Structuring AI vs Generative AI: Why the Distinction Matters
The rapid adoption of AI tools in regulated sectors has outpaced the development of governance frameworks. As a result, organisations are deploying AI without a clear understanding of the different risk profiles of different types of AI tool. The distinction between generative AI and structuring AI is fundamental to managing those risks.
What Generative AI Does
Generative AI creates new content. In the context of professional documentation, this typically means:
- Transcription tools that listen to conversations and generate text from audio, introducing errors from accents, background noise, and interpretation.
- Summarisation tools that read existing text and generate shorter versions, deciding what to include and what to omit, and choosing replacement language.
- Drafting tools that generate entirely new text from prompts, drawing on training data that may contain biases, outdated information, or irrelevant patterns.
Each of these introduces content that did not originate from the practitioner. The AI makes decisions about what to say and how to say it. Research has shown that these decisions can include fabricated information (hallucinations), systematic bias (gender, racial, or cultural), and misrepresentation of facts.
What Structuring AI Does
Structuring AI organises existing content. The practitioner provides all the input — their notes, observations, and professional assessments — and the AI arranges this into a professional format aligned with relevant regulatory frameworks.
The critical difference: no new content is generated. The practitioner's words, assessments, and professional judgements pass through the system and emerge organised, not rewritten. The AI's role is limited to structure, formatting, and framework alignment.
The Risk Profile Comparison
| Risk | Generative AI | Structuring AI |
|---|---|---|
| Hallucinations | Documented in research | Not possible — no content generated |
| Gender/racial bias | Demonstrated in LSE study | Not possible — language preserved |
| Transcription errors | Inherent to audio processing | Not applicable — no audio input |
| Client consent | Required for recording | Not applicable — no recording |
| Data retention | Up to 120 days with sub-processors | No retention after processing |
| Professional judgement | AI interprets and decides | Practitioner retains full control |
Why This Matters for Governance
Organisations procuring AI documentation tools should assess which category the tool falls into. A generative tool requires extensive safeguards: Data Protection Impact Assessments, bias auditing, hallucination monitoring, consent protocols, and robust human review processes. A structuring tool carries a fundamentally different — and lower — risk profile because the failure modes of generative AI do not apply.
This does not mean structuring tools require no governance. They still process personal data, require data protection compliance, and should operate within clear professional standards. But the specific risks that have been identified in recent research — hallucinations, bias, misrepresentation — are eliminated by design rather than managed by process.
Recommendations
- Classify AI documentation tools as either generative or structuring before procurement.
- Apply proportionate governance: higher safeguards for generative tools, appropriate controls for structuring tools.
- Ensure practitioners understand which type of tool they are using and what its limitations are.
- Require transparency from vendors about whether their tool generates new content or structures existing content.
Sources: Ada Lovelace Institute / University of Oxford (2026); LSE research on gender bias in AI social care tools (2025); GOV.UK Algorithmic Transparency Records.