The questions that actually separate providers now that everyone claims AI capability — and what the evasive answers tell you.
- Process and accountability
- Assets and continuity
- Security and fit
Process and accountability
1. What happens to AI output before it reaches me? 2. Who performs review, and what is their domain background? 3. What QA standard is applied, and can I see a report? 4. Who is accountable when delivered content is wrong?
The fourth question is the most revealing. Providers who have thought about it answer directly; providers who have not change the subject to their technology.
Assets and continuity
5. Do you build a terminology base from my content? 6. Do I own it, and in what format can I export it? 7. How is consistency maintained across revisions? 8. What happens to my assets if we stop working together?
These determine whether you are buying a service each time or building something. A vendor with no good answer here is one you will pay full price to indefinitely.
Security and fit
9. Does your NDA cover individual linguists and reviewers? 10. Is my content used for model training, and where is it retained? 11. Which of my content types would you route to a lower tier, and why? 12. Show me your process applied to a sample of my actual documents.
Question 11 is a good-faith test. A provider willing to tell you where you do not need their most expensive service is one whose recommendations you can weigh. Question 12 replaces the entire discussion with evidence.