Here you’re going to learn the basics of how to write good AI evaluation criteria inside of Endorsed.
When you set up a job, Endorsed will automatically generate initial criteria based on your job description.
Overview of adding criteria
Criteria are the key points the AI evaluates. Similar to ChatGPT, our technology reads and understands plain English. Enter criteria just as you would write a prompt for ChatGPT.
Step 1: Remove bad criteria
First, remove criteria that are either too subjective, generic, or difficult to evaluate from a resume or custom questions.
Examples:
"Is a good communicator" (hard to assess from a resume)
"Has experience writing HTML" (expected for a software engineer)
Step 2: Add ideal candidate criteria
Next, add 2-3 criteria that define your ideal candidate. This helps prioritize candidates who closely match your ideal profile.
Examples:
"Worked in a similar industry"
"Founded a company"
Endorsed doesn’t hard-filter candidates but ranks them higher based on these criteria.
Step 3: Add dealbreaker criteria
Finally, add dealbreaker criteria that automatically disqualify unsuitable candidates.
Example:
"Is not currently in school" (important for non-intern/entry-level positions)
Use these criteria to bulk reject applicants who don’t meet essential requirements.
Pro tips to increase evaluation AI accuracy
Split criteria
Break down complex criteria into simpler components.
Instead of: "Has 4 years of experience with marketing and finance"
Use: "Has 4 years of experience with marketing" and "Has 4 years of experience with finance"
Be specific
Define criteria clearly to avoid ambiguity.
Instead of: "Not a job hopper"
Use: "Worked at their last two companies for at least two years each"
In some cases, you may find that your criteria is 2-3+ sentences long. In those cases, be sure to not contradict yourself.
Choose the right amount of criteria
Aim for 8-10 criteria to effectively rank candidates. Too few criteria can make many candidates appear similar, while too many can dilute the importance of key criteria.
Labels & weights
Labels
These replace the criteria prompt on the candidate’s card for easier readability, especially for long criteria.
Weights
Adjust the impact of each criterion on the candidate’s ranking.
"Strongly Preferred" doubles the weight relative to "Preferred"
"Unweighted" means the criterion doesn’t affect the ranking.
Next Steps
You now know the basics of entering AI evaluation criteria. Proceed to run your evaluation.
Your next step is to refine your evaluation criteria.