Create a Theory of Change
Draft a theory of change that explains the program pathway, tests its assumptions, and identifies practical measures and risks.
On this page
- What You Need
- Share the program objective, population, main activities, intended outcome and relevant evidence. Add context or an existing logic model if available.
- Choose This Prompt When
- Use this to explain why and how change is expected. Choose Create a Logframe for a results table, or Draft a Theory of Change from a Program Narrative when the main input is an existing narrative.
Develop a practical theory of change for the program described in my material. Use its objective, population, activities, intended outcomes and evidence. Keep the pathway within that scope.
First check that the program scope, objective, population, main activities and intended outcome are clear and consistent. If an essential is missing or contradictory, identify the gap or conflicting sources and ask only the questions needed to resolve it. Stop before drafting a program pathway or filling assumptions and risks. An unfilled outline is optional.
Once the scope is settled:
- Map activities to outputs and intended outcomes with arrows. Explain how each link is expected to work, for whom, and under what conditions. Do not add unrelated impacts or present intended change as achieved.
- Distinguish supplied evidence, user context and proposed causal assumptions. Cite only supplied or verified sources that support the claim; omit invented source details, consultations, approvals and results. Mark unknowns. Keep conflicting evidence visible rather than selecting a preferred claim.
- Identify the assumptions and alternative explanations that matter most. Suggest proportionate ways to test uncertain links and identify whose perspectives are missing.
- Identify material implementation, external, sustainability, equity and measurement risks with practical responses. Avoid unsupported risk ratings; label any judgment provisional and explain its basis.
- Suggest a small set of measures with units, feasible candidate sources and safe, supportable disaggregation. Preserve supplied values, periods, denominators and target status. Do not infer causation from before-and-after change or assess sample representativeness or subgroup precision without the design and relevant counts.
Return a text logic map, a short causal narrative, compact assumptions and alternatives and risk-response tables, a measurement note, and focused stakeholder-validation questions. Label proposed methods and future validation as proposals, keep suggested counts consistent, and leave unsupported details unknown. Keep the draft proportionate to the program.
Worked Example
Illustration only. The setting and figures are invented to show the pattern.
Scenario: A program trains village health volunteers in two districts to promote early antenatal visits. The team wants a theory of change before designing its evaluation.
What You Give the AI: The program objective, the target women and districts, main activities, the intended outcome of earlier first visits, a short note from a past project, and a draft logic model with gaps.
What the AI Might Return: A text logic map from training and household visits through to earlier first visits, with a short narrative for each arrow. Assumptions include that volunteers are trusted and clinics are open on visit days. An alternative explanation is that a new transport route changed visit timing. A risk table covers volunteer turnover, and a measurement note names a small set of indicators with validation questions.
When to Choose This Prompt
- You need to explain why and how your activities should lead to the change you intend.
- The assumptions behind each link have not been written down or tested.
- You want rival explanations and risks beside the pathway.
When Not to Choose It
- You need a results table with indicators and means of verification. Create a Logframe
- The main input is an existing program narrative. Draft a Theory of Change from a Program Narrative
- The theory of change is drawn and the doubt is whether each link holds. Review My Theory of Change
Checking the Output
- Trace each activity in the map back to your material. Nothing should appear that you did not describe.
- Compare the narrative for each arrow with what you know. It should say for whom the link works and under what conditions.
- Look at how evidence is labeled. Supplied evidence, your context and proposed assumptions should be separate, and each cited source should be one you gave.
- Check the alternative explanations against the setting, and ask yourself whether the main one is missing.
- Read the risk table for implementation, external, equity and measurement risks. Any rating should be marked provisional with a reason.
- Make sure intended outcomes are not written as achieved.
Common Problems and Fixes
- The AI adds impacts far beyond your scope. Fix: Stop the map at the intended outcome you gave; wider change goes in as context.
- Assumptions read as facts. Fix: Mark each assumption as supplied, proposed or unknown, and add a way to test it.
- Sources and consultations appear that you never mentioned. Fix: Cite only what you supplied, and delete invented details.
- It starts drafting before the objective is clear. Fix: Settle the objective, population and main activities first. The prompt asks questions when they conflict.
Adapting It to Your Context
- Existing logic model: Paste it and ask for gaps and conflicts to be listed before any redraft.
- Conflicting stakeholder views: Give each view and ask for them to remain visible in the map.
- Short proposal deadline: Ask for the map, narrative and assumptions only, with validation questions to follow later.