How to create, organize, and scale your AI prompt library

How’s your prompt library looking? 

If you're like us, you're finding more and more use cases for improving the speed or quality of your work through well-tested generative AI prompts. But if your prompt library has become cluttered and difficult to navigate, it might be time for a new system.

No matter what size your current prompt library is, now is a great time to get it in order. 

What is a prompt library? 

Let’s quickly cover what a prompt library is.

The secret about prompts is that writing them might actually take more time than the original task. Repeat them, though, and that's where the efficiencies are gained.

A prompt is a way to capture your knowledge, your expertise, your approach — and then scale it. That's where your prompt library comes in: Quite simply, a collection of your best and most frequently used prompts.

If you don’t yet have a prompt library, you need one — that is, if you intend to save your intellectual property, scale your AI adoption, or share prompts with others.

Not having a well-organized prompt library could lead to: 

Lost knowledge

The most obvious risk of not having a centralized library is losing your personal intellectual property or valuable institutional knowledge. Maybe you have to scroll back through your chat history, or your chat history gets cleared, or you have prompts stored in various files and documents. 

In any of these cases, if you can't find it, you can't use it. And if you can't use it, you can't replicate it.

Redundancies

Maybe you have some prompts jotted down, but they aren't in a centralized place or organized in any useful way. Whether working solo or with a team, this could lead to multiple copies of the same or similar prompts across different files. This can make it more difficult to control versions of prompts and lead to inefficiencies.

Information silos

If everyone on your team is storing their prompts separately, there could be valuable information (prompts) hidden away that could benefit other team members. By controlling the repository of prompts and versions, everyone's work gets better.

Compliance breaches

The worst-case scenario is that unmanaged prompts could lead to breaches in organizational or legal regulations. This might be because of what is included in the prompt (confidential or sensitive information) or in what is excluded (such as a disclaimer to not produce harmful, biased, or inappropriate responses). 

This is also why prompt libraries should contain not just the inputs for the large language model (LLM), but also the entire prompt playbook or standard operating procedure — more on this later.   

Benefits of a prompt library

In contrast, the benefits of having a prompt library are:

• Documentation: These tried and tested prompts can be saved and stored for future use. 
• Centralization: All prompts are stored in a single place, making it easier to locate them and avoid redundancies.
• Efficiency and Scale: Prompts can be reused by you or another member of your team. That means team members spend less time recreating similar prompts, and can more easily use them at scale. 
• Version Control: Your prompt library becomes a central repository for prompts that have been used, or perhaps even approved, by your organization. As prompts are refined over time, the newest versions can be stored in this prompt library. 
• Quality Control: Maybe all text generation prompts for branded materials need to reflect a certain style or tone. Maybe certain prompts need to contain a disclaimer that the generative AI is “not to produce anything biased, harmful, untruthful, or inappropriate.” These requirements need to be woven into your prompt library so that prompt quality can be standardized. 

Whether you have a prompt library now or are starting one from scratch, let's look at the ways that you could organize it.

How to create a prompt library

Building a prompt library is pretty straightforward:

1. Document prompts you use often

To start, you might scroll back through your chat history with ChatGPT or the chatbot of your choice. Moving forward, try to write your prompts in a separate file — like a word document or notes app — so you can easily store it if you like the result and want to use it again.

2. Collect your best prompts

‍These might include prompts that performed particularly well for you, and they should definitely include prompts that help you complete common tasks. (Uncover common tasks — and opportunities to use AI — with this simple framework.)

3. Save and store prompts in a digital format

‍We explain how to organize prompts in more detail below, but your prompt library can exist in whatever digital format works best for you and your team: a word document, a spreadsheet, a folder of separate files, etc.

4. Scale gen AI by reusing prompts

To reuse a prompt, simply locate the prompt, copy and paste it into a new chat, make any edits or changes needed, and hit send.

5. Share prompts with others

We offer more tips for teams at the end of this article, but you can share your prompts by sharing access to your entire library, or by inviting team members to a third-party collaboration tool like BoodleBox. Tools like this make it easy to share prompts, chat with coworkers and generative AI at the same time, and even bring multiple LLMs together into a single thread.

Remember that prompts are intellectual property — they reflect your unique thinking, experience, and expertise. So consider carefully before sharing them publicly.

How to organize your prompt library 

A well-organized prompt library will allow you to enjoy the benefits of all your experimentation and learning thus far, better integrate your prompts into your day–to–day work, and more easily scale your generative AI use.

1. Pick your system  

Start by deciding how you want to categorize and describe your prompts. If you've already started collecting prompts, you may have established patterns here. 

Either way, take a step back and make sure the system you pick best serves you. 

A few options you could choose are: 

• By Function 
• By Discipline 
• By Project or Campaign 
• By Modality
• By Task 

Consistency is the most critical, but also consider how you currently organize your workload and navigate different projects and tasks. This will inform which of these options will make finding the right prompts at the right time (and adhering to your system) feel effortless.

‍

‍

By function

Function-based organization may be helpful for very narrow prompts. These typically begin with a verb and describe the action you’re asking AI to take. 

For example: 

• Summarize this text. 
• Transcribe this audio.  

These are sometimes called “AI use cases,” and can be useful if you tend to have very narrow tasks that you repeat often, or if you want to split projects into distinct tasks or even sub-tasks. 

The downside is that this isn't typically how we work. Even as we adapt to work collaborating with generative AI, many prompts are lengthy. They may include multiple steps or multiple functions. Teams may struggle to adhere to a function-based system with more lengthy and complex prompts.

By discipline

Organizing your prompt library by discipline would focus on describing prompts based on the broader business function they serve (marketing, sales, operations), or specific services or domains within a department (e.g. content, SEO, social media or acquisition, customer retention, sales enablement). 

Discipline-based organization can be useful if teams perform tasks that are very distinct from one another. But prompt authors might struggle to categorize prompts that may fall into more than one group: Many copywriting and editing prompts could be used across content and social media. Data analysis prompts might be valuable to marketing but also to operations. 

By project or campaign 

Project- and campaign-based prompts contain specific tasks and contexts that are relevant only to these prompts. This may be useful if you have a series of activities that you only need to perform occasionally: For example, prompts for planning and executing an annual event. 

But you may find it challenging to use a project- or campaign-based organization across your entire prompt library. And like discipline-based organization, they risk siloing great prompts from broader use. 

By modality

Categorizing your prompts by modality (text, image, video, audio) can be helpful if you're using a combination of different generative AI platforms with these different modalities. For example, if you use Runway for video generation or Midjourney for image generation. 

This may become less relevant as LLMs incorporate different modalities (like DALL-E’s integration into ChatGPT). Prompt library authors (and users) also have to distinguish between whether that modality is describing the input or the output: You might use Midjourney to generate images, but ChatGPT to analyze and interpret a screenshot of a chart. 

Lastly, modality-based organization, like function-based organization, creates very narrow use cases. If a prompt focuses on analyzing a report and writing bullet points for an upcoming presentation, how do you easily navigate the user from the image prompts (“generate this chart” or “interpret this graph”), to the related text-based prompts (“write three bullet points summarizing these findings for an executive team”)? 

By task 

Task-based prompts are outcome-based and our preferred prompt library system. They describe the end result that will be achieved which may contain multiple steps (actions) within it. 

This is valuable if you have groups of prompts built into a single process (e.g. content, SEO, social media, analytics, operations). Many successful examples of generative AI require a conversation — a series of prompts instead of a single, all-encompassing one. 

In fact, organizing around tasks allows prompt libraries to include a series of prompts, use multiple functions, address multiple modalities, and so on. Tasks can be used for specific projects and campaigns, and also applied to other uses outside of that specific context. 

Our typical workflow is project → steps within that project. We aren’t typically batching “summarization” tasks in isolation (unless you are). 

Consider: Does your to-do list look like this? 

Or more like this? 

The outcomes of these two approaches may look similar. But by adopting a prompt library system that aligns with your current workflow, you can more seamlessly – and more effortlessly — incorporate AI into your day-to-day work.  

Plus, when scaling a prompt library across a team, task-based prompts create a framework for a prompt standard operating procedure: A way for prompt authors or managers to ensure that certain steps are taken within each prompt’s use. These "prompt playbooks" ensure better alignment and better adherence to organizational standards. 

Not sure? Let AI organize it.

If you're still not sure which system is right for you, you might ask AI to uncover the patterns in your current prompt names and suggest an organizational system for you.

Note: If you believe that prompts are your own unique creations and intellectual property (as we do), avoid sharing your entire prompt library with generative AI — treat it like confidential, proprietary data. Instead consider sharing the names and descriptions of your prompts to create your new naming structure, without giving away all the specific details of your prompts.

2. Separate prompts from context blocks 

As you review your current prompt library, you may notice that you have prompts for specific events, projects, clients, audiences, or business functions. This may inhibit you from getting the full value out of your current prompts, and may silo important prompts from being used across other projects, roles, and functions. 

Ask: 

• Can that incredible buyer persona prompt be used across all of your audiences? 
• Can that content calendar prompt be used for every client? 
• Can a fine-tuned data analysis prompt be used on both marketing analytics and financial data? 

Instead, consider separating the prompt from the context blocks. 

Context blocks 

What are context blocks? In this case, we’re talking about sections of a prompt that add information, but that could be swapped out for a different context and allow you to reuse that prompt in a new way.

Things like: 

• Brand or client descriptions 
• Tone-of-voice summaries and examples 
• Audience descriptions and characteristics
• Etc.

These add nuance and important details to your prompts, and they’re incredibly essential to fine-tuned outputs. But they are best used in combination with prompts that perform specific functions.

We use color-coding to indicate sections of the prompt that must be reviewed, revised, and/or added before the prompt can be used.

‍

‍

3. Choose your naming convention  

Once you've selected your overall system, and identified and extracted your context blocks, you can work on renaming your prompts in a consistent, cohesive way.

Prompt names should follow the overall system you've established, along with any other valuable details that will help the user find the prompt they're looking for. This is where you can actually combine some of those systems we discussed earlier. For example:

Project: Task 

Discipline: Function

Discipline: Task 

I think this last option is the best option for most organizations, especially when combined with an “All Organization” option for company-wide prompts. 

How "Discipline: Task" naming  might look in practice: 

All Organization: Generate text in brand tone 

All Organization: Generate OKRs deck for your department 

Marketing: Create product launch campaign

Sales: Analyze monthly sales numbers 

Once all prompts follow a similar naming convention, you might choose to order them alphabetically for easier access, or use foldering or nesting to further organize prompts. For more on the features and functionality of your prompt library, scroll down to Where to Store Your Prompt Library.

Prompt details: What’s included in a prompt 

Don't feel the need to cram every possible detail into the prompt name. 

Instead, rethink what you include in your prompts. Your current prompts may look something like this: 

Instead, your prompts could look something like this: 

Building out your prompt library with additional details allows you to include the functions of each prompt, the modality, and the category or disciplines the prompt applies to (Marketing, Operations, Market Research, etc.), along with any other details related to its use. Include keywords or tags that might help when searching for valuable prompts.

This allows you to transform a simple prompt into a complete standard operating procedure — a Prompt Playbook. 

How to organize your team’s prompt library 

Now let’s get into specifics of organizing a shared prompt library. 

Tips for teams 

If a prompt library is going to be shared across a team or a company, there are a few additional considerations for how it will be managed and updated over time. 

Permission levels 

Who can add new prompts to the library? Who can update or modify prompts? These decisions may align with other file and knowledge management in your organization, and should absolutely align with your company’s generative AI policy.

Version control

Will you include any documented versioning for prompts or the prompt library? This might include the author, date changed, and iteration of the prompt (e.g. v1.2). Like in document management systems (DMS), version control can be helpful in team environments to track changes and updates.

Prompt sharing and intellectual property 

Consider the requirements and limitations of sharing prompts internally and externally.

• Internally: Are employees required to share prompts they create at work and for work projects? Do they own their prompts, or do these become the IP of the company? 
• Externally: Are employees permitted to share prompts externally, whether on LinkedIn, in blog posts, at speaking events, etc.? Are prompts employees’ own IP, company IP that should be kept confidential, or does it depend? 

The answers to these questions should align with your employee handbook’s guidance on how employee work and IP are handled. 

Governance and Standard Operation Procedures 

Your prompt library is another place to reiterate and reinforce prompting best practices. This might include things like: 

• Reminders to exclude confidential or sensitive information from prompts.
• Step-by-step instructions to anonymize data.
• An LLM selection flowchart.   
• Brand tone, vocabulary, grammar, or other style guidelines. (These might be part of a brand context block.) 
• Prompt requirements, such all prompts must include instructions to not produce harmful, biased, or inappropriate responses. 

With the addition of these guidelines and parameters, the prompt library is more than a collection of prompts — it’s your company’s handbook for prompt engineering. 

With all of these decisions, consider them in the same way you would any other business process or institutional knowledge. This is also why it's important for everyone on your team to have an understanding of and familiarity with generative AI — from the managers approving prompts to the users navigating the prompt library. 

Where to store your prompt library

Lastly, where and how should you store your prompt library? 

An unorganized prompt library (think: hundreds of prompts pasted into a Google Doc) offers maximum flexibility — but will leave your team hunting for the correct prompt. This also can lead to redundancies and failed versioning changes in your prompt library. Overly organized libraries, in contrast, can silo information and bury valuable prompts. 

Fortunately, where you store your prompt library, and the features that that system offers can help overcome these challenges. 

You can store your prompt library anywhere you might share another large repository of institutional knowledge: in a note-taking app, a shared file, an intranet or dedicated document management system. Anywhere that your team already stores files and shares knowledge. 

Whatever you select, key features to look for are:

• Search: No single organizational system will be perfect, so search is an essential feature of your prompt library. A robust search function will allow users to locate the right prompt, whether it's based on their task, function, or modality. 
• Tagging: Tagging with keywords can be another great way to help users sort, filter, search, or otherwise narrow down their library of prompts to the ones that best fit their needs. 
• Nesting: Consider folders, headings, or some other system for grouping or nesting prompts, depending on your team's specific needs. 
• Linking: The ability to inter-link sections of your prompt library may be useful. For example, to reference related prompts or jump between prompts and related context blocks. 

Once you've selected your prompt library’s organizational system and location, put a process in place to regularly review and update it. Like any file or knowledge management system, regular maintenance and feedback from your team can help ensure that it remains a useful and integral part of your team's day-to-day work. 

Have you organized your prompt library yet? What advice or lessons would you share? 

‍

➡️ P.S. If you found this post helpful, I’d appreciate if you left your feedback (or a like) on this LinkedIn thread. Thanks! 

‍

Editor's notes:

• ‍Content origin: All content (copy + images) was created by a human.
• This post was originally published in December 2023. It has been revised since then for accuracy and comprehensiveness.