Build an AI prompt library in 5 steps

Internal communicators know how to build systems. They’ve created templates for executive town halls, manager toolkits and change rollouts. But when it comes to AI, many teams are still improvising. They’re writing strong prompts one day and rebuilding them the next.

Catharine Montgomery, CEO of The Better Together Agency, told Ragan that her experience compiling an AI prompt library began out of necessity — and frustration.

“I kept getting strong results from AI tools and then losing them because I had no system for saving what worked,” Montgomery said. “I’d spend time recreating something I had already figured out.”

She realized the problem wasn’t the tools. It was the lack of structure.

“Over time, I started to see which prompts consistently worked for certain communications functions, and that’s when I began organizing them by function rather than by tool,” she said. “The structure of the library matters as much as what’s in it.”

Here are five steps comms pros can take to build a prompt library. They’ll help bring you from experimentation to indispensable infrastructure.

1. Capture what works and store it immediately.

Montgomery said comms pros should take note and file their prompts away when they’re effective — even if a formal library isn’t in place.

“Save your prompts the moment one works, even if you have no system yet,” she said. “You can organize later. The worst thing you can do is tell yourself you’ll remember it, because you won’t.”

Mary Yang, chief marketing officer at SquareX, suggested that embedding prompts into existing workflow tools can help kickstart the process and give useful prompts a place to live.

“If you have a project management tool like Trello or Asana, building out a prompt library as part of those workflows can work well,” Yang said.

2. Find a use case and repeat it.

When figuring out what’s worthy of the prompt library, communicators should look to tasks they’re frequently charged with completing.

Jennifer Hawton, public relations and executive communications manager at PEMCO, said that she takes executive talking points from town hall meetings and prompts an AI platform to turn those points into a video script.

“I’ll tell it to create a voice-over friendly, shorter, more conversational and easy-to-deliver on-camera script from this very long list of talking points,” Hawton said. “That specificity helps.”

Montgomery said that if you get a result you don’t like, don’t move on. Reverse engineer it to get where you need to go and document that in your library.

“Every prompt needs context around it,” she said. “Write down what you were trying to accomplish, which tool you used and what the output looked like. A prompt without that context is hard to reuse.”

3. Standardize the format of a good prompt.

The difference between a weak prompt and a repeatable one is structure. High-performing teams rely on format. Including a strong format into the prompt library, it makes it easier to build new prompts or refine existing ones.

Items that frequently appear in the prompt template include:

• Role
• Objective
• Audience
• Tone
• Constraints
• Output format

“The one thing that I’ve seen more than anything else when trying to help people start their AI journey is that they treat it like Google,” Hawton said. “And you have to give it what you want the context for it and what the no-go or the parameters are. They can be, ‘Don’t take anything from the web. Don’t take anything from outside the company. Don’t add anything older than 2024.’”

These templates can then form the backbone of your library.

Step 4: Build role-based assistants into the library.

Storing your prompts and embedding templates are great, but communicators should consider building custom AI assistants into the library to avoid rewriting prompts all the time. Yang said that communicators should view libraries as systems that work for them, as opposed to simple storage.

“A Word doc or a Google doc with a bunch of types of prompts just doesn’t allow you to really use the full scale of the tools today,” Yang said. “I think building out a gem for Gemini or a project for Claude or GPT for ChatGPT makes the most sense.”

Internal communicators could add the following kinds of assets to refine their AI assistants through consistency:

• Executive emails
• Town hall scripts
• Manager toolkits
• Change announcements

Yang added that she refines her GPTs by ensuring she’s asking the right questions to move the library from prompt reuse to a workflow design.

“In the instructions, I will make sure that I’m telling this GPT to ask me about the target reader that I want,” she said. “What’s the takeaway that I want them to have? Is the takeaway related to business or strategy? What’s a call to action if there is one?”

Step 5: Update your library to ensure it doesn’t stagnate.

Prompt libraries should be living documents that are frequently audited and updated.

“I’d push back on the idea that a prompt library is something you build once and move on from,” Montgomery said. “The communicators who get the most out of theirs are the ones who treat it as a working part of their process, something that the team actively updates and improves as the tools and the work evolve.”

Montgomery suggested that, before any updates are pushed out to a wider audience, a dedicated team of editors should look at the library and ensure that refreshed prompts are approved.

“We have what we call our AI Dream Team,” she said. “And any prompt going out to all teams gets tested and approved by them first. That step keeps the quality consistent across the whole team.”

Even the strongest prompt libraries won’t get everything right on the first go. Montgomery said that many of the most valuable lessons come from realizing what doesn’t hit the mark.

“Don’t underestimate the value of documenting what didn’t work,” she said. “We keep notes on prompts that produced weak outputs and why, and those notes often teach us more than the ones that worked.”

Sean Devlin is an editor at Ragan Communications.