Is the AI purpose-built for instructional design?
A purpose-built AI authoring tool produces different output than a general chatbot with a course-shaped prompt. Ask: "What is your AI, exactly — one model with a prompt, or a system that checks its own work?"
The vendor should describe an architecture, not a prompt library. Edplay runs a compound AI system that orchestrates 34 agents: specialist agents generate the content, and separate validator agents judge it against the approved plan. Pass or fail is computed by the system — never taken from a model's self-report. Follow up with: "Who built the instructional structure in?" and "Show me two courses on the same topic, built minutes apart." A general tool produces a strong first course and a second that diverges completely. A purpose-built system holds the same framework every time, because the framework lives above the model, not inside a prompt.
Josh Bersin, global industry analyst and founder of The Josh Bersin Company, puts the stakes plainly: "This thing we call training is actually very sophisticated and very complex. You've got to teach them about it, so they know why they're doing what they're doing." A tool that only generates content skips the "why." A tool trained on instructional design keeps it.
Data Breakdown:
| Dimension | Old Way (Articulate) | Bolt-On Way (ChatGPT) | Edplay Way |
|---|---|---|---|
| Instructional structure | Built by hand, page by page | Prose with no learning framework | AI trained on instructional design frameworks and learning science so every structure carries actual pedagogy |
| Fact validation | Manual SME check | The model self-reports it's correct | Every statement source-traced and judged verified, fixed, or rejected by the AI compound system of 34 agents |
| Consistency across courses | Depends on the builder's skill | First output strong, the second diverges | Governed by learning settings set up by the instructional designer, so every course follows the same standards |
| Content source | Whatever you build in | May pull from the public web | Generated only from your SOPs (standard operating procedures); contradictions caught at intake |
Architecture is not a detail — it's the whole result. Ask vendors: "Is this one model with a prompt, or a system of models that check each other?" A single-chat tool can't answer that, because there's nothing underneath the prompt to check.
Does it build from your learning strategy, or generate around it?
A design-first tool starts from your objectives and audience — it doesn't ask you to force-fit strategy onto content it already generated. Ask: "Does your tool start from my learning objectives and audience, or does it generate content and leave me to shape it afterward?"
The vendor should show a structured setup that captures your outcome before anything generates. In Edplay, a build runs through roughly ten setup questions and produces a storyboard you review first — the ID-native stage where you see the plan before a single page is generated. Follow up with: "What do I have to define before it builds?" and "Can I reach a usable draft without polished documentation?" You can get to roughly 80% of a finished course from a verbal conversation with a subject matter expert (SME), no knowledge base required.
A course builder at Region 7, an Edplay customer, puts the payoff in concrete terms: "I can easily get done in one day what would have taken a minimum of a week's worth of time. The course building side of Edplay satisfies one of the biggest needs we have, and that's time — drastically. I don't even know if we could calculate the amount of hours that were saved. It makes a very good course." That's the design-first result: the designer's strategy leads, and the build stops being the bottleneck. Designers create. Edplay builds. A tool that flips that order hands you content and makes the strategy your problem.
Does it validate content before anything ships?
An AI authoring tool built for enterprise shows its work — nothing gets in by assertion, and nothing ships unverified. Ask: "When the AI states a fact, how do you know it's true — does the system verify it, or does the model just tell you it's fine?"
The vendor should show validation layers, not a single generate button. Every upload to Edplay is analyzed at intake — duplicates, repeated content inside a file, and contradictions with what's already in your knowledge base are caught before anything enters. Every statement in the finished course is linked to its source and passes through multiple validation layers, where it is judged verified, fixed, or rejected. Follow up with: "How long to a reviewable draft, and how long to review it?" Edplay produces a course in about 7 minutes and expects 1 to 2 hours of human review — the review is the point, not a formality. SMEs can build within your framework, but they can't publish to the LMS (learning management system) without your sign-off.
Ask vendors: "Show me the step where a human approves before anything reaches a learner." If everything generates on one click with no validation layer and no sign-off, you're back to prompt-and-hope — fast output, no quality gate, and a rebuild waiting for you.
Will it work with your LMS?
Integration is where most AI tools quietly fail — a course you can't get into your LMS is a demo, not a deployment. Ask: "What exactly do you export, and how does it get into our LMS?"
The vendor should name formats, not wave at compatibility. Edplay exports SCORM 1.2 and SCORM 2004 (Sharable Content Object Reference Model — the packaging standard most LMS platforms use), supports LTI (Learning Tools Interoperability — the cleaner, modern integration that needs no change for learners), and offers embed links and an open API. Follow up with: "Does it work with Workday?" and "When I edit a course, does the change reach learners automatically?" Workday has been tested; courses go in via SCORM or API, and edits — including find-and-replace fixes — flow through to where learners take the course, with no change to your tech stack.
Ask vendors: "Which SCORM version, and what happens on every edit after launch?" A vendor who says "we support SCORM" without naming a version, or who makes you re-upload the package every time content changes, has an integration problem you'll inherit.
What happens with compliance-regulated content?
An AI authoring tool built for compliance traces every line to your source — it doesn't ask you to trust content you can't verify. Ask: "When we build compliance training, can you show me where every statement came from?"
The vendor should prove control over the source, not just the output. Edplay generates only from the material you provide, checks content against compliance documents you upload — OSHA (Occupational Safety and Health Administration) regulations, HSE (Health and Safety Executive) guidelines — and source-traces every statement, so nothing enters by assertion and nothing gets pulled off the public web. The stakes are concrete: a serious OSHA violation costs up to $16,500, and willful or repeated violations reach $165,000 each. Content built on a version of the procedure that no longer exists is exposure, not training. Follow up with: "How do you catch a fact the model invented?" In an independent expert review, Edplay reports zero fabricated facts across five live courses — including four deliberate traps set to catch the system inventing content.
Ask vendors: "What happens to our sensitive procedures — do they reach a third-party model?" No AI-built course carries a formal regulatory stamp, and an honest vendor says so; the approach is demonstrating alignment against the regulatory source if audited, with the final accuracy check held by your SME. Edplay redacts sensitive data before anything reaches a third-party model. If a vendor can't show you where content came from, that's a red flag.
What reporting and data visibility should you expect?
Reporting determines whether you can prove the training worked and whether your data stays yours. Ask: "What can you show me about completion, engagement, and where learners struggle — and is our data ever used to train your AI?"
The vendor should show live insight and a clear data position. Edplay has built-in AI reporting you query in natural language — completion, engagement, quiz performance — and its in-course tutor, Lia, turns every learner question into a labelled signal: what confuses people, where the gaps are, what to teach next. That means your knowledge base learns from your own workforce, rather than sitting static. Follow up with: "Where is our data hosted?" and "Is our content ever used to train your models?" Edplay is ISO-certified, hosts on US, EU, or UK servers, staffs a full-time DPO (data protection officer), and never trains any AI model on your content.
David Wakefield, Founder and CEO of Sibme, ties visibility to the authoring workflow itself: "Edplay has been instrumental in optimizing Sibme's course authoring capabilities. With 123 hubs and 193 external users engaging at over 80%, we've gained content insights, tracked user engagement, and streamlined learning delivery." Enterprises from Cincinnati Children's Hospital to Zebra and Sherwin-Williams run training built this way. Ask: "Is my company's data ever used to train your models?" If a vendor hedges on that question, walk away.
Key Takeaways
• A purpose-built AI authoring tool runs an architecture, not a prompt — Edplay's compound system of 34 specialist and validator agents computes pass or fail against the approved plan, where a general chatbot produces a strong first draft and an inconsistent second one.
• A design-first tool starts from your learning objectives, audience, and a storyboard you approve before generation — so your strategy leads and the system executes, rather than handing you content to force-fit.
• Real validation means nothing enters by assertion and nothing ships unverified: every statement is source-traced and judged verified, fixed, or rejected, with a human sign-off before anything reaches a learner.
• LMS integration must be specific — SCORM 1.2 and 2004, LTI, embed links, an open API, tested Workday support, and edits that flow through to learners automatically — not a vague "we support SCORM."
• Compliance-ready authoring generates only from your source, traces every line, catches invented facts (Edplay reports zero fabricated facts across five audited courses), and redacts sensitive data before it reaches a third-party model.
• Effective reporting shows live completion, engagement, and learner-struggle signals you can query in plain language, backed by a clear position that your content never trains the vendor's AI.