> Requirements, evidence, missing information and search reports: a method for using AI with your database.

Source: https://hice.ai/en/blog/web-ai-ai-recruiting
# Search and compare candidates with ChatGPT or Claude

- Define requirements before requesting a shortlist
- Compare people using evidence and open questions
- Summarize a search's progress
- A conversation does not replace a database
- Reuse the database in your next search
- Tools to put the workflow into practice

## Define requirements before requesting a shortlist

Separate necessary requirements from preferences: skills, experience, location and verifiable constraints. Ask the assistant to search accessible Hice data and show evidence and missing information for each candidate. Availability should not be inferred merely because a profile is public.

Connector search and model analysis are not automatically the same engine as Hice's native matching. Check available operations before requesting a function. If an attachment or semantic search is inaccessible through the connector, use the app's supported flow rather than expecting the model to read content it has not received.

## Compare people using evidence and open questions

Use a table with requirement, profile evidence and information to verify. Request an explanation for each suggestion rather than only a score. A useful request is: 'Compare these three candidates for this search and prepare questions that clarify their differences.'

Keep facts separate from interpretations. 'Has worked with Java' can be documented; 'will be the best team member' needs more evidence. A shortlist is a proposal for the recruiter rather than an automatic hiring decision. Review the supporting information before sharing a summary with the client.

## Summarize a search's progress

For a recruiting update, identify the search and period. Request linked candidates, completed steps, open activities and missing information. Specify whether the report is for the internal team or the client, as the useful level of detail may differ.

Check counts against the same record set. Saved profiles are not the same as contacted people or completed interviews. Ask ChatGPT or Claude to draft a narrative from accessible data, then verify names and status before sharing the document.

## A conversation does not replace a database

Chat can help organize information, but a conversation alone does not create a shared system of records, relationships and permissions. Paste a candidate list and the assistant works with that copy; it does not automatically receive subsequent changes in your business software.

With Hice connected, the assistant consults accessible records through connector operations. The database remains the place for identities and relationships; chat is an interface for reading, analyzing and proposing actions. For records just saved through the extension, request a fresh read: an earlier answer does not update itself.

## Reuse the database in your next search

Before searching for new people, consult the profiles you already collected. Use explicit requirements and verify that information remains current. Someone not selected for one assignment may suit another; preserve the reason for the earlier assessment without treating it as a permanent judgment.

Measure the whole process: saved profiles, complete records, contacted people and candidates actually evaluated. Record count alone does not measure database value. AI connected to Hice can help read accessible records, but missing information requires verification or a conversation with the person.

## Tools to put the workflow into practice

- ATS with ChatGPT
- ATS with Claude
- Candidate database

## Put this guide into practice with free software

HICE connects candidates, clients, consultants, timesheets and expenses. The free plan includes up to 10 employees and 5,000 CVs for companies, and 500 CVs for freelancers.

### Related solutions

### More Articles

#### Create AI business reports from your software records

Define filters and formulas for sales, recruitment and project reports without relying on prebuilt dashboards.

#### Company API keys for AI chat: OpenAI and Anthropic

Two AI workflows, administrative configuration and API costs separate from chat subscriptions.

#### CRM and ATS through MCP: operations and approvals

What to test in a business connector and how to distinguish reads, writes and app functions.

#### How to organize prospects found on the web

From company website to CRM record: retain the source, link contacts and define the next step.

#### Prepare billing data with AI: what to verify

Check hours and rates and distinguish drafts, billing status, issuing and sending the document.

#### Choose a candidate or client database: a practical trial

Criteria to check for a free database, an extension and a connection to AI assistants.