What Is an AI Prompt?
The instructions you give Holly are called prompts. How much you get out of Holly depends heavily on how clearly you write them.
Think of Holly's AI like a knowledgeable new colleague who has read extensively about government HR but knows nothing about your agency specifically. Vague instructions produce generic results. Specific instructions, with the right context, produce something much closer to what you need — on the first try.
What Makes a Good Prompt?
A strong prompt does three things:
States the task clearly — What do you want Holly to produce or do?
Provides relevant context — What does Holly need to know to do it well?
Specifies constraints — Are there things the output should or shouldn't include?
Examples
Say you're drafting a new class spec for a position in your agency.
Vague prompt:
Write a class spec for an HR analyst.
Specific prompt:
Write a class spec for a mid-level HR Analyst position at a mid-sized California county. The role focuses on classification and compensation work — reviewing position allocation requests, conducting salary surveys, and maintaining the classification plan. This is a journey-level position, above entry-level but below a senior analyst. We use a formal tone consistent with civil service specs.
The second prompt gives Holly the role level, the core duties, the agency context, and the tone. That's what produces a useful first draft instead of a generic one.
The same principle applies when editing. Rather than "rewrite this section," try:
Rewrite the distinguishing characteristics section to clarify how this role differs from the HR Analyst I. The Analyst I handles routine transactions; this role involves independent judgment and leads smaller projects.
In ChatMOU, instead of asking a broad question:
What does the MOU say about overtime?
Try:
Under the SEIU MOU, how is overtime calculated for employees who work a 9/80 schedule? Does comp time accrue differently than cash payment?
How Much Context to Include
More context is almost always better, but it doesn't need to be long. A few targeted sentences beat a paragraph of vague background. When in doubt, ask yourself: if I handed this prompt to a colleague who had never worked at my agency, would they have enough to go on?
Useful context to include when relevant:
The classification series or level (entry, journey, senior)
Your agency type and size (county, city, special district)
The tone or style you want (formal, plain language, consistent with existing specs)
What the output will be used for (a draft for review, a final posting, internal documentation)
What you don't want (avoid jargon, don't include duties that belong to a different class)
You don't need all of these every time. Match the context to the complexity of the task.
When the output isn't what you expected
AI output won't always be right on the first try — that's normal. The goal isn't an immediate perfect result; it's getting to a useful one efficiently.
Read it critically before you start editing
Before making changes, read the output through once and identify what's actually wrong. Is the tone off? Is the position level described incorrectly? Is a section missing? Pinpointing the issue helps you write a better follow-up instruction.
Give a specific follow-up instruction
Rather than rerunning the prompt from scratch, tell Holly what to fix.
Instead of: This isn't right, try again.
Try: The knowledge and abilities section reads too broadly — it could apply to any analyst role. Revise it to reflect duties specific to classification and compensation work, like conducting salary surveys and evaluating position allocation requests.
Know when to step in yourself
Sometimes the fastest path is editing the output directly. If Holly got 80% of a spec right but the remaining section isn't landing after a follow-up instruction, it may be quicker to rewrite that section yourself. AI works best as a drafting partner — it's not a replacement for your editorial judgment.
How to review AI output before using it
Before using any AI-generated content in an official document, decision, or communication:
Read the full output, not just the sections you asked about.
Check that any specific claims — about duties, requirements, legal standards, or contract language — are accurate for your agency and jurisdiction.
Compare the output against your existing specs, MOUs, or other source documents where relevant.
Apply your professional knowledge. If something doesn't sound right, trust that instinct and investigate.
Note: Holly's AI is designed to support your professional judgment, not replace it. Always review AI output before using it in official decisions, reports, or communications.
If you notice a pattern of inaccurate or unhelpful results in a specific part of Holly, let us know at [email protected].
FAQs
What if I disagree with Holly's output? Your judgment takes precedence. Holly's AI is a tool to support your analysis, not override it.
Can I use the same prompt across different classifications? Yes, and that consistency is one of the benefits of AI — it applies the same methodology every time. Just make sure to update the role-specific details (level, duties, agency context) for each classification.
Does the length of my prompt matter? Length matters less than specificity. A three-sentence prompt with the right details will outperform a long prompt that's vague.
For more on how AI works across Holly's features, see How AI Works in Holly, or reach out to [email protected].