Prompting, or crafting a specific query or question to guide an AI model, has quickly emerged as a vital skill for individuals who want to get the most relevant response from an AI tool. But learning how to prompt effectively can take effort, and it can vary depending on the product.
Prompting Search Copilot with thoughtful questions, clarity, and precision ensures you get the answers you need as quickly as possible.
When you prompt our technology, a few things happen in rapid succession. Search Copilot uses a custom mix of our own proprietary models as well as RAG (Retrieval Augmented Generation) to identify the data that is relevant to your query. We then use a large language model to process that information and generate a response.
This process changes as the bar for “magic” keeps moving up. We frequently update our tool with new features and capabilities, and we will continue to rapidly innovate.
When you prompt Search Copilot for an answer, it’s scanning your entire knowledge base as well as all the software integrations you’ve connected, whether that’s your Read AI meeting reports, Google Docs, Gmail, Outlook, Slack, HubSpot, Notion, and more.
The tool is also flexible enough to support prompts from various roles and team members—product leaders, engineers, marketers, customer service agents, and every other collaborator. For that reason, prompting well may look different depending on your role, but general themes apply to everyone.
To ensure you can find the most value right away, follow these tips for writing excellent prompts and you’ll get the most relevant and actionable results.
Filters allow you to specify a time frame and/or platform for your search, helping to focus the data and ensures that you get your answer even more quickly.
For instance, if you're asking about a long-term client but are only interested in your most recent conversations, apply a filter for the last 90 days upfront and you’ll narrow the pool of information. You can also filter to content that you created (versus content that is shared with you), as well as only search content that you did not create. For example, you can limit your search to meetings you have access to, but might not have attended. This gives you an easy way to glean insights from shared reports and documents without any heavy lifting.
In the same vein, straightforward and well-defined queries are always best. For example, asking, “What is the status of [project X]?” can yield immediately impactful results. Similar queries to understand project limitations, topics discussed, foundational research, and other relevant information are best served with clear inquiries. Start your work by imagining the output, and work backwards. This will give you stronger questions out of the gate.
While Search Copilot can handle detailed inquiries, it’s important to keep in mind that complex or layered questions may require follow-ups or refinements to get the most precise answers. Keep at it—you’re still saving time. And remember, all answers can be easily shared with others, while the foundational documentation is kept private to you by default (meeting and document titles will be visible, however).
A major benefit of Search Copilot is that it is built specifically to support the examination of Meeting Reports in addition to all the other information types (documents, emails, messages, and more). That’s because we recognize that meetings are where crucial decisions and insights are made, yet this valuable information tends to remain locked away, accessible only to those who were present.
When prompting for data within a meeting report, it’s best to include the specific meeting name in your query, or the date when the meeting transpired. If you're only looking for answers from a single meeting, the best way to find what you’re looking for is to go to the meeting report page, and use the "Search this report" filter.
Because Read AI has access to your entire knowledge base, as well as your Meeting Reports and transcriptions, it can uniquely provide insight into subject matter expertise. You can prompt the tool to provide details on who might be the right resource on a particular topic, as well as the details on that specific topic. When prompting the technology, avoid words such as “the most knowledgeable” in favor of “who knows about…” which will provide more options for consideration.
Whether you want Read AI to write its response in the form of an email, a detailed report, a memo, or a poem, it can deliver—when you’re clear about the output and format you’re looking for. Ask it to draft the response in an email form, or as an outline, or with summarized bullets. This ensures you get the exact response you need without having to reorganize and edit, making it even easier to share insights with relevant stakeholders and ensure everyone is on the same page.
There is always more to know about your workplace, whether that’s details about a specific product or an acronym you’re less familiar with. Search Copilot can evaluate a team’s entire knowledge base and provide clarity on the function of a certain product, feature, or term. When prompting for this type of information, a standard prompt such as “What is…” or “Tell me about…” works well.
When it comes to prompting, precision is one of your sharpest tools. If you have a specific meeting, document, or batch of emails in mind when asking a question, include it in your prompt. Asking it about specific sources can ensure it’s focusing on the right information. For example , you might say something like, “What are the biggest concerns from my meetings with [insert client], or ‘What’s the latest status after the [insert date] meeting with [insert person].”
If your first prompt doesn't get you the response you expect, try rephrasing your query, adding more context, or asking follow-up questions to drill down into the information you need. For example, if your query is, "What are my client’s current priorities?" returns too many results, try adding more specifics: “What priorities did our client raise during our review meeting last week?”
You may also want to query Search Copilot a few times over a few days, if the project or topic is rapidly changing. Our internal authorization service runs half a billion checks a day, so the information will be updated alongside your other documentation.
This is just the beginning of internal knowledge search at Read AI. In the near future, we’ll be adding analytical queries (such as counts and metrics), continuous chat, and many more powerful features that are in development now. Stay tuned! If you have feedback to share, you can always reach our product team here. We’d love to hear from you.
Straightforward and well-defined queries are always best. Start your work by imagining the output, and work backwards. This will give you stronger questions out of the gate.
© 2024 Read AI, Inc.
%2C%20Color%3DOriginal.png)
