AI Requirements Gathering: Faster, Cleaner Elicitation

In my BA practice, requirements gathering is the work that defines the role. Done well, it produces clear, validated, traceable requirements that lead to successful implementations. Done poorly, I’ve seen it produce ambiguous specs that drive scope churn, rework, and stakeholder frustration. AI changes the math. In my experience the same elicitation effort produces sharper, more complete, more validated requirements - and I can run more elicitation cycles within the same project budget.

In this guide I cover how I use AI to augment every stage of requirements gathering, the prompts and patterns that have worked for me, and the failure modes I’d avoid.

The synthesis stage sees the largest compression. Documentation also compresses dramatically.

AI helps:

• From an org chart and project description, suggest a stakeholder register.
• Identify likely influencers, blockers, and beneficiaries.
• Suggest interview priorities and sequencing.

A useful prompt:

“From this org chart and project description, suggest a stakeholder register. For each: role, likely influence, likely interest, suggested interview priority, key concerns based on the project’s nature.”

The BA validates with on-the-ground organisational knowledge. AI suggests a starting framework; the BA fills in political dynamics.

For each interview, AI generates:

• Tailored interview guides based on the stakeholder’s role.
• Probing questions for known concerns.
• Background research summary.
• Question sequencing for emotional and informational flow.

A useful prompt:

“I am interviewing the CFO about a project to modernise the financial reporting system. Generate a 45-minute interview guide. Include: opening rapport-building questions, current-state pain questions, future-state vision questions, constraints and concerns, closing questions about success criteria. Tone: respectful of executive time, deeply curious.”
 

AI does not run interviews well; humans do. AI augments by:

• Capturing transcripts (Otter, Fireflies).
• Suggesting follow-up questions in real-time (with care - can be distracting).
• Running parallel async elicitation (surveys, async interviews) at scale.
• Generating summary notes immediately after the session.

For most BAs, the live session remains classical. The augmentation is at the edges.

This is where AI delivers the most value. Patterns:

Theme clustering: AI groups raw input into themes. The BA validates and names themes.
Contradiction detection: AI surfaces contradictions across stakeholders that the BA might otherwise miss.
Gap detection: AI compares requirements against a template (security, compliance, performance) and flags gaps.
Frequency and prioritisation: AI tracks how often themes appear and across which stakeholders.

A useful synthesis prompt:

“Below are 18 stakeholder interview transcripts. Cluster requirements into 8-10 themes. For each: name, frequency across stakeholders, dominant viewpoint, contested points, supporting quotes. Flag contradictions across stakeholders. Identify gaps where standard categories (security, performance, scalability, accessibility) are unaddressed.”

Strong BAs always read at least 10% of source material manually to validate AI’s clustering.

AI generates:

• BRD or FRS drafts from synthesised themes.
• Use cases or user stories from requirements.
• Acceptance criteria expansion.
• Traceability matrices.

The pattern from AI BRD Generator covers BRD specifics.

A useful prompt:

“From these synthesised requirements, generate a BRD. Sections: business context, scope, functional requirements, non-functional requirements, assumptions, constraints, dependencies, acceptance criteria. Tone: precise. Length: 3,000 words. Maintain traceability to source interviews.”

Validation is where AI helps verify requirements meet stakeholder needs:

• AI generates validation questions per requirement.
• AI scans requirements for ambiguity.
• AI suggests validation scenarios.
• AI tracks sign-off status across requirements.

A useful prompt:

“Below are 80 requirements. For each, generate one validation question that confirms the requirement matches stakeholder intent. Identify any requirement that is ambiguous and suggest specific clarification.”

The BA runs validation sessions with stakeholders armed with AI-prepared questions.

For most BAs starting out, a meeting capture tool plus Dovetail AI plus a general LLM covers 80% of needs.

These are the patterns I see most often in AI-augmented elicitation. I’d flag the first two as the ones that have bitten teams I’ve worked with hardest.

• AI fabricated quotes. I’ve seen this break stakeholder trust quickly. Always pull verbatims from real transcripts.
• Generic clustering. AI tends toward generic theme names. I push for specifics every time.
• Missing the surprise. AI clusters can flatten the most important user voice. I always read manually.
• Skipping validation. AI synthesis is a draft. Stakeholder validation remains essential.
• Documentation drift. AI-generated requirements that no one reviews drift from reality.
• Over-formal documentation. AI tends to generate verbose BRDs. Trim aggressively.
• No traceability. Without traceability, requirements changes cascade unpredictably.