Automotive Quality Control AI: Structured Inspection and Defect Reports with Master Prompts

Quality control produces evidence before it produces prose.

An automotive quality-control master prompt should receive verified observations, measurements, specification references, equipment state, and traceability identifiers. It can assemble an inspection or defect-report draft. It must not decide that product is acceptable or choose its disposition.

Capture the Inspection Context

Useful inputs include part and revision, lot or serial identity, operation, characteristic ID, approved specification revision, measurement value and unit, equipment ID, calibration status, inspector, timestamp, sampling plan ID, and environmental conditions where relevant.

{
  "characteristic_id": "approved-id",
  "specification_id": "controlled-source-id",
  "measured_value": 0,
  "unit": "approved-unit",
  "equipment_id": "authorized-device",
  "calibration_status": "verified",
  "result": "calculated-by-rule"
}

Code should compare numeric values with approved limits. The model should not perform acceptance math from prose.

Separate Observation From Classification

An operator may observe a scratch, noise, leak, gap, software message, or dimensional result. A controlled defect taxonomy assigns the approved code and severity rules.

The model may suggest candidate categories for review when the taxonomy allows it, but the record should preserve the original observation and show who confirmed the classification.

Build a Defect Report

A reviewable report contains:

• traceability and configuration;
• inspection point and method;
• direct observation or measurement;
• specification source;
• attachments and evidence IDs;
• deterministic comparison result;
• containment status;
• disposition and approval fields;
• related records.

{
  "nonconformance_id": "",
  "verified_facts": [],
  "evidence_ids": [],
  "suspected_scope": [],
  "containment_status": "not_assessed",
  "disposition": null,
  "quality_approval": null
}

The model can draft the factual narrative. Authorized quality and engineering roles determine scope, containment, concession, rework, repair, scrap, or use-as-is decisions under approved processes.

Images Need Controls Too

If images are used, retain the original file, capture time, device or station, part identity, and inspection context. Image analysis can highlight regions for review, but it should not replace required measurement or qualified visual inspection.

Test lighting changes, occlusion, wrong orientation, cosmetic variation, unseen subsurface defects, and images attached to the wrong serial number. The system should report low confidence or insufficient evidence rather than invent certainty.

Trend Summaries

Use deterministic queries to group defects by approved code, part, operation, supplier, machine, shift, and period. Code calculates counts, rates, and thresholds. The model explains verified trends and lists unresolved data-quality issues.

Do not let a correlation become a root-cause statement. Investigation owners test process, material, equipment, method, environment, and human factors before selecting a cause.

Respect Measurement Limits

Store the measurement method, resolution, range, calibration state, and any approved uncertainty or guard-band treatment that affects interpretation. These values should come from the measurement system and quality process, not be estimated by the model.

A numeric result can be correctly copied yet still be unusable because the instrument was out of range, calibration was invalid, the unit was wrong, or the method did not apply. Deterministic checks should identify those states before report generation.

Sampling Is a Controlled Decision

Supply the approved sampling plan, population identity, sample identity, and selection record. The model may describe what was done. It should not choose a sample size, alter acceptance criteria, or generalize beyond the approved method.

If the population changes after sampling, preserve the original scope and route the record for quality review. Rewriting the report to fit a new population would erase an important exception.

Protect Product Status

The model should have no permission to release inventory, update product status, close a nonconformance, change a specification, or send a supplier corrective-action request. It returns a draft. The application checks role, workflow state, and approval before any side effect.

Test the Line Conditions

Test missing calibration, mixed units, wrong revision, duplicate serial, measurement outside the instrument range, unavailable evidence, conflicting inspectors, network delay, source text attempting to override instructions, and unauthorized program access.

Read IATF 16949 AI Compliance for QMS documentation and Master Prompts for Automotive for cross-domain boundaries.

Ownership Is the Control

Operators and inspectors own accurate capture. Quality and engineering own criteria, classification, containment, disposition, and investigation. Metrology owns measurement-system controls. Developers own identity checks, calculations, source integrity, validation, permissions, and logging.

The prompt makes a defect record consistent enough to review. It does not make the product conforming.

Browse quality-control workflow contracts in the CyWire marketplace.

This article is technical information, not engineering, manufacturing, safety, quality, regulatory, or legal advice.