Prompt Engineering Frameworks: CO-STAR, RACE, RISEN Compared

Search “prompt engineering framework” and you'll hit a handful of competing acronyms — CO-STAR, RACE, RISEN, and a few others — each presented as though it's the definitive method. It's worth saying upfront: they're not competing methods. They're the same underlying idea, independently packaged into different mnemonics by different authors. This page compares them honestly, including where sources disagree on the exact letters, and shows the one pattern they all reduce to.

CO-STAR

Context, Objective, Style, Tone, Audience, Response format. Originated from a GovTech Singapore prompt engineering competition and is one of the more consistently defined frameworks — most sources describing CO-STAR agree closely on all six components. Its strength is separating Style from Tone from Audience, which is more granular than most alternatives and useful for writing-heavy prompts where those three genuinely differ.

RACE

Role, Action, Context, Expectationin most descriptions, though some sources swap Expectation for Execute or add a fifth letter. Unlike CO-STAR, RACE's exact definition varies more between the guides that reference it — worth knowing if you see two sources disagree on what the letters stand for, since it's a community term rather than one with a single canonical source.

RISEN

Role, Instructions, Steps, End goal, Narrowing.More oriented toward multi-step tasks than CO-STAR or RACE — the “Steps” and “Narrowing” components explicitly push toward breaking a task into ordered pieces and then constraining the output, which makes it a reasonable pick for complex analytical or planning prompts specifically.

Other names you'll see

A handful of other acronyms circulate with less consistency across sources — variations that add a persona/style axis, or reorder the same four or five components. Rather than catalog every variant, it's more useful to notice the pattern: nearly all of them contain some version of role, context, the actual task, and a format or constraint element. The acronym is memorable packaging around the same short list.

The one pattern underneath all of them

Line them up and the overlap is obvious:

• Who — Role (CO-STAR's Context partially, RACE's Role, RISEN's Role)
• Why / background — CO-STAR's Context, RACE's Context
• What — CO-STAR's Objective, RACE's Action, RISEN's Instructions/Steps
• Boundaries — CO-STAR's Style/Tone/Audience, RACE's Expectation, RISEN's Narrowing
• Shape of the answer — CO-STAR's Response format, implied in RACE and RISEN

That's five buckets, which is exactly the role / context / task / constraints / format pattern covered in how to write better prompts for AI. None of the acronyms are wrong — they're relabelings of the same five ideas, and the acronym you'll actually remember and use consistently is the right one to pick. If none of them stick, skipping the mnemonic and just working through the five plain-English buckets works equally well.

Using the pattern without memorizing an acronym

Frameworks help until remembering the framework becomes its own overhead. The actual value is in the completeness — hitting all five buckets — not in reciting the right acronym under pressure. A tool that applies the underlying pattern automatically gets you the benefit of any of these frameworks without requiring you to memorize one:

TL;DR

• CO-STAR, RACE, and RISEN are the most commonly referenced prompt frameworks — they're independently named, not competing methods.
• CO-STAR is the most consistently defined across sources; RACE varies more between guides.
• All of them reduce to the same five buckets: who, why, what, boundaries, and the shape of the answer.
• Pick whichever acronym you'll actually remember, or skip the acronym and use the plain five-part pattern directly.
• The value is completeness across all five buckets, not which mnemonic you used to get there.