Best Practices: Smart Snippets

Last updated: April 15, 2026

Goal

Designed to help your team generate clear, contextual, and accurate AI-written messaging using agent outputs without breaking personalization or logic.

Why It Matters For You

  • Smart Snippets only see agent outputs and AI Research Observations, not the agent’s full question or research context.

  • Without proper context, snippets can generate incorrect or awkward messaging.

  • Small agent changes can silently break snippets if not handled carefully.

Example:
“A snippet without context is guessing — not personalizing.”

What People Do Wrong

  • Use text-output agents inside Smart Snippets.

  • Reference {agent output} without explaining what the agent was answering.

  • Reference AI Research Observations without explaining what was researched or what the output represents.

  • Update agents without re-adding them to snippets.

  • Forget to reset variables after agent changes.

  • Use variables inside fallback text (fallbacks must be static text only).

  • Run email steps before agent research completes, causing snippet variables to be missing.

Unify’s Best Practice Recommendations

  • Use True/False or Multi-Select agents whenever possible.

  • Always explain what the agent is answering or what the AI Research output represents inside the snippet guidance.

  • If an agent changes, fully remove and re-add it to the snippet.

  • Include a fallback when agent output is none or unknown.

  • Add a minimum 2‑day delay before email steps that use agent research to ensure agent outputs are available.

  • Keep fallback text static—do not include any variables in fallback text.

How to Apply / Set Up / Build

Step

What to Do

Why It Works

1

Select a structured-output agent

Ensures predictable results

2

Add context explaining the agent’s question

Gives the snippet clarity

3

Reference the agent output clearly

Prevents incorrect assumptions

4

Define a static fallback (no variables allowed)

Avoids broken messaging

5

Re-add the agent after updates

Ensures variables stay in sync

Good Example

“Research was conducted to determine what system the company uses for HR, payroll, or ERP. The company likely uses: {agent output}…”

Bad Example

“Based on {agent output}, write a personalized email…”

Pro Tips

  • Snippets cannot infer meaning — spell it out.

  • Never assume {agent output} is self-explanatory.

  • AI Research Observations work like agent outputs — always explain what was researched and provide fallback text for missing variables.

  • If any template variable in a snippet is missing, the entire snippet uses the fallback—ensure all referenced variables will have values for your records.

  • If results look messy, check the agent output type first.

  • Treat agent updates like breaking changes.

Measuring Success / How To Know You’re Doing This Right

  • Cleaner, more relevant AI-generated emails

  • Fewer personalization errors or hallucinations

  • Faster iteration without snippet breakage

Templates

Template Name

Guidance

Question

Smart Snippet (Good)

Research was conducted to determine what system the company uses for HR, payroll, or enterprise resource planning. The company likely uses: {agent output}.

Write a personalized email explaining why {your company name} is a better fit than {agent output}. This email should:
• Address their current tool by its specific name
• Highlight key advantages of switching
• Be conversational and value-focused

If no system was identified, return this: {FALLBACK}.

N/A

Smart Snippet (Bad)

Based on {agent output}, write a personalized email explaining why {your company name} is a better fit than {agent output}.

The question is asking what software tool they use. If none, use the fallback.

- No explanation of what the agent was answering
- No real context for the snippet

N/A

Smart Snippet with AI Research

Write a concise 1–2 sentence actionable summary of the AI Research findings for this prospect. Use only the information in these research outputs:
- Research output 1: {AI_RESEARCH_OUTPUT_1}
- Research output 2: {AI_RESEARCH_OUTPUT_2}

Requirements:
• Sentence 1: state the single most important insight (what matters and why)
• Sentence 2: recommend one concrete outreach angle based on that insight
• Keep it neutral, specific, and avoid adding facts not present in the research outputs

If research outputs are missing, return: {FALLBACK}

N/A

Related Resources