Prompt Guide

Creating Your Own Prompt

Connect AI allows you the ability to ask a question in plain English and get a clear, data-backed answer quickly. If you can send a text, you can use Connect AI. Educators can ask questions, and get immediate role-specific answers. You can cross reference multiple data sets with a single prompt, and degenerate reports, summaries and agendas in minutes. You can further refine your insights through conversational follow-up questions. 

To get the most accurate and insightful results from Connect AI, keep a few simple best practices in mind. Here is how to unlock its full potential:

Language Considerations

• Assessments: 
  • Use clear, specific assessment names as they appear in Homeroom.
  • Avoid ambiguity where multiple assessments exist. 
  • Generic terms may pull from
  • the wrong source
• Unique Program Names:
  • Use the full, official program name as it appears in Homeroom.
  • If results look wrong, ask "What programs do you have data for?" before re-prompting.
  • For programs that share a name with a common term (e.g., "Bridges"), add a qualifier —"Bridges Math Intervention program."
• Core Action Words
  • There are four core action words for Connect AI. Use them to get efficient results: summarize, compare, identify, and visualize. 

Functional Considerations

• Start Broad, Then Narrow
  • Enter a general prompt first to see what data is available.
  • Layer conditions incrementally: grade level →school → program →threshold.
  • Narrowing step-by-step helps you catch misinterpretations before they compound.
  • Follow up by asking for key findings and next steps.
• Validate Before Sharing Externally
  • Any result headed to leadership, parents, or state systems should be validated first.
  • Validation prompt: "What data did you use to generate this, and does it account for [specific variable]?"
  • If a number seems off by more than ~5%, investigate before sharing.
• Know the Boundaries
  • Connect AI draws only from data in the Homeroom Connect ecosystem — not external state databases or unintegrated systems.
  • If a state metric doesn't match, it's likely a calculation methodology difference, not an error.
  • Results are a starting point. Human review before external distribution is always best practice.

• Prompt Guide.pdf100 KB