> For the complete documentation index, see [llms.txt](https://manual.formwise.ai/formwise/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://manual.formwise.ai/formwise/advanced-features/prompt-engineering.md).

# Prompt Engineering

Write better prompts, chain steps, and get consistent outputs in FormWise.

Prompt engineering is how you turn your expertise into **reliable outputs**. In FormWise, that means writing instructions the model can follow every time.

Use this page as your checklist before you publish a SmartForm or CoPilot.

### The mental model

Every generation is the same loop:

1. **Inputs** (form fields or the chat conversation)
2. **Context** (datasets, backend data sources, crawled URLs, search results)
3. **Instructions** (your prompt)
4. **Output** (what the user sees)

Your job is to control steps 2 and 3.

### SmartForms vs CoPilots (what changes)

#### SmartForms

SmartForms shine when you want **structured input → structured output**. You control the user inputs and the format.

Typical prompt goals:

* Generate a doc, plan, or checklist
* Produce content in a specific format
* Run multi-step workflows using prompt chaining

#### CoPilots

CoPilots shine when you want **guided conversation**. You control the assistant’s role, boundaries, and behavior.

Typical prompt goals:

* Ask smart follow-ups
* Diagnose a situation before answering
* Keep the user on a specific workflow

### Prompt anatomy that works

Use this structure in both SmartForms and CoPilots:

1. **Role**: who the assistant is
2. **Goal**: what “success” looks like
3. **Inputs**: what the model will receive
4. **Process**: how to think (steps, checks, constraints)
5. **Output format**: exact structure, headings, bullets, JSON-like blocks, etc.

{% hint style="info" %}
If you care about consistency, over-specify the **output format**. Format beats “be helpful” every time.
{% endhint %}

### The 80/20 rules (use these everywhere)

* **Write for a junior assistant.** No implied steps.
* **One task per step.** Split big jobs into multiple prompt steps.
* **Provide context before variables.** Tell the model what `@step_1` represents.
* **Add constraints.** Length, tone, audience, and do/don’t lists.
* **Force a structure.** Headings and bullets reduce randomness.

### Prompt chaining (SmartForms)

Prompt chaining lets you break a workflow into steps. Each step can use the output of earlier steps.

Common pattern:

* Step 1: extract or summarize
* Step 2: generate the main output
* Step 3: rewrite, format, or repurpose

{% hint style="warning" %}
Each step is a separate AI call. More steps means more tokens and more timeout risk.
{% endhint %}

### Quick links

<table data-view="cards"><thead><tr><th>Guide</th><th data-card-target data-type="content-ref">Open</th></tr></thead><tbody><tr><td><strong>Intro to Prompt Chaining</strong></td><td><a href="/formwise/advanced-features/prompt-engineering/intro-to-prompt-chaining-in-formwise.md">Intro To Prompt Chaining in FormWise</a></td></tr><tr><td><strong>Advanced Prompt Chaining</strong> (tokens, structure, strategy)</td><td><a href="/formwise/advanced-features/prompt-engineering/advanced-prompt-chaining-guide.md">Advanced Prompt Chaining Guide</a></td></tr></tbody></table>

### Practical templates (copy, then adapt)

#### SmartForm prompt template

```
You are a [role].

Goal:
- Create [output] for [audience] that achieves [objective].

Inputs:
- [Field 1]: @q1answer
- [Field 2]: @q2answer

Rules:
- Use a [tone] tone.
- Avoid [things to avoid].
- If critical info is missing, ask 1–3 clarifying questions.

Output format:
## Summary
- ...

## Recommendations
1. ...
```

#### CoPilot behavior template

```
Role: You are a [role] CoPilot.

Workflow:
1) Ask questions until you have [required info].
2) Confirm the goal in one sentence.
3) Produce the output in the format below.

Boundaries:
- Do not guess facts you cannot verify.
- If unsure, say what you need next.

Output format:
- Short answer
- Step-by-step plan
- Next action
```

### Debugging prompts (fast)

* **Bad structure?** Tighten the output format first.
* **Wrong assumptions?** Add required inputs and clarifying questions.
* **Too long?** Add limits (bullets, word count, “no more than 5 items”).
* **Inconsistent tone?** Define tone + add 1–2 short examples.
* **Hallucinations?** Add data sources, web intelligence, or “cite your source”.