AI Prompt for a Program Logic Model
This AI prompt for a logic model turns the program facts you supply into a table of inputs, activities, outputs and short, medium and long-term outcomes. It then tests the logic: it flags every activity with no linked outcome, every outcome with no activity that could plausibly produce it, and the assumptions the model depends on. It does not invent the program for you.
The prompt
Act as a structured drafting assistant for program design. Draft a program logic model using only the facts I supply below. Program: [PROGRAM NAME] Problem or need: [PROBLEM OR NEED] Who is served and where: [POPULATION AND SERVICE AREA] Inputs (staff, funding, space, partners we have confirmed): [INPUTS] Activities (what the program does): [ACTIVITIES] Outputs (what we will deliver or count): [OUTPUTS] Intended changes, short, medium and long term: [INTENDED OUTCOMES] Timeline: [TIMELINE] Evidence we have that this approach works, with sources: [EVIDENCE AND SOURCES] Assumptions we know we are making: [ASSUMPTIONS] Outside factors that could affect results: [EXTERNAL FACTORS] Part 1. Create a table with these columns: Inputs | Activities | Outputs | Short-term outcomes | Medium-term outcomes | Long-term outcomes Use one row per activity or connected pathway. Include every activity and outcome I supplied, even where a connection is missing, and keep disconnected entries visible. Where I gave you nothing for a cell, write "not supplied." Do not invent activities, outcomes, targets, measures or evidence. Do not add an activity to make an outcome look supported. Do not present a proposed connection as proven. Part 2. Test the logic: 1. List every activity with no linked outcome. 2. List every outcome with no activity that could plausibly produce it. 3. List the weak steps between outcome stages, where the jump from one column to the next is large. 4. List any output that has been placed in an outcome column, or the reverse. Part 3. List the assumptions and external factors this model depends on, using only the ones I supplied, and say which pathway each affects. Then list questions my team should answer. Phrase gaps as questions, not as recommendations. End with a short checklist for validation by program staff and by people from the community we serve.
Works in ChatGPT, Claude, Gemini and other AI chat tools. Replace each bracketed placeholder with your own details before you send it.
What to fill in
- [INPUTS]
- The resources you actually have or have confirmed: staff roles, funding sources, space, equipment and partner contributions.
- [ACTIVITIES]
- What the program does, stated as actions: for example weekly workshops, home visits or one-on-one coaching. Include how often and for how long if you know.
- [OUTPUTS]
- The direct, countable products of the activities, such as sessions held or people enrolled. Outputs are records of delivery, not evidence that anything changed.
- [INTENDED OUTCOMES]
- The changes you expect in participants, families, systems or conditions, grouped by when you expect them. Describe changes, not services.
- [EVIDENCE AND SOURCES]
- Whatever supports your approach, with where it comes from: your own past results, an evaluation or published research you have read. Write "none yet" if that is the case. Do not ask the AI to supply research.
- [ASSUMPTIONS] and [EXTERNAL FACTORS]
- What has to be true for the program to work (participants can get to the site, partners will refer) and what is outside your control (funding changes, local conditions).
Why this prompt works
An AI tool asked to "write a logic model for a youth program" will produce a tidy, generic table that describes no real program. This prompt takes the opposite approach.
- Bounded source material. The AI arranges facts you supply into the table. "Not supplied" gives it an alternative to filling an empty cell with something plausible.
- Disconnected entries stay visible. The prompt forbids adding an activity to prop up an unsupported outcome. The gap is the finding.
- A second pass that tests the model. Part 2 asks the AI to critique the table it just built. Checking each activity for an outcome, and each outcome for an activity, is tedious for people and well suited to an AI.
- Outputs separated from outcomes. Confusing the two is a common logic model error. Workshops held are an output. Skills gained are an outcome. Step 4 looks for the mix-up directly.
- Six columns. Splitting outcomes into short, medium and long term makes it easier to see where the expected chain of change has a missing step.
- Gaps as questions. The AI is told to ask, not recommend. Deciding how to close a gap is a program decision.
These controls shape the draft. They do not establish that your program theory is sound.
Why it is appropriate to use
Program design is the point in the grant cycle where you can still change the plan. A logic model built now exposes weak connections while there is time to fix them, and the finished model feeds the need statement, work plan, evaluation plan and budget.
Using AI here is appropriate because it is organizing and stress-testing your own thinking. It is good at noticing that an outcome has nothing leading to it. It has no knowledge of your community, your participants or what will actually work for them.
Do not use AI to decide what a community needs, to choose outcomes without the people affected, to claim an approach is evidence-based or to produce evidence. Program staff and people from the community you serve must review the model before it goes into a proposal.
Keep case notes, participant information and confidential budgets out of the prompt, and check the funder's AI policy. The logic model guide explains each component in more depth, and the template library has a blank logic model you can fill in.
How to check and revise the output
- Check completeness. Every activity and outcome you supplied should appear. Remove anything the AI added.
- Check delivery. Can the inputs actually support each activity? Is each output a direct record of an activity?
- Trace each pathway backward. Start from a long-term outcome and ask what would have to happen first, column by column, until you reach an activity.
- Check reach. Long-term outcomes should stay within what the program can credibly influence. If the last column describes change across a whole community, ask whether your activities can carry that.
- Resolve the flagged gaps with people. Discuss each one with staff and community reviewers. Change the program or change the stated expectations. Do not ask the AI to smooth it over.
- Compare with the budget and timeline so the model, the money and the calendar describe the same program.
Frequently asked questions
- Can I use this prompt without a finished program plan?
- Yes. It is useful for organizing the decisions you have made and exposing the ones you have not. Leave unresolved fields marked as such. The table is not a finished plan until staff and community review has settled the main gaps.
- Why does this logic model have six columns?
- A common format has inputs, activities, outputs, outcomes and impact. This prompt splits outcomes into short, medium and long term so the expected chain of change is easier to examine. If your funder supplies a logic model template, use their format.
- What if the AI adds an outcome I did not supply?
- Remove it and check for other additions. If the idea seems useful, treat it as a question for staff and community discussion, not as part of the program.
- What is the difference between an output and an outcome?
- An output is what the program delivers and can count, such as sessions held or people served. An outcome is the change that results, such as improved skills or a change in behavior or conditions.