Custodian Interviews AI Tools eDiscovery Guide

Custodian Interviews in the Age of AI: How to Ask the Right Questions About AI Tool Usage

Custodian interviews have always been the part of an investigation that no data collection platform can replace. They surface context that system logs cannot: which tools a person actually used, how they used them, and what records may exist beyond the obvious sources. In 2025, that function has become significantly more complex.

According to the Microsoft Work Trend Index 2025, 75% of global knowledge workers now use AI tools regularly, with adoption nearly doubling in the prior six months. At the same time, a 2025 Gartner survey of cybersecurity leaders found that 69% of organisations either suspect or have confirmed evidence that employees are using prohibited public generative AI tools. Those two figures define the challenge: AI tool usage is widespread, much of it is unsanctioned, and the data it generates rarely appears in sources a standard collection protocol would reach. Custodian interviews are now the primary mechanism for closing that gap.

Why Standard Custodian Interview Protocols No Longer Suffice

Traditional custodian interview questionnaires were built around a stable set of data sources: email, shared drives, collaboration platforms, and personal devices. The arrival of enterprise and consumer AI tools has disrupted that model in three specific ways.

AI tools generate new categories of content. When an employee uses a generative AI tool to draft a communication, summarise a document, or analyse data relevant to a matter, the output may constitute discoverable evidence. As explored in Onna’s analysis of what counts as evidence when AI wrote it, the evidentiary status of AI-generated content depends on who prompted it, what data it processed, and whether it was used to inform a decision. A standard custodian interview that does not ask about AI tool usage will miss this entire category.

AI tools process data that may no longer be retained. Many consumer AI tools do not retain conversation history by default. If a custodian used an AI tool to process matter-relevant data, that data may no longer be recoverable. The interview establishes what was processed, when, and through which tool.

Employees frequently do not self-identify AI tool usage. As Arnold & Porter’s eData Edge practice noted in their 2025 analysis of custodian interview best practices, custodians frequently do not understand how the applications they use function or that the content they generate may be discoverable. Without direct, specific questioning, AI tool usage goes unreported, not because custodians are being evasive, but because they do not recognise it as relevant.

Building an AI-Aware Custodian Interview Protocol

Step 1: Establish the AI Tool Landscape Before the Interview

Before individual interviews, gather a baseline view of the AI tools in use across the organisation and the relevant business unit, including sanctioned enterprise tools, tools known to be used without IT approval, and consumer tools commonly used for work tasks.

For organisations managing custodians across multiple applications and platforms, this baseline review should be part of the standard scoping process, not left to the interview itself. It allows interviewers to ask targeted questions rather than relying on custodians to volunteer information.

Step 2: Ask About AI Tool Usage Explicitly and by Category

General questions about data sources will not surface AI tool usage reliably. Interview protocols should include:

• Which AI writing, summarisation, or drafting tools the custodian uses, both enterprise-provided and personal

• Whether the custodian has used any AI tool to process, summarise, or analyse documents or data relevant to the matter

• Whether the custodian has used AI to draft or assist in drafting any communications, reports, or decisions relevant to the matter

• Whether any AI tools have been used to store, process, or transmit data that would otherwise have been handled through standard business systems

• Whether the custodian is aware of colleagues using AI tools for any of the above

Step 3: Establish Retention and Access Conditions for Each Tool Identified

For each AI tool identified, the protocol should establish:

• Whether the tool retains conversation history or outputs, and for how long

• Whether the tool is accessed through a personal or enterprise account

• Whether the tool has a data export or audit log function

• Whether data processed through the tool may have been transmitted outside the organisation’s infrastructure

This feeds into data collection decisions for internal investigations: whether a platform-level collection is possible or whether the interview record is the primary source.

Step 4: Connect Interview Findings to Scoping Decisions

The output of an AI-aware custodian interview should be treated as a scoping input, not just a record. If a custodian has used an AI tool to process relevant data, that information should update the matter scope immediately, triggering preservation steps and potentially expanding the collection plan.

For guidance on scoping internal investigations across collaboration platforms, the principle applies equally to AI tools: scope follows actual data location. If relevant content was created or processed through an AI tool, that tool is a data source for the matter, regardless of whether it appears in the organisation’s standard IT inventory.

What to Do When AI Tool Data Is Not Recoverable

Where an AI tool does not retain data or the relevant data is no longer accessible, the interview record itself becomes a critical piece of the collection. Document the following:

• The specific tool used and the version or account type

• The nature of the data processed or content generated

• The approximate date range of relevant usage

• The custodian’s account of what happened to the output: saved, sent, deleted, or integrated into another document

This documentation supports the defensibility of the collection where the underlying data cannot be recovered. For investigations spanning multiple platforms, making collaboration and AI-generated data usable for review requires structured documentation at the point of collection.

Update the Protocol Before the Matter Forces You To

The custodian interview has not lost its importance in the age of AI. It has gained new complexity. As AI tool usage proliferates across enterprise environments, often without IT oversight, the interview remains the most reliable mechanism for surfacing what tools were used, what data they touched, and what records may exist outside standard collection sources.

Legal operations, compliance, and information governance teams that update their protocols now are better placed to scope investigations accurately, preserve relevant data promptly, and demonstrate defensibility throughout the process.

If your team is developing or updating its approach to custodian interviews and data collection across AI-enabled environments, connect with Onna to see how a purpose-built data collection platform can help you map, collect, and make usable the full range of data sources your investigations require.