Evaluating an AI training vendor: the 8 questions that separate real from theatrical

AI training is the easiest category to fake. The vendor demo looks impressive. The credentials look real. The price is reasonable. Then delivery happens and the cohort produces nothing.

Here are the 8 questions to ask before signing.

1. Who wrote your curriculum?

Good answer: named individuals at the vendor, with public proof of work (book, articles, conference talks, prior roles).

Bad answer: "we license it" / "our team built it" / vague references to anonymous senior staff.

You're paying for someone's accumulated craft. If you can't name the person, you don't know what you're buying.

2. What's your reference artifact after the cohort?

Good answer: a book or detailed workbook the learner takes home and references months later.

Bad answer: the slide deck. Slide decks die in SharePoint folders.

3. How do you measure behavior change after the cohort?

Good answer: 30-day survey + reused-prompt tracking + cross-departmental influence tracking.

Bad answer: "we send a satisfaction survey" / "we don't measure post-cohort, that's the client's job."

Vendors who don't measure behavior change don't optimize for behavior change. They optimize for the demo.

4. Show me a real cohort outcome from a similar customer.

Good answer: anonymized but specific. "At Customer X (federal civilian agency, 80 staff), Day-30 behavior-change rate was 67%; here's the post-cohort prompt library they built."

Bad answer: generic testimonials. "Our customers love us!"

5. What's your refund / make-good policy?

Good answer: if 30-day behavior change drops below X%, vendor offers a refresher session at no cost OR a partial refund.

Bad answer: no make-good. Or refund tied to attendance, not outcomes.

6. Who delivers the training?

Good answer: named lead facilitator with public bio (LinkedIn, IMDb, book authorship). Their hours are reserved on the SOW. No substitution without your written approval.

Bad answer: "our senior team" / facilitator named after contract signing / facilitator substitution language permissive.

7. Section 508 / accessibility posture?

Good answer: vendor publishes a VPAT 2.5. Captions on all video. Screen-reader-compatible slides. Color contrast verified.

Bad answer: "we comply with accessibility" without documentation.

For federal civilian customers this is a contract requirement, not optional.

8. How do you handle our sensitive data in examples?

Good answer: sanitized examples only. No CUI/PII in training materials. Documented data-handling policy. (Apache-3's is at apache3.com/compliance.)

Bad answer: "we use customer scenarios" without a sanitization process. Or evasive about their data-handling.

What an Apache-3 / LearnTrainAI response looks like

1. Curriculum by Snake Blocker and Naveen Dhillon. Published as *Prompt to Product*, April 2026.
2. The 380-page book is the reference artifact.
3. Day-30 behavior change survey + 60-day check-in + 90-day retention check (in the SOW).
4. Anonymized customer outcomes available under NDA after fit call.
5. Refresher session at no cost if Day-30 behavior change drops below 50%.
6. Named facilitator on every SOW. No substitution without written approval.
7. VPAT 2.5 published. Captions on all video.
8. Documented data-handling. No customer CUI/PII in materials.

See the curriculum → · Reserve a slot →