Amplify Prompting Best Practices
Bullhorn Amplify Assistant and Amplify Chat both support custom prompts to help staffing teams streamline their workflows. Unlike external AI tools, these operate directly within Bullhorn ATS, ensuring alignment with existing workflows. This article contains tips and best practices for building effective prompts, either ad-hoc on records or via the Amplify Prompt Studio.
Getting Started: Implementation Tips
Start with Recruiter Feedback
Meet with recruiters to understand where they spend time creating or reworking content. Build prompts around their most common or time-consuming tasks. Focus on small, high-impact use cases to build trust and adoption.
Standardize the Prompt Library
Create, test, and refine your prompt library before rolling it out. Organize prompts by use case, role type, or workflow to make them easy to find and apply consistently.
Encourage Testing and Editing
Train recruiters to review and revise AI-generated content before saving. Reinforce that the generated output is a starting point, not a final product, in order to build confidence and ensure accuracy.
Prompting Best Practices
Data Quality
A foundation of good data is the most important factor in Amplify generating a useful message. Make sure that your Candidate, Contact, and Job records are fully filled out. In addition, select the appropriate fields to be included in the context to the LLM
An LLM, or Large Language Model, is a deep learning model that has been pre-trained on vast amounts of data and is used to power generative AI. in the Data section of the Amplify Admin > Prompts tab.
Choose the Right Role
Roles are used to guide the LLM’s understanding of a prompt and to help it generate more accurate and relevant outputs. There are pre-defined roles within Amplify Admin. If none of these fits your prompt’s needs, you can add more roles via the Data Management tab. When creating a new role, be specific and consider the context in which the role will be used. Roles should be relevant to the task or question being asked.
Iterate
Take advantage of the opportunity to generate and regenerate when editing prompts in the Prompt Studio. As you refine your prompts, it is important to test them by generating, making some tweaks, generating again, and so on. You can generate your prompts multiple times using actual Bullhorn records before saving in order to refine the parameters and get to the output you need.
Converse
After using a prompt with Amplify Chat or Amplify Assistant, follow your request up with additional context by conversing with Amplify as if you were speaking to a person. As you provide more information, Amplify will continue refining its response for you. Be specific, use natural language, and provide context that may not have been included in the original prompt to tailor Amplify's response to your needs.
Still Not Seeing Expected Results?
If you're following the best practices above and still not getting the results you need, we recommend starting from scratch. The LLM may be receiving conflicting instructions from the role, prompt, or follow-up messages.
Here are a few tips we've had the most success with:
Create a New Role
Roles help shape the LLM's behavior, so a new role can make a big difference. From the Data Management tab in Amplify Admin, you can create a new role to define how the LLM should act.
A typical role defines the point of view or voice the LLM should use.
“You are an expert recruiter for a healthcare staffing firm. Be sure to write professionally.”
You can also include specific grammar and formatting rules if they're relevant at the role level, such as:
-
Do not use any sentences from the original job description verbatim.
-
NEVER state the name of the client or hiring company.
-
NEVER state the specific industry for the job.
Create a New Prompt
Try building a brand new prompt in the Prompt Studio. A typical prompt defines what the LLM should do.
“Write an offer letter to the candidate. Sign the email with the recruiter’s contact information.”
When writing your prompt, include clear, detailed instructions. If there's a specific call to action or template you want the LLM to follow, include that directly.
“At the very end, always include this exact ending: "Not ready to apply? Contact Bullhorn Recruiting at bullhorn@recruiting.com or call 555-234-2345""
Simpler templates tend to perform best. From the Prompt Studio, you can test and iterate until you find a template that works well with your data.
Include all Relevant Data Points
Don't forget the data! In the Prompt Studio, data points must be selected as "default" to be included in the preview pane on the right side of the screen.
-
For primary entities, only default-checked fields are used.
-
For secondary entities, all selected fields are automatically in scope.