you want to do more of – or worry about less?
What challenge are you facing, or what opportunity are you exploring? Why does it matter?
The Contracts & Governance inbox receives a high volume of enquiries from colleagues. Many repeat the same questions about templates, approval steps or policy interpretation, but still need careful, accurate responses.
This takes time away from more complex or strategic work.
How does this work today? Where does it work well, and where are the difficulties?
Staff read each enquiry, search internal guidance, templates and previous correspondence, then draft a response.
The process is accurate but slow, and similar questions are answered repeatedly.
What information, data, context or expertise would be needed for AI to help here?
- Anonymised historical enquiries.
- Contracts and governance policies, templates and guidance.
- Examples of previous high-quality responses.
Where could AI help, and what would it need to do?
AI could group recurring enquiries, suggest FAQ topics and draft response templates, and match questions to existing policies or guidance.
It could also flag areas where guidance appears unclear or incomplete. Staff would check and refine every output.
If this worked well, what would improve?
- Faster handling of routine enquiries.
- More consistent tone and advice.
- Better visibility of recurring questions and pressure points.
- Less duplicated effort across the team.
What could go wrong, and what would need particular care?
All enquiries would need to be anonymised before use.
Staff would need to review AI-generated material, especially where policy interpretation is involved, to catch errors or invented guidance.
How would you know whether this was genuinely useful?
- Recurring categories identified across a substantial share of enquiries.
- A validated FAQ and response-template library that staff use.
- Shorter turnaround times for routine questions.
- Positive feedback from colleagues on clarity and consistency.
What is the smallest useful thing you could do next?
Run a small pilot with a limited anonymised set of historical enquiries. Use AI to identify recurring themes and draft an initial FAQ set, then review the outputs internally and test them with colleagues.
If this worked, what would you have more time or headspace for? What would you want to do more of – or worry about less?
- Less time spent re-explaining the same processes.
- Better visibility of where colleagues most often need help.
- More confidence that routine questions are handled consistently.
- More time for complex, strategic work.
What specific research question, gap, or hypothesis are we addressing, and why does it matter to our field?
Think about the knowledge, method, or application you want to advance, and why this is timely or significant.
Which part of the research lifecycle do we want to explore – and how do we currently work in that area?
Choose a part such as literature review, study design, data collection, analysis, writing or dissemination. Note the methods, information or tools involved.
Where do we run into difficulties, limitations or unnecessary effort in this part of the research?
Consider where the work is repetitive, time-consuming, difficult to do consistently, or where existing methods or processes may limit rigour, progress or reproducibility.
Where could AI help with the challenges we identified, and what would it need to do?
Could it find or compare information, explore ideas or alternatives, prepare or analyse data, work with qualitative material, develop or review outputs, or communicate research for different audiences?
What small, testable use of AI could we try?
Sketch what AI would need to draw on, how you would test the idea, and what a useful output or result would look like. Aim to learn something without trying to build a complete workflow.
What ethical, disciplinary or integrity issues should we keep in mind?
Think about privacy, accuracy, bias, authorship and transparency. What would need checking, documenting or human judgement? Are there approvals or requirements that should shape the test?
What wider priorities or commitments should shape this use of AI?
Look at funder expectations, disciplinary standards, institutional and wider strategic priorities, and open research principles. Which are relevant here?
How would we know whether using AI here was genuinely worthwhile?
Decide what you would look for before running the test. This might include research quality or rigour, reproducibility, time or effort, the usefulness of the outputs, or feedback from the people involved.
What is the next useful step?
Choose one practical action to test or refine the idea. You might gather the right material, try it on a small scale, compare approaches, or involve someone whose expertise or perspective would help.
After testing, what did we learn about the research or the way we work?
What surprised you? Did the test reveal assumptions, limitations, gaps or new possibilities?
What specific research question, gap, or hypothesis are we addressing, and why does it matter to our field?
How can we make complex research on topics such as marine plastics, renewable energy or health innovation more accessible and engaging to local communities?
Which part of the research lifecycle do we want to explore – and how do we currently work in that area?
The focus is dissemination and engagement.
Researchers often prepare text-heavy reports or academic presentations. These may be shared with community stakeholders but can be difficult for non-specialist audiences to understand or use.
Where do we run into difficulties, limitations or unnecessary effort in this part of the research?
- Limited time to prepare tailored materials.
- Difficulty simplifying technical language without losing accuracy.
- Inconsistent outreach formats across research teams.
Where could AI help with the challenges we identified, and what would it need to do?
AI could help draft plain-language summaries, suggest content for visual explainers and generate FAQs or key messages for different audiences.
What small, testable use of AI could we try?
Use one completed study on microplastics in coastal waters as a small test.
Give AI the research material and ask for a 300-word plain-English summary plus three key messages for a community workshop. Researchers would review the outputs for accuracy, nuance and suitability before use.
What ethical, disciplinary or integrity issues should we keep in mind?
Outputs need to be fact-checked to avoid oversimplification or misrepresentation.
Researchers remain responsible for accuracy and for preserving nuance before anything is shared.
What wider priorities or commitments should shape this use of AI?
AI could help identify connections with relevant regional priorities, funder strategies or wider frameworks such as the Sustainable Development Goals. In this case, likely areas include civic and community engagement, sustainable communities, public participation in science and making research more accessible beyond academia.
How would we know whether using AI here was genuinely worthwhile?
- Positive feedback from community partners on clarity and relevance.
- Technical content simplified without losing accuracy.
- More useful two-way discussion between researchers and communities.
What is the next useful step?
Trial the approach with one project. Test the materials with community partners, refine them in response to feedback and decide whether the approach is useful enough to repeat elsewhere.
After testing, what did we learn about the research or the way we work?
The test might reveal where researchers assume too much prior knowledge, where nuance is lost and which formats make discussion easier.
Which research question, societal challenge or domain-specific problem are we tackling? Why does it matter now?
Consider how different disciplines, communities or stakeholders might understand or frame the issue.
How do we currently approach this challenge, and what does each discipline or stakeholder contribute?
Describe the knowledge, methods, data, experience or perspectives involved. Where do these contributions already connect, and where are they still separate or in tension?
What makes it difficult to work together effectively?
Think about differences in terminology, methods, data, priorities, expectations or ways of working. Where is unnecessary effort being created, understanding breaking down, or useful connections being missed?
Where could AI help people work across these differences, and what would it need to do?
Could it compare perspectives, connect information or datasets, bridge differences in terminology, explore scenarios, surface patterns or disagreements, or make complex ideas easier to work with?
What small, testable use of AI could we try?
Sketch what AI would need to draw on, whose knowledge or perspectives need to be represented, how you would test the idea, and what a useful output or result would look like. Aim to learn something without trying to build a complete workflow.
What ethical, social or disciplinary issues should shape how we test this?
Think about sensitive information, bias, uncertainty and community trust. Whose knowledge, interests or experiences could be oversimplified, excluded or misrepresented? Where are transparency or human judgement particularly important?
How does this connect with wider priorities, stakeholder needs or public value?
Look at relevant strategic or funder priorities, the people, communities or sectors affected, and any wider social or environmental commitments. Who could benefit, and whose needs or interests might otherwise be overlooked?
How would we know whether using AI here was genuinely worthwhile?
Decide what you would look for before running the test. This might include better integration of different contributions, clearer shared understanding, reduced friction, stronger outputs or decisions, time or effort, or feedback from the people involved.
What is the next useful step?
Choose one practical action to test or refine the idea. You might gather material from another discipline, involve someone currently missing from the process, compare approaches, or try the idea on a small scale.
After testing, what did we learn about the challenge or the way we work together?
What surprised you? Did the test reveal assumptions or tensions, expose gaps in knowledge or skills, or suggest new possibilities for collaboration?
Which research question, societal challenge or domain-specific problem are we tackling? Why does it matter now?
How can we support sustainable coastal tourism while protecting marine biodiversity?
How do we currently approach this challenge, and what does each discipline or stakeholder contribute?
Ecologists contribute survey data and biodiversity expertise. Tourism researchers contribute tourism planning and impact models.
These are often developed separately, so ecological evidence can remain disconnected from tourism planning and have limited influence on decisions.
What makes it difficult to work together effectively?
- Different disciplinary languages and priorities.
- Ecological and tourism datasets that are difficult to align.
- Policymakers receive fragmented or incomplete evidence.
Where could AI help people work across these differences, and what would it need to do?
AI could connect ecological survey data with tourism impact models, explore development scenarios and summarise the trade-offs in policy-relevant language.
What small, testable use of AI could we try?
Use one coastal site and a small set of ecological and tourism data.
Test whether AI can generate a few tourism-development scenarios that include ecological safeguards and summarise the trade-offs in plain language. Ecologists and tourism researchers would review the outputs together.
What ethical, social or disciplinary issues should shape how we test this?
Scenarios could oversimplify complex systems or give too much weight to assumptions from one discipline.
Domain experts should review the outputs, and community and industry perspectives need to be included before conclusions are drawn.
How does this connect with wider priorities, stakeholder needs or public value?
AI could help map the work against relevant regional strategies, funding priorities and wider frameworks such as the Sustainable Development Goals. Likely areas include the maritime economy, sustainable communities, biodiversity protection and balancing environmental and economic priorities.
How would we know whether using AI here was genuinely worthwhile?
- Ecological and tourism evidence can be brought together more clearly.
- Clearer shared understanding of the trade-offs across disciplines.
- A stronger evidence base for regional decision-making.
- Ecological protection and economic development considered together rather than separately.
What is the next useful step?
Pilot the approach at one coastal site using a limited set of ecological and tourism data. Review the outputs with researchers and relevant stakeholders, then refine the test in response to what they find.
After testing, what did we learn about the challenge or the way we work together?
The test might expose assumptions within each discipline, gaps between the datasets and points where shared scenarios help researchers find common ground.
Using the Canvas
Choose one thing you want to explore and work through the questions on your own, with a colleague or as a group.
Start with how the work happens now, rather than with an AI tool or solution. Be specific about the information, expertise and judgement involved. If a possible use of AI emerges, define a small test and decide what you would need to see before taking it further.
These are the same Canvases I use in training and workshops. I’m sharing them here so the framework is available beyond those sessions.
What people take from the Canvas
“Discussing the different possibilities and in my head seeing how I could apply them to my research, teaching and life.”
“Using the canvas tool helped solidify learning and improve confidence.”
Start a conversation
The best way to explore a potential project is simply to talk it through. We can discuss what you have in mind, where AI could make a useful difference and how I might help.