“Which Parts Did the AI Write?” How to Answer Before Someone Asks

Sooner or later, a contracting officer, reviewer, or legal team will ask which parts of your proposal were AI-generated. Most teams can’t answer that today.

The problem usually isn’t careless AI use. It’s that the record never got built. Content moves through chat tools, shared drives, and personal accounts with no source tracking, no review history, and no documented disposition. When a question arrives, there’s nothing to point to.

This session walks through what an AI audit trail actually needs to contain, where it breaks in a typical proposal workflow, and how to close those gaps with process rather than prohibition. That includes the shadow AI problem directly: most teams don’t have a training gap, they have a policy gap.

What you will learn

• What provenance requires: The minimum viable audit trail has five fields captured at the section level: source material, tool and model, who reviewed it, what changed, and when it was approved.
• How to define AI-generated: Drafted by a model, edited by a human, summarized from past performance, and rewritten for tone each require different labels and different review expectations.
• Where the trail breaks: Copy-paste from unmanaged chat tools and shared drives breaks the record regardless of what your approved platform logs.
• Why policy beats training: A short, clear use policy covering allowed tools, data restrictions, and documentation requirements reduces risk more than another training session does.
• What data handling now means contractually: Whether your tools train on proposal content and where that content is stored has moved from an IT question to a contracting officer question.

Who should attend

Proposal managers, capture leads, and compliance officers at GovCon firms who use AI tools in their proposal workflow and need a defensible record for reviews, protests, or data rights disputes.