Governance, Ethics & Legal Practical insights
AI testimonials need a real customer behind them
Verify customer experiences before publishing AI-assisted testimonials, preserve the original statement and distinguish a fictional example from evidence.
A testimonial is evidence of someone’s experience. Generative AI can produce convincing language without that experience existing at all. Before optimizing a quotation for a landing page, establish who said it, what happened and what the organization has permission to publish. Plausible wording cannot supply missing evidence.
Start with the original statement
Keep the interview recording, written feedback or approved case-study response. Record the context: product version, use case and relevant period. If the only source is an AI prompt requesting a happy customer quotation, there is no customer testimonial to edit. Use a clearly identified hypothetical example elsewhere instead of assigning invented praise to a fictional customer.
Protect meaning during editing
AI may shorten a genuine statement, but compare the result with the original. Watch for stronger results, removed qualifications and implied certainty. ‘Our team found onboarding easier’ must not become a quantified productivity claim without supporting evidence. NIST’s generative AI risk profile discusses information-integrity risks; this review is a practical editorial safeguard against them. [1]
Approve the full presentation
Have the appropriate person approve the final quotation and the authorized name, role, company and image use. Show the actual placement, including the headline and nearby numerical claims. A truthful sentence can acquire a misleading meaning when paired with a result that came from another project. Ask a legal reviewer about applicable advertising rules where the intended use requires it.
Maintain the evidence after launch
Store the approved version with its source and usage scope. Set a review trigger for major product changes, expired permission or a customer correction. If a statement cannot be verified, remove or replace it with supported product information. This gives the page credible proof rather than a persuasive paragraph that the team cannot explain when a reader asks where it came from.
Sources and evidence
Sources checked on 4 October 2026. Proposed workflows and hypothetical examples are editorial analysis.
From insight to practice