Which prompt framework fits your task?

What a prompt framework actually is

A framework is a reusable slot structure — a named checklist of the things worth specifying, arranged for one kind of task.

CO-STAR asks for Context, Objective, Style, Tone, Audience and Response format. AIDA asks you to move a reader through Attention, Interest, Desire and Action. STAR asks for Situation, Task, Action and Result. None of these is a magic phrase. What they do is stop you leaving out the things that change an answer.

That is the honest case for frameworks, and it is a real one. The most common reason AI output disappoints is not that the model is weak — it is that the prompt left the audience, the format and the constraints unstated, so the model chose them, and chose differently each time. A framework is a prompt for the person writing the prompt.

How to choose one

The most common mistake is matching a framework to the topic. Frameworks match the shape of the task, and the same topic can take several shapes: writing a pricing page, explaining a price rise to existing customers, and analysing whether to raise prices at all are three different jobs that want three different structures.

Ask what the reader is meant to do

If they should act — buy, click, reply — you want a persuasion structure: AIDA, PAS, or one of their relatives. If they should understand, you want an explanatory structure. If they should decide, you want an analytical one that surfaces trade-offs instead of conclusions.

Ask whether you are producing or investigating

Producing a deliverable — an email, a page, a post — suits a content framework. Investigating something — comparing options, diagnosing a problem, working through a decision — suits a reasoning framework, and those two families behave very differently. A reasoning framework applied to a tweet produces a tweet with a methodology section in it.

Ask how much you already know

If you can already state the audience, the format, the length and the constraints, you do not need a framework — you need to write those four things down. Frameworks earn their keep on the tasks where you know what you want but not what to specify.

The 7 domains

Frameworks group by the kind of work they structure. The finder above searches all 60; this is the map.

The frameworks worth knowing by name

Most people need four or five. These are the ones that come up constantly, and what each is genuinely for.

One task, three frameworks

The clearest way to see what a framework does is to hold the task constant and change only the structure. The task: tell existing customers we are raising prices by 12%.

With APE — Action, Purpose, Expectation

Produces a short, direct notice. Action: write an email announcing the increase. Purpose: so customers hear it from us before they see it on an invoice. Expectation: under 150 words, plain, states the date and the new figure. The result is an operational message that gets the fact across without ceremony.

With CO-STAR — Context, Objective, Style, Tone, Audience, Response

Produces something written for a specific reader. The Audience slot forces you to say which customers — three-year subscribers on legacy pricing read this very differently from last month’s signups — and the Tone slot forces a decision you would otherwise leave to the model. The output is longer, warmer, and acknowledges the relationship.

With SCQA — Situation, Complication, Question, Answer

Produces an explanation. Situation: pricing has been unchanged for four years. Complication: infrastructure and support costs have risen. Question: how do we keep the service at this standard? Answer: a 12% increase from March. Same facts, but the reader arrives at the number having already followed the reasoning.

A framework is only as good as what you put in it

The most common failure is a perfectly structured prompt full of empty slots. "Audience: our customers. Tone: professional. Style: engaging." Every box is filled and nothing has been specified — a model already assumes it is writing professionally for your customers.

A slot is doing work when the answer could have been different. "Audience: finance leads who have never used a data warehouse" tells a model what to explain and what to assume. "Audience: professionals" tells it nothing. The test is simple: if the opposite of what you wrote would be absurd, you have not said anything.

Three slots people habitually waste:

• Tone. "Professional" is the default. "Direct, and willing to say the thing people are avoiding" is an instruction.
• Style. "Engaging" is a hope. "Open each section with a concrete example before any explanation" is checkable.
• Response format. "Well-structured" could mean anything. "Three sections of about 200 words, then a five-row comparison table" cannot be misread.

Where these came from

Most prompt frameworks are older than prompting, which is a point in their favour: they were tested on human readers for decades before anyone pointed them at a model.

AIDA was described by the advertising writer E. St. Elmo Lewis around 1898 as a model of how a customer moves from noticing an advertisement to buying. STAR comes from behavioural interviewing, where it exists to stop candidates describing what they would hypothetically do instead of what they actually did. MECE is management consulting’s contribution — a discipline for breaking a problem into parts that neither overlap nor leave gaps. SCQA comes from Barbara Minto’s work on structured business writing, built for readers senior enough to want the answer before the analysis.

The genuinely new ones are the reasoning frameworks — chain of thought and ReAct among them — which came out of machine-learning research rather than writing practice and have no human-communication ancestor. That split is worth knowing, because it predicts where each family works: the old ones shape what a reader experiences, the new ones shape what a model computes.

Every framework, by domain

All 60 in the library — 6 available free and 54 on a paid plan. The finder above searches every one of them regardless of plan.

When a framework does not help

Three cases, and they are more common than framework advocacy admits.

When you already know what you want. If your prompt names the audience, the format, the length and the constraints, a framework adds ceremony. Check it with the prompt scorer — anything in the eighties does not need scaffolding.

When the fit is forced. A persuasion structure applied to a technical explanation produces an explanation that is trying to sell you something. If the finder above returns everything at low confidence, that is the honest signal that no framework fits, and a plain well-specified prompt is better.

For very short output. A subject line, a tweet, a button label. The overhead of a six-part structure exceeds the output.

How the matching works

The finder is deterministic. No model is called: it is pattern analysis in PHP, so the same description always returns the same ranking. That matters because it means you can edit your description, re-run it, and know that any change in the result came from your edit.

Each framework carries a hand-written set of weighted patterns — the phrasings that indicate its kind of task, plus negative patterns that stop it claiming tasks belonging to a neighbour. Your description is scored against all 60, and the ranking is what comes out. The bar on each result is its score relative to the winner, so you can see whether the top match won decisively or by a nose.

Two honest limitations. It matches phrasing, not meaning: describe a task in words none of the patterns anticipate and it will under-match, which is why the tool says "no strong match" rather than promoting a two-point result. And it is tuned by hand, so it is better on common task types than unusual ones.

This is the same scorer the Frompting generator uses to pick a framework. What you see here is what the product would actually do with your task.

Common questions

Do I have to use a framework?

No. A prompt that names the task, audience, format, length and constraints is excellent without one. Frameworks are a route to that, not a requirement.

Can I combine two frameworks?

Rarely worth it. Two structures usually fight — one wants a narrative arc, the other wants exhaustive categories. Pick the one matching the primary job and borrow a single slot from the other if you need it.

Why did it pick a framework I have never heard of?

Because the match is on task shape, not on fame. The obscure ones are often specialised and fit narrow tasks better than the famous general-purpose ones. The expansion under each name tells you what it asks for.

What if two results score almost the same?

Then either would work, and the choice is stylistic. That is what the strength bars are for — a close second means a genuine tie, not a near miss.

Does the framework change the answer, or just the prompt?

The prompt, and then the answer through it. The framework decides what the prompt asks for; specifying an audience and a format is what changes the output.

Glossary

• Prompt framework — a named slot structure for writing a prompt.
• Acronym expansion — what the letters stand for, and the actual content of the framework. CO-STAR is only useful once you know the six slots.
• Domain — the family of work a framework is built for.
• Zero-shot — prompting with no examples. What most people do.
• Few-shot — including one or two examples of the output you want.
• Chain of thought — asking the model to reason before concluding.
• Slot — one element a framework asks you to fill, such as Audience or Tone.
• General-purpose prompt — a well-specified prompt with no framework. The right answer more often than framework advocacy suggests.