How to create an application screening agent in Microsoft Copilot

João Barros 01 de October de 2026 4 min read

This tutorial shows how to create an agent in Microsoft Copilot that performs initial screening of applications (CVs/resumes) and identifies the best candidates according to defined criteria. It is useful to reduce manual work and speed up recruitment, with examples of prompts, configuration and basic integration.

Prerequisites

Account with access to Microsoft Copilot / Copilot Studio.
Set of sample CVs in text or PDF files.
List of evaluation criteria (e.g.: years of experience, skills, location).
Basic knowledge of prompts and API usage (optional for automation).

Step 1: Define criteria and output format

Before creating the agent, clarify the criteria of interest: mandatory skills, minimum years of experience, certifications, willingness to travel, language. Decide the output format you want — for example: Score (0-100), skill tags, 2-sentence summary and recommendation (Yes/No/Maybe).

// Example of JSON output format the agent should produce
{
  "score": 0-100,
  "recommendation": "Sim|Talvez|Não",
  "matched_skills": ["Azure","Power BI"],
  "summary": "Resumo de 1-2 frases"
}

Step 2: Create the agent in Copilot Studio

Go to Copilot Studio and choose to create a new agent/assistant. Name it, for example: "TriagemCandidaturasBot". In the initial state, focus on configuring the personality and system instructions with clear language and limits of responsibility (e.g.: "This agent performs initial screening, it does not replace human assessment").

System prompt exemplo:
"És um assistente de triagem de candidaturas. Lê o CV, extrai competências e experiência, e devolve um JSON com score, recommendation, matched_skills e summary. Usa critérios: 3+ anos de experiência, 'Azure' e 'Power BI' como skills desejadas, residência em Portugal preferida."

Step 3: Create analysis prompts and test with samples

Create a standard prompt that includes the CV (extracted text) and the criteria. Test with several samples to adjust how Copilot weights each criterion. Use clear instructions to avoid free-form responses and ensure the JSON format.

Prompt de entrada exemplo:
"Analisa o texto abaixo (CV). Usa os critérios: 3+ anos de experiência, competências: Azure, Power BI. Produz apenas o JSON no formato especificado. CV: "

Step 4: Adjust weights and rules (prompt engineering)

If results do not reflect the correct priority, adjust the prompt to assign weights or explicit rules. For example, make skills mandatory or increase the impact of years of experience on the score calculation.

// Exemplo de regra no prompt para cálculo de score
"Regra: base = 50. +20 se tiver 3+ anos. +15 por cada competência chave encontrada (Azure, Power BI). -10 se residir fora de Portugal. Normaliza para 0-100."

Step 5: Integrate with the application intake workflow

To automate, integrate the agent with where you receive CVs (email, form, ATS). Initially you can copy/paste CVs in tests. For automation, use the Copilot API or Power Automate connectors to send the CV text to the agent and store the resulting JSON.

// Pseudocode básico para chamar a API do Copilot (exemplo genérico)
POST /copilot/agents/{agentId}/invoke
Headers: Authorization: Bearer 
Body: { "input": "Analisa o CV: " }

Verify the output

Validate the outputs with a set of known CVs. Check that the JSON is well-formed, that the scores make sense and that the recommendation aligns with a human review. Test edge cases: short CVs, CVs with many irrelevant keywords, and PDFs with poor text extraction.

Conclusion

After testing, adjust rules and integrations to reduce false positives and ensure transparency in decisions. As next steps, consider: connecting the agent to your ATS, creating logs for auditing and adding human feedback to improve the model. Tip: start with clear rules and then refine the prompt as problematic examples arise — what is the first criterion you want to prioritize?