Ad hoc comparison prompts produce ad hoc results. One day you remember to rank your criteria and the comparison is sharp; the next you fire off "which is better?" and get a confident guess. The cure for inconsistency is a structure you run every time, so the quality of a comparison depends on the method rather than on how careful you happened to feel that morning.
The model below, FRAME, organizes comparison prompting into five stages: Frame, Rank, Assemble, Map, and Elect. Each stage has a clear job, and each exists to neutralize a specific failure that comparisons are prone to. The value of naming the stages is that they become a habit you can run from memory and a vocabulary a team can share.
This method is the constructive spine beneath the practices in Habits That Make AI Comparisons Hold Up Under Pressure; here it is assembled into one repeatable arc.
Stage One: Frame the Decision
Every comparison serves a decision, and the decision sets the standard for "better."
Name what you are deciding
Before listing options, state the choice and what you will do with the answer. "I am selecting a queueing system for a service that will run for years at moderate scale" anchors everything downstream. Without this, the model has no basis for judging relevance, the gap behind many of the failures in Seven Ways Comparison Prompts Quietly Go Wrong.
Confirm comparability
Check that the options occupy the same category and serve the same job. If they do not, the framework's first output is a reframed question, not a table.
Stage Two: Rank the Criteria
This is where most of a comparison's quality is won or lost.
Make the axes explicit and ordered
List what matters and put it in priority order. Ranking tells the model how to resolve trade-offs the way you would, instead of guessing. Unranked criteria leave the model to weight them by what is most discussed online rather than what decides your case.
Tie criteria to the decision
Each criterion should trace back to the framed decision. A criterion that does not affect the choice is noise that dilutes the comparison.
Stage Three: Assemble Symmetric Evidence
The comparison can only be as fair as its inputs.
Equalize the information
Give every option the same fields, depth, and recency. Asymmetric input is the most common silent distortion. Where you cannot equalize, instruct the model to flag where it reasons from missing data.
Demand traceable cells
In this stage, the prompt fills the comparison but attaches a source or assumption to each cell and leaves true unknowns blank. The output is an evidence table, not a verdict.
Stage Four: Map the Trade-offs
Before any recommendation, surface the structure of the decision.
Make conflicts and conditions visible
Ask the model where criteria pull against each other and under what conditions each option wins. This converts the comparison into a decision map rather than a flat ranking. The crossover points it reveals are usually where the real choice lives, a theme developed in The Axes That Decide Comparative Analysis Prompts.
Still no verdict
The map stage deliberately withholds a recommendation. Keeping the verdict out until the evidence and trade-offs are settled prevents an early conclusion from biasing the analysis.
Stage Five: Elect the Winner
Only now does the framework ask for a decision.
Scope the recommendation to your conditions
Feed the verified evidence table and the trade-off map back, then ask for a recommendation under your specific constraints. The verdict arrives grounded in inspected evidence rather than first impressions.
Verify before acting
Before you act, confirm the load-bearing numbers yourself. The framework structures the comparison; you own the facts that move the decision. Whether the whole method is working is judged by the signals in Judging Comparison Quality With the Right Signals.
Applying FRAME at Different Weights
The method is not all-or-nothing. Each stage can run heavy or light depending on what the decision demands.
The light version
For a quick, reversible choice, FRAME collapses to a single prompt that still respects the order: state the decision, name a couple of ranked criteria, ask for a short comparison with conditions, and accept a verdict. You get most of the benefit—a comparison anchored to a real decision and ranked criteria—at a fraction of the effort. The discipline survives even when the ceremony does not.
The full version
For a decision a team will commit to, every stage runs in full: framed decision, ranked criteria tied to that decision, symmetric and traceable evidence, an explicit trade-off map, a separately scoped recommendation, and human verification of the facts. The stages become distinct prompts, with verification between the evidence and the verdict. This is the form that earns trust for consequential choices.
Why the Order Matters
The sequence is not arbitrary; each stage depends on the one before.
Earlier stages constrain later ones
You cannot rank criteria meaningfully without a framed decision, because the decision is what makes a criterion relevant. You cannot map trade-offs without assembled evidence, because there is nothing to weigh. And you cannot elect a winner honestly without a trade-off map, because the map is what tells you whether a single winner even exists. Running the stages out of order—electing before mapping, ranking before framing—reintroduces exactly the failures the method exists to prevent, the ones catalogued in Seven Ways Comparison Prompts Quietly Go Wrong. The order is the method.
Where people break the sequence
The most common violation is jumping straight to Elect—asking "which is better?" before any framing, ranking, or mapping has happened. This is the default everyone reaches for, and it is precisely the habit FRAME exists to interrupt. The second most common is skipping Map, producing a flat ranking that hides the conditions under which the answer flips. Both shortcuts feel efficient and both quietly degrade the comparison. If you find yourself tempted to skip a stage, that temptation is usually a signal the decision matters more than you are admitting, not less.
Adapting FRAME to Your Domain
The five stages are universal, but what fills them is not.
Keep the structure, localize the content
A comparison of marketing channels, a comparison of cloud providers, and a comparison of job candidates all run through the same five stages, but the criteria, the evidence sources, and the conditions differ entirely. Treat FRAME as a container you fill with domain-specific content rather than a fixed recipe. The portability comes from the sequence; the relevance comes from what you pour into it. This is the same separation of structure from specifics that lets a single method serve wildly different decisions without becoming generic.
Frequently Asked Questions
What does the FRAME acronym stand for?
Frame the decision, Rank the criteria, Assemble symmetric evidence, Map the trade-offs, and Elect the winner. Each stage neutralizes a specific comparison failure, and running them in order produces a reliable, inspectable result.
Do I have to run all five stages every time?
For consequential decisions, yes. For quick choices, Frame and Rank carry most of the value. The later stages—symmetric evidence, trade-off mapping, and a separate election step—scale with how much the decision matters.
Why keep the verdict out until the final stage?
Because an early recommendation anchors the reasoning and turns analysis into advocacy. Withholding it until evidence and trade-offs are settled keeps the analysis honest and the conclusion grounded.
How is this different from just asking for a comparison table?
A raw table mixes evidence, trade-offs, and verdict into one pass, where they contaminate each other. FRAME separates them into stages so each is done cleanly and the final decision rests on inspected, not improvised, reasoning.
Can the framework handle conditional answers?
Yes—that is the Map stage. It explicitly asks under what conditions each option wins, producing a decision map rather than forcing a single universal verdict that may not exist.
Where do humans stay in the loop?
In assembling symmetric evidence and, critically, in verifying load-bearing numbers before acting. The model structures and reasons; people own the facts that drive the decision.
Key Takeaways
- FRAME gives comparison prompts a repeatable five-stage structure: Frame, Rank, Assemble, Map, Elect.
- Framing the decision sets the standard for "better"; ranking criteria wins most of the quality.
- Symmetric, traceable evidence prevents the asymmetry and fabrication failures.
- Mapping trade-offs and conditions turns a flat ranking into a real decision map.
- Withholding the verdict until the final stage keeps an early conclusion from biasing analysis.
- The model structures the comparison; humans verify the facts that move the decision.