Generative AI in L&D — What It Can Build, What It Cannot Judge, and Where the Line Sits
83% of instructional designers now use generative AI. Research also shows that 78% of AI-generated lesson plans require significant adjustments before they are fit for use. Both statistics are true at the same time, and the gap between them is where quality is either protected or quietly lost. The distinction is not whether to use…
1. The Confidence Problem: Why GenAI Is Most Dangerous When It Is Most Useful
Generative AI is the most capable content production tool L&D has ever had access to. It is also the most confidently wrong tool L&D has ever had access to. The combination is what makes it genuinely risky in ways that previous productivity tools were not.
A spell-checker flags uncertainty. A search result offers a source to evaluate. GenAI produces fluent, well-structured, grammatically correct content with no indication of where it is confident and where it is confabulating. In a production workflow under time pressure, a well-formed AI output gets approved faster than a rough draft that visibly needs work. The quality problem becomes invisible until a learner, a subject matter expert, or a regulator finds it.
Key Distinction
GenAI is excellent at the tasks that feel like instructional design but are actually production tasks. It is unreliable at the tasks that are actually instructional design: determining whether a learning objective produces the right cognitive outcome, judging whether a scenario has the ambiguity that drives genuine learning, and writing feedback that changes behaviour rather than simply confirming or correcting. These require human expertise — not because the AI lacks capability but because it lacks the pedagogical judgement to know when its output is pedagogically wrong.
83%
of instructional designers now use GenAI, widespread adoption with uneven governance
65%
time savings in lesson planning from GenAI, the production efficiency gains are real and significant
78%
of AI-generated lesson plans required significant adjustments to align with learner context and standards
58%
better results from GenAI tools for designers who invest in prompt engineering, the skill gap is significant
2. What GenAI Builds Well: The Genuine Time Savings
The efficiency gains from GenAI in L&D production are real and should be captured. Research consistently shows significant time savings on the production tasks that are structurally predictable: outline generation, quiz question drafting, script first drafts, voiceover production, translation, and accessibility description generation.
GenAI excels · Automate with review
Course outline and structure generation from source documents
Given approved source material, GenAI can generate a structured course outline rapidly. The output requires instructional review — the sequencing and emphasis will not automatically reflect the learning objective hierarchy — but a first draft produced in minutes is significantly faster to review and reshape than one produced from scratch. This is a genuine production accelerator when the review process is built in.
GenAI excels · Automate with review
Assessment question drafting at volume
GenAI can generate large question banks rapidly, and research shows up to 95% time savings in assessment generation. The output consistently skews toward recognition questions — multiple choice with plausible distractors — rather than application or analysis questions. A subject matter expert review that reclassifies and replaces the weakest items is still significantly faster than generating the full bank from scratch.
GenAI excels · Automate with review
Voiceover production and multilingual translation
Text-to-speech quality has reached a level appropriate for most corporate eLearning contexts. AI translation combined with human localisation review produces multilingual content at a fraction of traditional production cost. These are tasks where the GenAI output quality is consistently high enough that the production time saving is captured without a proportionate quality cost — provided the localisation review step is not eliminated.
3. What GenAI Cannot Judge: Where Human Expertise Is Non-Negotiable
Research on AI in instructional design consistently identifies the same limitation: GenAI-generated content frequently lacks deep understanding of learning science principles and struggles with contextual adaptation. The content looks correct. The pedagogy is weak.
Human judgement required · Always
Learning objective specification at the right cognitive level
GenAI generates learning objectives that are grammatically correct, measurable, and formatted to a standard template. They are frequently at the wrong level of Bloom’s taxonomy for the intended outcome — producing knowledge objectives for programmes designed to change practice, or analysis objectives for content intended to develop foundational awareness. A learning objective at the wrong level produces a misaligned assessment, which produces a misaligned module. The error is invisible in the output and consequential in the learning design.
Human judgement required · Always
Scenario design with the right ambiguity
GenAI generates scenarios that are structurally plausible and resolve cleanly. The clean resolution is the problem. Scenarios that produce genuine learning require the ambiguity that makes the learner think, a protagonist who is not clearly right or wrong, competing values that cannot both be satisfied, a situation where the learner’s instinct and the correct response diverge. GenAI’s training produces cooperative scenarios. Good instructional designers produce uncomfortable ones. The difference is what determines whether the learner has actually practised the decision.
Human judgement required · Always
Feedback that explains rather than scores
GenAI feedback confirms correct answers and identifies incorrect ones. Effective feedback explains the specific reasoning that would have produced a better response; why this choice leads to that outcome, what the learner’s response reveals about their mental model, and what they should think differently next time. This requires knowing why a particular wrong answer is wrong in this specific context, which requires knowing the learner’s likely misconception, which requires instructional expertise that GenAI does not carry.
Human judgement required · Always
Determining whether training is the right intervention
GenAI will produce a training programme for any brief it is given. It cannot determine whether the performance problem being addressed is a knowledge problem, a skill problem, an environment problem, or a motivation problem, and therefore whether training is the right intervention at all. The most expensive mistake GenAI can produce is a well-crafted module that solves the wrong problem confidently.
“GenAI is most dangerous when it produces content that looks like the output of an expert. An instructional designer reviewing AI-generated learning objectives, scenarios, and feedback must be evaluating them against pedagogical standards — not against the standards of grammar, structure, and fluency, which AI consistently meets and instructional quality does not consistently require.”
4. The Governance Question: Who Reviews What, and When
The governance failure in most GenAI L&D deployments is not technical — it is structural. Teams adopt GenAI tools for efficiency, reduce review time to capture the efficiency gain, and discover the quality problem when a learner, a manager, or a subject matter expert raises it.
Effective GenAI governance in L&D requires a defined framework: which production tasks GenAI can complete without human review, which require instructional quality review before use, and which require subject matter expert review in addition to instructional review. The framework should also address information security — many GenAI tools use input data in model retraining, which creates intellectual property and confidentiality risk when proprietary organisational content is used in prompts.
The governance framework does not slow production. It protects the efficiency gain from being eroded by the rework that unreviewed AI output consistently generates when it reaches learners. Research consistently shows that truly customised, contextually appropriate instructional content still requires human oversight — the question is where in the workflow that oversight is most efficiently applied.
Qquench AI Automation Practice · GenAI Governance · 25+ Years
Before adopting GenAI tools for eLearning content production, Qquench helps L&D teams define the governance framework that determines which tasks the AI owns, which require instructional review, and which require SME review — so the efficiency gains are captured without the quality losses that unreviewed AI output consistently produces.
The governance decision comes before the tool adoption. Most teams discover they need it after the first quality incident.
5. The Qquench Approach: Define the Line Before You Deploy the Tool
Qquench’s GenAI integration framework maps the L&D workflow task by task — identifying which steps are production tasks that GenAI can accelerate, which are design decisions that require instructional expertise regardless of AI capability, and which require both AI production and human review in combination.
The framework produces a defined workflow: AI generates the first draft, the instructional designer reviews against pedagogical criteria rather than editing for grammar and fluency, and the subject matter expert validates factual accuracy. The review is calibrated to what the AI gets wrong rather than what it gets right — which means reviewers are checking objective levels, scenario ambiguity, and feedback quality, not sentence structure.
A global pharmaceutical company adopted a GenAI authoring tool to accelerate production of regulatory training content across 40+ modules. Initial quality reviews caught consistent patterns: learning objectives written at knowledge level for modules designed to change pharmacovigilance reporting behaviour, scenarios that resolved before the learner reached the decision point that drove the compliance failure, and feedback that confirmed regulatory requirements without explaining the consequence of the behaviour being discouraged. The production speed had improved significantly. The instructional quality had not — because the review process had been scoped to factual accuracy rather than pedagogical judgement. A review framework redesigned around the specific failure patterns GenAI produces for this content type reduced the rework cycle from three rounds to one, without slowing the production speed that had prompted the tool adoption.
Explore on Qquench
In Summary
GenAI produces 65% time savings in lesson planning and requires significant adjustments to 78% of its lesson plans. Both are true. The line between them is where human instructional judgement is the non-negotiable input: learning objective specification at the right cognitive level, scenario design with the ambiguity that drives genuine learning, feedback that explains rather than scores, and the foundational determination of whether training is the right intervention at all. GenAI accelerates the production tasks reliably. It cannot perform the design judgements that determine whether the production was worth doing. Governance frameworks that define this line before tools are adopted capture the efficiency gain without the quality loss. Frameworks built after the first quality incident cost more than the adoption did.
Qquench · 25+ Years · Fortune 100 · Global
Find out whether your GenAI workflow has defined where human instructional judgement is required or whether the efficiency gain is being captured at the cost of pedagogical quality that your learners will notice before you do.
Qquench’s GenAI governance audit maps your current workflow against the tasks GenAI gets reliably right and those it gets confidently wrong, and produces a review framework that protects quality without slowing production.
Frequently Asked Questions
Q1
What is generative AI best at in the eLearning content production workflow?
Generative AI excels at the production tasks that are structurally predictable and volume-intensive: generating course outlines, drafting quiz questions, producing first-draft scripts, converting text to audio, translating content, and creating accessibility descriptions. Research shows 65% time savings in lesson planning and up to 95% in assessment question generation — the gains are real for the right task types.
Q2
Where does generative AI produce plausible but wrong outputs in L&D?
The most dangerous GenAI failure in L&D is confident production of content that is pedagogically weak: learning objectives at the wrong Bloom’s level, scenarios that resolve too cleanly to drive real learning, assessment questions that test recognition rather than application, and feedback that confirms without explaining. Research found 78% of AI-generated lesson plans required significant adjustments — not because content was factually wrong, but because it lacked pedagogical judgement.
Q3
Do instructional designers need to understand AI to work effectively with GenAI tools?
Yes, AI literacy and prompt engineering are now core instructional design competencies. Research shows designers who invest in prompt engineering achieve 58% better results from GenAI tools. The ability to critically evaluate AI-generated content against pedagogical standards is as important as the ability to produce original content.
Q4
What governance does an L&D team need before deploying GenAI in content production?
At minimum: a defined list of which production tasks GenAI can complete without review, which require instructional quality review before use, and which require SME review as well. Proprietary information protection is also critical — many GenAI tools use input data in model retraining, creating IP and confidentiality risk when organisational content is used in prompts without governance controls.
Q5
How does GenAI change what instructional designers spend their time on?
GenAI removes the time cost of production tasks that were never genuinely design work — first-draft scripting, outline generation, accessibility descriptions, translation, quiz question generation. Instructional designers who use GenAI effectively spend more time on objective specification, scenario design, feedback quality, and measurement frameworks. The design work becomes more valuable because production is faster.
Q6
Has Qquench integrated GenAI into its eLearning production workflow?
Yes, with 25+ years and 1,256+ hours of eLearning delivered for Fortune 100 clients globally, Qquench uses GenAI for production tasks that do not require instructional design judgement, with defined human review gates at every point where pedagogical quality depends on expert assessment. The governance framework is built before tools are adopted — not retrospectively after quality problems emerge.
QS
Qquench Specialists
AI Automation and Learning Design · Qquench
Qquench Specialists is the collective voice of Qquench’s learning design and AI practice. With 25+ years delivering award-winning eLearning for Fortune 100 clients globally, we write from practice, not position papers.