AI Prompt Frameworks: CRISPE, TAG, RACE, and When to Use Each

Prompt frameworks are tools for structuring instructions, not requirements. Many well-performing prompts do not use any named framework — they simply include enough specificity about role, context, task, constraints, and format. Frameworks become useful when you want a repeatable template for a category of tasks, when you are training a team on consistent prompting practices, or when you find that freeform prompts consistently miss one structural element.

If frameworks feel like overhead for your current tasks, skip them and use the ChatGPT Prompt Generator to build structured prompts from plain task descriptions. The generator applies the structural principles automatically.

CRISPE stands for: Capacity (what role the AI should take), Role (the persona or expertise), Insight (background information the AI needs), Statement (what you want it to do), Personality (tone and style), Experiment (how many options to produce).

Applied example: "[Capacity] You are an experienced email copywriter. [Role] Write as a direct-response specialist. [Insight] This is a re-engagement campaign to SaaS users who have not logged in for 60 days. The product is a project management tool. The primary concern of this segment is setup complexity. [Statement] Write a re-engagement email under 150 words. [Personality] Warm, direct, no guilt-tripping. [Experiment] Give me 2 options."

CRISPE works well for complex content tasks where tone, persona, and context all matter. It is heavier than needed for simple formatting or data extraction tasks.

TAG stands for: Task, Action, Goal. It is intentionally lean — suitable for simpler, clearly defined requests where you do not need a detailed role or tone specification.

Applied example: "[Task] Summarize the following article. [Action] Extract the 5 most important points as a numbered list. [Goal] The summary will be used in a newsletter for B2B marketers — keep it under 100 words total."

TAG is useful for research, summarization, data extraction, and structured formatting tasks. For creative or tone-sensitive writing tasks, it is usually too minimal.

RACE stands for: Role, Action, Context, Execute. Similar to CRISPE but more action-oriented — it emphasizes what the model should do and the context it needs to do it well.

Applied example: "[Role] You are an SEO content strategist. [Action] Create a content brief for an article targeting 'content marketing for startups'. [Context] The audience is early-stage B2B SaaS founders with no dedicated marketing team. The content goal is organic traffic and email list growth. [Execute] Output: H1, 6 H2s with one sentence on each, 5 secondary keywords, word count range, and one differentiation angle."

RACE works well for SEO and content planning tasks where a clear action and defined context produce reliable structured output.

For tasks where style, format, or voice consistency matter most, providing examples of the desired output is often more effective than any named framework. This is called few-shot prompting.

Applied example: "Write a LinkedIn post in the style of this example: [paste example post]. The topic is [topic]. Audience: [describe]. Keep it under 200 words."

Few-shot prompting is particularly effective for brand voice matching, style replication, and producing output in a specific format that is easier to show than to describe. One or two examples is usually enough — more examples do not consistently produce better results and can confuse the model about which pattern to follow.

For channel-specific prompting guidance with examples, see: ChatGPT prompts for marketing · ChatGPT prompts for SEO · ChatGPT prompts for blog writing.