Choosing the Right Use Case
The use case tells VantagePrompt what kind of work you are doing, so template, framework, expansion, and scoring all tune to match. There are six textual use cases — auto, coding, reasoning, creative, marketing, and structured — and picking an explicit one also skips the classifier call, because you are stating the intent instead of asking a model to guess it.
The use case you pick tells VantagePrompt what kind of work you are doing, so the optimizer tunes its template, framework, and expansion to match. Here is what each of the six textual use cases is for, and how "auto" classifies your intent automatically.
By Andrei Bădulescu, Founder at VantagePrompt·Updated
Every prompt you paste into VantagePrompt runs through the same pipeline: classify, select a template, pick a framework, expand, score. The use case you choose is the first lever you pull. It tells the optimizer what kind of work you are doing so the rest of the pipeline tunes itself for that job.
There are six textual use cases: auto, coding, reasoning, creative, marketing, and structured. (Image, audio, and video are separate modes covered in their own guides.) Pick the wrong one and you still get a usable prompt, but the expansion optimizes for the wrong things. Pick the right one and the structure, instructions, and scoring all line up with what you actually want.
How does the "auto" use case work?
Leave the selector on auto and VantagePrompt classifies the input for you with a single lightweight model call. It reads your rough text and decides two things: the professional domain (code, writing, marketing, analysis, data, and so on) and the intent (what you are trying to do). It also extracts whatever concrete dimensions your text already states or strongly implies — task, target, format, audience, tone, length, constraints, success criteria, and the output shape.
The important rule: the classifier does not invent missing context. If your input never mentions an audience, the audience dimension stays empty. It only captures what is actually there. That classification then drives template selection and the framework choice downstream.
Auto is the right default when you are not sure, or when one prompt mixes concerns (a feature spec that is part coding, part product writing). Let the classifier read the room. Switch to an explicit use case when you already know the job and want to skip the guesswork.
What does each explicit use case optimize for?
Choosing a specific use case does two things. First, it overrides the domain and intent directly instead of asking the model to guess — which means it skips the classifier call entirely, saving that round-trip. Second, it points the expansion at a focus that fits the work:
| Use case | Optimizes for | Reach for it when |
|---|---|---|
| coding | Language and framework versions, input/output contracts, error handling, tests. | Write, refactor, or debug this code. |
| reasoning | Chain-of-thought, step-by-step decomposition, self-verification. | Analysis, math, planning — any "think it through" task. |
| creative | Tone, voice, audience, narrative structure, emotional register. | Stories, scripts, any writing where style carries the weight. |
| marketing | Audience persona, persuasion structure, CTA, brand voice. | Ad copy, landing pages, emails, product messaging. |
| structured | Output schema, field definitions, validation rules. | Machine-parseable results you pipe into another tool. |
| auto | Nothing fixed — a classifier reads domain and intent from your text. | You are unsure, or one prompt mixes concerns. |
What changes when I switch the use case on the same input?
Say you paste this rough input:
summarize this customer support threadOn auto, the classifier reads it as a writing/summarization task and expands toward a clean prose summary. But your actual goal might be a row in a dashboard. Switch the use case to structured and the same input gets optimized for a parseable shape instead:
{
"summary": "one-sentence resolution",
"sentiment": "positive | neutral | negative",
"action_items": ["..."],
"escalated": true
}Nothing about the input changed — only the use case. That single choice is the difference between text a human reads and data a program consumes. The same goes for picking coding over auto when you want tests and error handling baked in, or marketing over creative when you want a CTA instead of a mood piece.
How do I decide which use case to pick?
- Not sure, or the input is mixed → auto. Let the classifier decide.
- Code in, code out → coding.
- "Show your work" / analysis / planning → reasoning.
- Story, voice, vibe → creative.
- Selling something, persuading someone → marketing.
- Output another tool will parse → structured.
Whichever use case you land on, VantagePrompt also picks a fitting prompt-engineering framework automatically and scores the result. You do not have to choose the framework yourself — see the frameworks guide for what that buys you.
Frequently asked questions
- What are the six textual use cases in VantagePrompt?
- auto, coding, reasoning, creative, marketing, and structured. Image, audio, and video are separate modes with their own tool selectors and their own guides.
- When should I leave the use case on auto?
- When you are not sure, or when one prompt mixes concerns — a feature spec that is part coding and part product writing, for example. Auto classifies domain and intent for you with a single lightweight model call. Switch to an explicit use case when you already know the job and want to skip the guesswork.
- Does the auto classifier invent context I did not give it?
- No. It extracts only the dimensions your text already states or strongly implies — task, target, format, audience, tone, length, constraints, success criteria, and output shape. If your input never mentions an audience, the audience dimension stays empty.
- Does picking an explicit use case make optimization faster?
- Yes. An explicit use case overrides domain and intent directly, which skips the classifier call entirely and saves that round-trip.
- What happens if I pick the wrong use case?
- You still get a usable prompt, but the expansion optimizes for the wrong things — for example prose a human reads when you wanted data a program consumes. Re-run with the right use case; nothing about your input has to change.
Sources
Put it into practice.
Run this technique in the optimizer.