Trust and Transparency
Inline Evidence
Show the evidence for each claim right next to it, so people can judge it without leaving the page.
Overview
People trust outputs when evidence is immediately visible, not hidden behind generic confidence scores. Listing sources at the bottom fails because it leaves individual claims unlinked, allowing hallucinated details to blend in unnoticed. Inline Evidence embeds the citation directly alongside each claim or result, which also makes missing evidence obvious. This matters most for high-consequence decisions, such as financial or legal ones.
When to use
Use when
- Verifiable records: answers are built from documents, data or records users can independently verify, such as contracts, policies or case files.
- Consequential claims: outputs will be acted on directly or shared with external stakeholders, and have high consequences, such as financial or legal ones.
Not needed when
- Generative drafting: the content is purely creative or exploratory, with no factual source material to cite.
- Weak relationships: sources only loosely relate to the specific statement. Weak citations degrade trust faster than omitting them.
Questions to ask
Ask these at kickoff, when design work starts. Tick them off as your team answers.
Example
An AI extracts key terms from a vendor contract. Each field carries its evidence.
See it in context · full screen
4.1 Invoices shall be issued monthly in arrears.
4.2 Payment shall be due within forty-five (45) days of receipt of invoice.
4.3 Late payments accrue interest at 1% per month.
- 1Each extracted term carries its supporting passage right underneath it.
- 2Selecting a term opens the source at that passage, highlighted, without leaving the page.
- 3A term with no supporting passage is flagged instead of looking as solid as the rest.
Anti-pattern: Detached sources
The same summary with a list of sources at the bottom. Nothing connects a term to the passage it came from, so the unsupported termination notice looks exactly as solid as the rest.
Anti-patterns to pay attention to
- Evidence lives in a separate tab, panel or download.
- Sources are listed for the whole answer, not per claim.
- A claim with no supporting evidence looks the same as one with evidence.
References
Current guidelines · EU AI Act2 references
For orientation only, not legal advice. Applies to high-risk systems. EU AI Act links open a readable version of the official text published on EUR-Lex; ISO/IEC 42001 is a paid standard.
- EU AI Act Art. 13(3)(b)(vii) Instructions for use must include information that helps deployers interpret the system's outputs.
- EU AI Act Art. 14(4)(c) People overseeing the system must be able to interpret its results correctly.
Evidence
Survey & researchSurvey insights, respondent quotes
“Aside confidence score, being able to see attached sources from which the AI referenced helps in both verification and trust of the answer provided.”
Survey respondent, open-ended answer
Survey insightSeeing the source documents was the second most popular form of transparency, after a plain explanation. Combined, the two were the clear majority choice over a percentage.
The survey is still open. Add your perspective ↗