# Screening rules by conversation

The Intervieux agent drafts screening rules from a plain-language description, checks your real field names and job setup first, back-tests the draft against recent completed interviews so you see a match rate before anything is saved, and only creates the rule once you approve it.

The rules engine underneath is genuinely capable: nested AND/OR logic, five scoring dimensions plus challenge results, salary and experience filters, five action types. That capability is also why a rules builder UI can be slow to work in when you are not sure of the exact field names or what a rule would actually catch. Describing the rule in a sentence and letting the agent look up the real fields and test the logic against real candidates first removes most of that friction, without changing what the engine underneath can do.

## What screening rule authoring covers

Six tools cover the full cycle. list_screening_rules shows what already exists, in priority order, so a new rule does not duplicate one you forgot about. get_hiring_context pulls your open jobs, interview templates, challenges, and the engine's actual field, operator, and action catalog, which the agent is required to check before drafting anything, so a proposed condition never references a field or challenge id that does not exist. test_screening_rule runs the draft against your own recent completed interviews without saving it, returning how many candidates it would have matched and a sample of who. create_screening_rule saves the rule once you have seen that number and approved it. update_screening_rule and delete_screening_rule handle changes and removal, both scoped to a rule you name explicitly.

## A walkthrough: turning a hiring habit into a live rule

An employer says: "I want anyone who scores above 8 overall and solves our take-home challenge to get tagged strong_candidate and moved to qualified automatically." The agent calls get_hiring_context first, confirms the take-home challenge's real id and that overall score is a valid field, then drafts the two-condition rule with two actions. Before saving anything, it calls test_screening_rule and reports back: "this would have matched 7 of your last 40 completed interviews, including these three." The employer asks to tighten it to a score above 8.5, the agent re-tests and reports 4 matches, and only then does the approval card appear: "Create screening rule 'Strong technical + challenge pass': 2 condition(s), 2 action(s)?" Once approved, the rule runs automatically on every interview that completes from that point forward.

## What the agent will not skip

A rule is never created without first being back-tested against real recent interviews, and the match numbers are shown to you before the create step, not after, since the whole point of the back-test is to catch a rule that's too broad or too narrow while it still costs nothing to fix. Creating, updating, or deleting a rule always requires your approval, and update or delete only happen when you explicitly name the rule to change, never as a side effect of drafting something else. Deleting a rule cannot be undone, and the approval card says so plainly before you confirm it. The agent never invents a field name, operator, or challenge id; get_hiring_context is the only source it draws from, so a drafted rule references only things that genuinely exist in your account, and a rule referencing a challenge you retired months ago simply won't get drafted in the first place.

## Frequently asked questions

### Can the agent create a screening rule without showing me how many candidates it would match?

No. test_screening_rule runs first and reports a match rate against your recent completed interviews before create_screening_rule is ever called, so you see the numbers before the rule exists.

### What fields can a screening rule use?

All 5 AI scores, challenge score, experience level, salary estimates, key skills, and strengths or weaknesses, combined with 14 operators and nested AND/OR logic up to 5 levels and 25 conditions. The agent pulls the real field catalog before drafting so it never guesses at a name that doesn't exist.

### Can I ask the agent to change an existing rule instead of writing a new one?

Yes. update_screening_rule changes fields, conditions, actions, priority, or the enabled state on a rule you name. It's used for that case specifically, not for drafting a rule from scratch.

### Do screening rules run automatically once created?

Yes, a saved rule runs automatically as each interview completes going forward. It does not retroactively apply to interviews finished before it existed.

### What actions can a screening rule actually take?

Five: tagging a candidate, setting their pipeline stage, setting a priority, flagging them, or sending a notification. A single rule can combine more than one action, as long as its conditions actually match.

## Related pages

- [Your AI recruiting agent](/recruiting-agent)
- [Pipeline automation](/recruiting-agent/pipeline-automation)
- [Hiring funnel analysis](/recruiting-agent/hiring-analytics)
- [Durable org memory](/recruiting-agent/org-memory)
- [Frequently asked questions](/faq)

## Turn a screening habit into a rule

Describe the condition to the agent, review the back-test numbers, then approve it once the match rate looks right.

Start practicing free: https://www.intervieux.ai/register · Hire with Intervieux: https://www.intervieux.ai/employers/signup