A logic model template you build by prompt in any GenAI tool — eight copyable prompts, six worked examples with data fields, and the 5 components explained.
Most logic model templates are a five-column table that looks finished and measures nothing. This page is a set of prompts you paste into any GenAI tool — Claude, ChatGPT, or Gemini — to draft your model, stress-test its weak links, and turn each outcome into a real instrument. A prompt gets you the structure in a minute. Persistent IDs, real intake, and follow-up are what get you the evidence.
The short answer
Two definitions worth keeping straight — then the five columns every template shares.
A logic model is a one-page visual that maps how a program turns resources into change. It lays out program theory in five columns — Inputs, Activities, Outputs, Outcomes, and Impact — with causal arrows running left to right. It becomes a measurement tool only when each column is tied to a matching instrument: an intake question, a survey, or a follow-up tracker.
Inputs, Activities, Outputs, Outcomes, and Impact. Inputs are the resources a program uses; activities are what it does; outputs are the countable results; outcomes are changes in knowledge, behavior, or condition; impact is the long-term change it contributes to. Some funders use four components, folding long-term impact into the outcomes column.
Build it live
A static template gives you blank columns. These give you a working session. The order matters — each prompt builds on the answer before it.
Start vague on purpose. A sentence is enough for the tool to fill in the causal logic between the boxes.
I run a [program type] program. We [main activities] for [population], and the change we're aiming for is [outcome]. Draw my logic model as five columns left to right — Inputs, Activities, Outputs, Outcomes, Impact — with one or two lines in each column. Keep it to a single page I could show a funder.
The boxes are easy. The arrows between them rest on assumptions that rarely get written down. This is where most models are thin.
Now look at the arrows between the columns, not the boxes. For each step — activities to outputs, outputs to outcomes, outcomes to impact — name the assumption that step is riding on. Then tell me which arrows are the weak links: the ones that sound right but that I have no evidence for yet.
A flagged assumption is only useful if you can test it. This converts each one into something a small team can actually track this year.
For each arrow you flagged as weak, give me one thing I could measure this year to know whether the assumption is holding. For each, name the indicator, the instrument (intake question, short survey, or follow-up), and when I'd collect it. Keep it to things a small team can realistically run.
An outcome with no instrument behind it is a wish. This forces every outcome and impact column to name how it will be measured — or admit it can't be.
Go column by column through Outcomes and Impact. For each one, write: the exact indicator, the instrument that captures it, who reports it, and the collection moment (intake / exit / 90-day / 6-month). Flag any outcome that has no realistic instrument — I want to see which ones are wishes rather than outcomes.
Everything you want to report later has to be captured at first contact. This drafts the intake that makes equity reporting and pre/post comparison possible.
Draft the intake form for this program. It must capture (a) a baseline measure for every short-term outcome in my model, and (b) the variables I'll need to break results down later — gender, race/ethnicity, age, income band, and geography. Write each as an actual question with answer options, and mark which questions are the baseline measures.
Over-claimed impact fails evidence review. This pulls the impact column back to what one program can genuinely cause and detect.
Review my Impact column. Which claims are bigger than one program can plausibly cause on its own — the ones an evaluator would push back on? Rewrite each into an outcome I can genuinely cause and detect within my program's span of control, and explain each change in one line.
When a funder uses different words, rewriting your outcomes breaks year-over-year comparison. A crosswalk keeps both reports possible from one set of data.
A new funder uses different outcome language. Here is their wording: [paste the funder's outcomes or framework terms]. Build a crosswalk that maps my existing outcomes to theirs without rewriting mine, so I can report to both and still compare this cohort to last year's. Show it as a two-column table: my outcome / their term.
The finished model is also a draft of your proposal. This turns the boxes and arrows back into the paragraph a grant or board deck needs.
Turn this finished logic model into the narrative a grant proposal needs: one tight paragraph that walks from inputs to impact in plain language, then a two-sentence version for a board slide. Use my real program details, not generic filler.
The honest part
A GenAI tool can draft the whole model and even the instruments — that is the easy 5%. The other 95% is the part that survives a reporting cycle.
What the prompt gives you — the 5%
What a data system gives you — the 95%
A prompt drafts the model. A record proves it. The draft is real value — it gets a defensible structure on the page in minutes. But evidence comes from the same person answered twice, a year apart, with one ID holding it together. That is the work the prompt hands off.
Logic model examples
The difference between a template that gathers dust and one that produces evidence is the bottom row of every card — the data field that makes each outcome measurable. Borrow the structure, swap in your program. Four are social-sector; two adapt the same discipline to other settings.
Job-readiness training for unemployed adults 18–24
School-based mentoring for middle-schoolers
Diabetes self-management program
Mobile food pantry + nutrition education
Internal reskilling program at a mid-size employer
First-generation retention initiative at a university
How to build one
The fastest way to a defensible model is to start from the change you can realistically cause, then trace back to the resources it takes.
The one long-term change your program contributes to. Keep it to one.
The measurable changes you can actually detect. One per outcome.
The countable signals that the activities happened.
What the program does to produce those outputs.
The resources each activity requires, with an indicator on every outcome.
Prompt 01 above runs this whole sequence for you in a single pass — then prompt 04 attaches the indicator to every outcome before you finalize.
Best practices · Logic model design
Every template uses the same five columns. These six rules — applied at design time, not at reporting time — are what make the difference.
Principle 01
If the template says the program will increase financial literacy, the intake survey must include a baseline literacy measure. Outcomes without instruments are reporting debts, not outcomes.
Principle 02
The ID is assigned at enrollment, not reconstructed later from exports. Without it, pre-post, cohort, and longitudinal comparison all fail.
Principle 03
Gender, race, age, geography, income — every variable needed for equity reporting must be in the intake form. Retroactive capture means re-surveying.
Principle 04
Long-term outcomes need surveys at 90 days, 6 months, or 12 months. The template must specify cadence and instrument or those columns never populate.
Principle 05
Reduced community poverty is too large for one program. Stop at outcomes the program genuinely causes — not aspirational impact no instrument can detect.
Principle 06
When a new funder uses different language, build a crosswalk — do not rewrite the outcomes. Cohort-to-cohort comparison breaks the moment the template changes its own words.
The common thread: every rule requires alignment between template and data system before the program starts. After enrollment, each rule costs ten times as much to enforce.
Template vs. architecture
Word and PDF templates from major foundations satisfy grant submission. They do not satisfy measurement. Here is where the gap opens — row by row, every cycle.
| Capability | Word / PDF template | A measurement system |
|---|---|---|
| Outcome–survey alignmentDoes each column map to a data field? | Manual, after approval | Generated from the columns |
| DisaggregationBreak outcomes down by demographic | Only if intake captured it | Structured at intake · any variable becomes a filter |
| Persistent participant IDsOne person across every survey | None · manual matching | Assigned at first contact |
| Pre-post comparisonBaseline vs. exit, matched | Spreadsheet match · drops 15–30% | Automatic via ID |
| Cohort-to-cohort comparisonLanguage consistency across cycles | Breaks when funders change | Crosswalks, not rewrites |
| Qualitative integrationOpen-ended linked to quant | Separate file · manual theming | Linked and analyzed by participant |
| Reporting prep timeFrom data to funder-ready | 11–14 days per cycle | Hours, not days |
| Follow-up deploymentTracking at 90 days / 6 months | Manual email · 20–40% response | Scheduled automatically |
What goes wrong
Most logic models fail not at the design stage but at the seam between design and collection. Five we see repeatedly.
A template can't validate whether your causal claims are defensible — that is the job of a theory of change, which is why rigorous evaluation needs both. For the full picture across a portfolio, see nonprofit impact measurement.
Frequently asked
A logic model is a one-page visual that maps how a program turns resources into change — Inputs, Activities, Outputs, Outcomes, and Impact, with causal arrows running left to right. It serves program design, funder communication, and the blueprint for what data to collect. It becomes a measurement tool only when each column is attached to a matching instrument.
Inputs, Activities, Outputs, Outcomes, and Impact. Inputs are the resources the program uses; activities are what it does; outputs are the countable results; outcomes are changes in knowledge, behavior, or condition; impact is the long-term change it contributes to. Some templates compress these to four by merging short- and long-term outcomes.
Inputs, Activities, Outputs, and Outcomes — folding long-term impact into the outcomes column. The five-component version splits outcomes into short-term outcomes and long-term impact. Both describe the same causal chain; choose the format your funder requests.
Yes — a GenAI tool can draft a structurally correct five-column model from one sentence about your program. The prompt library on this page walks through drafting, stress-testing the weak links, and attaching an instrument to every outcome. What a prompt cannot do is assign persistent participant IDs, deploy the intake form, or keep outcome language consistent across cycles. Use a prompt to draft; use a data system to operationalize.
Start with: "I run a [program] for [population], aiming for [outcome]. Draw my logic model as five columns — inputs, activities, outputs, outcomes, impact — left to right." Then follow up with "show the assumption riding on each arrow, and tell me which are the weak links." Asking for the weak links is what turns a static diagram into a stress test. The full eight-prompt sequence is above.
Yes — major foundations and university extensions publish free Word and PDF logic model templates, and they are fine for grant submission. Their limit is that a Word file is disconnected from your data: it produces a design document, not a measurement architecture. The prompt library here drafts the same five-column structure and goes further, naming the instrument behind every outcome.
Work backward: name the impact, then the outcomes you can realistically detect, the outputs that signal them, the activities that produce them, and the inputs they require. Write one measurable change per outcome, keep claims within your span of control, and attach an indicator to every outcome before finalizing. Prompt 01 above runs this sequence in one pass.
A sample logic model for a workforce program lists staff and curriculum as inputs, cohort training as activities, sessions completed as outputs, improved interview confidence as a short-term outcome, and employment at 90 days as impact. This page includes six worked examples — workforce, youth, health, food security, employee upskilling, and student success — each with a matching data field for every column.
A logic model describes the program — what it does and produces. A theory of change argues for it — why the activities should produce the outcomes. Most funders request a logic model at application; rigorous evaluation needs a theory of change underneath. See theory of change vs logic model.
No. A generated model is a design document; a measurement system is what makes its columns produce evidence. The draft is the easy 5% — structure and suggested instruments. The other 95% is persistent IDs, deployed surveys, baseline and exit linked by participant, follow-up on schedule, and consistent language across cohorts. The prompt gets you to the starting line.
Your next logic model
A prompt drafts the model in a minute. Sopact Sense is where that draft becomes the data architecture — the logic model and the measurement system are the same document.
