Why this is hard to get right
The Real Problem With Onboarding Training
Maria is a Customer Success Manager at a mid-sized B2B SaaS company. Her team just doubled in size after a Series B raise, and she's responsible for getting six new hires productive within 30 days. The existing onboarding doc is a 47-page PDF nobody reads past page three.
She knows the solution is structured microlearning — short lessons, knowledge checks, real examples. She's used other AI tools before. She typed: "Create an onboarding module for new Customer Success hires." The AI gave her a five-paragraph essay. No objectives. No lesson structure. No quiz questions. Just a wall of text she'd have to reformat for hours.
The core problem isn't the AI. It's the prompt.
Without constraints, the AI defaults to the most generic interpretation of "onboarding module." It doesn't know she needs mobile-first delivery. It doesn't know her lessons cap at four minutes. It doesn't know her team's biggest failure point is miscommunication during escalations. It doesn't know she needs multiple-choice questions with correct answers already flagged.
Maria tried adding a little more detail. She wrote: "Make it for SaaS companies and include quiz questions." The output improved slightly — but the quizzes had five options per question, the tone was stiff, and the examples referenced enterprise software her team doesn't use.
The prompt still wasn't doing the job she needed done.
The breakthrough came when she restructured the entire request. She specified the role (instructional designer), the audience (new Customer Success hires), the format (5-lesson module, 4 minutes each, mobile-first), the exact topics (security, data handling, escalation rules), and the output structure (title, key points, 120-160 word script, knowledge check).
The AI's output changed completely. Lessons were the right length. Examples referenced real SaaS workflows. The knowledge checks were targeted, not generic. She had a draft worth building in under 20 minutes.
What made the difference was structure and specificity. A well-crafted prompt acts like a creative brief. It tells the AI what role to inhabit, who the audience is, how long each piece should be, what format to use, and what success looks like. Every detail you leave out becomes a gap the AI fills with assumptions — and those assumptions are almost always wrong for your specific team.
This is why the prompt template matters as much as the tool you're using to generate it.
Common mistakes to avoid
Targeting 'Employees' Instead of a Specific Role
When you write 'new employees' instead of 'new Customer Success hires' or 'incoming SDRs,' the AI generates examples that fit no one in particular. Role-specific prompts produce role-specific examples. The escalation rules for a CSM are completely different from those for an engineer. Vague audiences force the AI to pick the middle ground, which is useful to no one.
Skipping Lesson Length and Delivery Format
Leaving out constraints like '4 minutes per lesson' or 'mobile-first' means the AI will write however long it feels is appropriate — often 800+ words per lesson. That's incompatible with microlearning. Set hard limits in the prompt. State the word count, the delivery medium, and the time budget. The AI will stay within those boundaries if you define them.
Asking for 'a Module' Without Defining Output Structure
Requesting 'a training module' without specifying what the output should contain — titles, scripts, key points, knowledge checks — produces a prose summary instead of a ready-to-build artifact. Define every component you need. The After Prompt on this page explicitly requests title, key points, a 120-160 word script, one real-work example, and three multiple-choice questions per lesson.
Forgetting to Specify Knowledge Check Format
Asking for 'a quiz' without stating question type, count, and whether you want correct answers flagged results in unusable output. The AI might write open-ended essay questions, 10-question tests, or questions without answer keys. Specify multiple choice, three questions per lesson, with correct answers marked. That's what turns AI output into a deployable assessment.
Not Naming the Tools and Platforms the Team Uses
Generic onboarding content references generic tools — Slack, Excel, generic CRMs. If your team uses Salesforce, Notion, and a specific ticketing system, name them in the prompt. Otherwise, new hires get trained on workflows that don't match their actual day one experience, which creates confusion instead of reducing it.
Omitting the Pass Standard for Assessments
If you don't tell the AI what score constitutes a pass, it can't calibrate difficulty. A module designed for 100% accuracy writes very different questions than one designed for 80%. Set your pass threshold explicitly — for example, 'questions should be passable at 80% correct by someone who read the lesson once.' This produces questions that are challenging but fair.
The transformation
Create an onboarding training module for new employees about our company policies.
You’re an instructional designer. Create a **5-lesson microlearning module** for **new Customer Success hires** at a **B2B SaaS** company. 1. Define **3 measurable objectives** for the module. 2. For each lesson, provide: **title**, 2–3 key points, a **120–160 word script**, and one real-work example. 3. Add **1 knowledge check** per lesson (3 questions, multiple choice) with the correct answers. **Tone:** friendly, direct, and professional. **Constraints:** mobile-first, 4 minutes per lesson. **Topics:** security basics, customer data handling, meeting notes, escalation rules, and acceptable tools.
Why this works
Role Assignment Anchors Tone
The After Prompt opens with 'You're an instructional designer.' This single sentence shifts the AI out of general assistant mode and into domain expert mode. An instructional designer writes objectives before content, sequences lessons logically, and builds checks that test comprehension — not just recall. Without this role assignment, the AI produces content that reads like a summary, not a lesson plan.
Numbered Output Structure Prevents Prose Dumps
The After Prompt uses a numbered list of three deliverables — objectives, per-lesson components, and knowledge checks. This forces the AI to treat each element as a discrete output rather than weaving everything into a paragraph. The result is a module you can paste directly into your LMS or training template without reformatting.
Word Count Constraint Enforces Microlearning Format
The instruction '120-160 word script' is a hard boundary the AI respects. It prevents lesson scripts from ballooning to 500 words, which would break the 4-minute delivery constraint. Short word counts also force the AI to prioritize the most important points — exactly what microlearning requires.
Explicit Topic List Removes Guesswork
The After Prompt names five specific topics: security basics, customer data handling, meeting notes, escalation rules, and acceptable tools. Without this list, the AI picks topics it considers generally important for onboarding. With the list, each lesson maps to a real gap in your team's first-30-days experience.
Tone and Constraints Block in a Single Line
The 'Tone: friendly, direct, and professional. Constraints: mobile-first, 4 minutes per lesson' block packs four requirements into one compact instruction. This pattern — grouping related constraints together — keeps the prompt scannable while ensuring the AI doesn't miss any parameter. It also makes the prompt easy to edit when requirements change.
The framework behind the prompt
The Learning Science Behind Microlearning Prompts
Microlearning is grounded in cognitive load theory, developed by John Sweller in the 1980s. The core principle: working memory can process only a limited amount of new information at once. Lessons that exceed roughly 5-7 minutes of focused content begin to overload working memory, which reduces retention and transfer.
The 4-minute per lesson constraint in the After Prompt isn't arbitrary. It maps directly to cognitive load research showing that short, focused learning bursts with immediate retrieval practice (knowledge checks) outperform long-form instruction by a measurable margin. A 2015 meta-analysis by Hattie and Donoghue confirmed that retrieval practice — what we call knowledge checks — is among the highest-impact learning interventions available, with an effect size nearly double that of re-reading or summarizing.
The Bloom's Taxonomy framework also matters here. When the After Prompt asks for 'measurable objectives,' it's implicitly asking for objectives written at the right cognitive level. Objectives like 'list the five escalation rules' (recall level) are easier to check but have low transfer value. Objectives like 'identify when a CSM should escalate versus resolve independently' (application level) require the learner to actually think — and they produce better on-the-job behavior.
The ADDIE model (Analysis, Design, Development, Implementation, Evaluation) is the foundational framework for instructional design. A well-structured AI prompt collapses the Analysis and Design phases into a single step. By specifying your audience, objectives, constraints, and content scope in the prompt, you're performing ADDIE analysis before the AI writes a single word.
Spaced repetition research (Ebbinghaus, 1885; updated by Cepeda et al., 2006) shows that spreading learning across multiple short sessions dramatically outperforms a single long session. A 5-lesson module structure — even when completed in one sitting — mimics this spacing effect by forcing discrete encoding events between each lesson.
Understanding these principles helps you write better constraints into your prompts. You're not just formatting requests — you're applying evidence-based learning design through the structure of your instructions.
Prompt variations
You're a sales enablement specialist. Create a 4-lesson microlearning module for new SDRs at a B2B software company entering their first two weeks.
- Define 3 measurable objectives for the full module.
- For each lesson, provide: a title, 3 key points, a 130-150 word script, and one real prospecting scenario.
- Add 1 knowledge check per lesson (3 multiple-choice questions) with correct answers marked.
Tone: energetic, direct, and practical. Constraints: desktop and mobile delivery, 5 minutes per lesson. Topics: ideal customer profile, cold call structure, objection handling basics, and CRM data entry rules.
Design questions at a difficulty that a focused new hire passes at 80% on first attempt.
You're a technical instructional designer. Create a 3-lesson microlearning module for new backend engineers joining a remote-first fintech startup.
- Write 2 measurable learning objectives per lesson.
- For each lesson, provide: a title, 3 key points, a 150-180 word script, and one code-adjacent real-world example.
- Add 1 knowledge check per lesson (3 multiple-choice questions) with correct answers and a one-sentence explanation for each answer.
Tone: precise, collegial, and concise. Constraints: text-based delivery, 6 minutes per lesson, no video assumed. Topics: security access protocols, code review standards, and incident escalation procedures.
Focus examples on the first sprint, not abstract concepts.
You're an HR instructional designer. Create a 5-lesson compliance microlearning module for all new hires across departments at a 200-person manufacturing company.
- Define 3 compliance-focused learning objectives for the module.
- For each lesson, provide: a title, 2 key policy points, a 100-130 word plain-language script, and one realistic scenario showing the policy in action.
- Add 1 knowledge check per lesson (3 multiple-choice questions) with correct answers.
Tone: clear, neutral, and non-legalistic — translate policy into plain language. Constraints: mobile-first, 3 minutes per lesson, zero assumed technical knowledge. Topics: workplace safety reporting, harassment and discrimination policy, data privacy, attendance rules, and expense reimbursement.
Write for an 8th-grade reading level. Every script must reference a consequence of non-compliance.
You're a product educator. Create a 3-lesson microlearning module for existing Customer Success team members learning a newly released analytics dashboard in a SaaS platform.
- Define 2 behavioral objectives per lesson — what the learner will do differently after each lesson.
- For each lesson, provide: a title, 3 feature-specific key points, a 120-150 word walkthrough script, and one before-and-after workflow example.
- Add 1 knowledge check per lesson (3 multiple-choice questions) with correct answers.
Tone: encouraging, efficient, and practical — assume the learner is busy. Constraints: desktop-first, 4 minutes per lesson, assume learners already know the old workflow. Topics: navigating the new dashboard, running custom reports, and sharing report exports with clients.
Every example must reference a real CSM task, not a hypothetical.
When to use this prompt
Customer Success Leaders
Standardize onboarding for new CSMs across regions while keeping examples tied to real customer workflows.
Product Managers
Train internal teams on a new feature rollout with short lessons and quick checks for understanding.
Sales Enablement Teams
Build role-specific onboarding for SDRs and AEs with consistent scripts and scenario-based examples.
Engineering Managers
Create lightweight onboarding on security and tooling rules without writing every module from scratch.
HR and People Ops
Ship policy training that stays consistent across departments and supports faster ramp time.
Pro tips
- 1
Specify the job role and first-30-days tasks so the examples match real work.
- 2
Name the top 5 mistakes new hires make so the module prevents them early.
- 3
Set a pass standard, like 80% correct, so the AI designs checks that match your bar.
- 4
List the tools your team uses so the content avoids wrong platforms and steps.
Once you've built a single microlearning module with this prompt, you can chain modules into a full onboarding curriculum by adding a sequencing instruction to the top of your prompt.
Try this addition: 'Before generating the module, output a 3-row table showing: Module Name, Learning Objective, and Prerequisite Module. Then generate Module 1 in full.'
This forces the AI to map dependencies before writing content — a standard instructional design practice that prevents lessons from assuming knowledge the learner hasn't received yet.
Other advanced techniques:
-
Branching scenarios: Add 'For the real-work example in each lesson, write a 2-option branching scenario — one correct path and one common mistake path, each with a 2-sentence consequence.' This turns passive scripts into decision-based learning without needing an authoring tool upfront.
-
Spaced repetition hooks: Add 'At the end of each lesson, include one reflection question the learner should answer in their own words — no right answer required.' This supports retention without adding formal assessment overhead.
-
Facilitator notes: Add 'After each lesson script, add a 3-bullet facilitator note for managers covering: what to reinforce, a common misconception to address, and one question to ask in a 1:1.' This extends the module's value beyond self-paced delivery.
The core prompt structure works across industries, but three variables shift depending on your sector: compliance weight, technical depth, and scenario realism.
Healthcare and regulated industries: Compliance language must be precise. Add: 'Every lesson must cite the specific regulation or policy it supports (e.g., HIPAA Section 164.308). Avoid paraphrasing regulatory language — quote it directly, then explain it in plain terms.' Knowledge checks in regulated training often require a higher pass threshold — set it at 90% or 100% for safety-critical content.
Financial services: Scenarios need to reflect real customer interaction risks. Add: 'Each scenario must involve a customer interaction where the wrong action creates measurable financial or reputational risk.' This grounds abstract compliance rules in consequences the learner actually cares about.
Retail and frontline teams: Simplify the script length and reading level. Use: '80-100 word scripts, 6th-grade reading level, no industry jargon.' Frontline workers need content they can absorb between shifts, not during dedicated study time.
Professional services and consulting: Learners are often high-autonomy and resistant to prescriptive training. Adjust tone to: 'Write as a peer sharing hard-won experience, not as an authority issuing instructions.' Frame knowledge checks as 'judgment calls' rather than right/wrong assessments.
Run through this checklist before sending your prompt to any AI model. Each item corresponds to a failure mode that produces unusable output.
Audience and Role
- [ ] Named the specific job title, not just 'new employees'
- [ ] Specified the company type or industry
- [ ] Noted any prior knowledge the learner already has
Format and Constraints
- [ ] Set a lesson count (3-5 recommended)
- [ ] Defined maximum minutes or words per lesson
- [ ] Specified delivery format (mobile-first, desktop, in-person)
Content Scope
- [ ] Listed the exact topics to cover (5 or fewer per module)
- [ ] Named any tools, platforms, or processes by their actual names
- [ ] Flagged any topics to exclude or handle carefully
Output Structure
- [ ] Requested each component explicitly (title, objectives, script, example, quiz)
- [ ] Specified knowledge check format (multiple choice, question count, answer key)
- [ ] Set a pass threshold if assessments will be graded
Tone
- [ ] Included at least two tone descriptors
- [ ] Added a behavior-based tone example if precision matters
If you check all 15 boxes, your prompt is ready to generate a first-draft module worth building.
When not to use this prompt
When Not to Use This Prompt Pattern
This prompt works best for structured, repeatable knowledge transfer — the kind of content where there's a right answer and a clear behavior change you're targeting. It's not the right tool for every training need.
Avoid this pattern when:
- You're building complex skill simulations. Microlearning scripts can't replace hands-on practice for technical skills like software demos, surgical procedures, or high-stakes client conversations. Use the AI to generate briefing content, but build the simulation separately.
- The content is legally sensitive or jurisdiction-specific. AI-generated compliance training requires legal review. Don't ship it directly — use it as a first draft that a compliance officer validates line by line.
- Your audience needs dialogue and discussion. Microlearning is a one-way format. For topics that require debate, perspective-sharing, or emotional processing (like DEI training or leadership development), a facilitator-led session produces better outcomes than a self-paced module.
- The topic changes faster than your update cycle. If your product roadmap shifts monthly, microlearning modules become outdated quickly. Living documents or wiki-style resources are better formats for fast-changing content.
When in doubt, ask: does this topic have a correct answer that a new hire can learn and demonstrate? If yes, microlearning fits. If the topic requires judgment developed over time, use a different format.
Troubleshooting
The AI produces a long essay instead of structured lesson components
Your prompt is missing an explicit output structure. Add a numbered list of exactly what you want inside each lesson: title, key points, word-counted script, real-work example, and knowledge check. If needed, add: 'Format each lesson as a clearly labeled section. Do not combine components into paragraphs.' Structure in the prompt produces structure in the output.
Knowledge check questions are too easy or too generic
Add a calibration instruction: 'Write questions that require the learner to apply the concept, not just recall a definition. Each question should have one clearly correct answer and two plausible-but-wrong distractors.' Also specify: 'Base questions on the real-work example in each lesson, not the key points alone.' Scenario-anchored questions are harder to guess and test real understanding.
Lesson scripts run too long and exceed the time budget
Add a word count enforcement line: 'Each script must be between 120 and 160 words. After writing each script, count the words and display the count in parentheses.' Asking the AI to self-report the word count creates accountability and catches overruns before you have to re-prompt. If scripts still run long, reduce your key points from three to two per lesson.
Examples reference tools or workflows your team doesn't use
Name your actual tools in the prompt. Replace generic references by adding: 'All examples must reference these specific tools: [your CRM], [your project management tool], [your communication platform]. Do not substitute or reference other tools.' Specificity in the tool list eliminates generic substitutions that confuse new hires on day one.
The tone feels formal and academic instead of friendly and direct
Move the tone instruction to the very first line of your prompt, before the role assignment. Then replace adjective-only descriptors with a behavior-based instruction: 'Write the way a knowledgeable colleague explains something to a new teammate on their first week — helpful, conversational, and never condescending.' Behavioral tone models produce more consistent results than lists of adjectives.
How to measure success
How to Evaluate the AI Output
A strong first-draft microlearning module passes these checks before you move it into an authoring tool.
Structure checks:
- Every lesson has all five components: title, key points, script, real-work example, and knowledge check
- Scripts fall within your word count range — count them manually if the AI didn't self-report
- Objectives are measurable — they start with a verb and name a specific behavior (not 'understand' or 'appreciate')
Content quality checks:
- Examples reference your actual role, tools, and context — not generic placeholders
- Knowledge check questions test application, not just recall — the correct answer requires thinking, not skimming
- Distractors are plausible — wrong answers should reflect common misconceptions, not obvious errors
Tone checks:
- Read one script aloud — if it sounds like a legal document, the tone instruction didn't land
- Check for jargon your new hire wouldn't know on day one
Completeness check:
- All five topics from your list appear in the module — confirm none were dropped or merged without reason
If more than two checks fail, revise the prompt and regenerate rather than editing the output manually.
Now try it on something of your own
Reading about the framework is one thing. Watching it sharpen your own prompt is another — takes 90 seconds, no signup.
Build a role-specific onboarding module with knowledge checks — ready to load into your LMS.
Try one of these
Frequently asked questions
3 to 5 lessons is the standard range for onboarding microlearning. Fewer than 3 lessons rarely covers a meaningful skill cluster. More than 5 in a single module often signals scope creep — consider splitting into two modules. The After Prompt on this page uses 5 lessons at 4 minutes each, totaling about 20 minutes of structured learning. Match lesson count to your topic list, not the other way around.
Yes — and you should. Replace 'B2B SaaS' with your industry, swap 'Customer Success hires' with your specific role, and update the topic list to reflect your team's actual first-30-days tasks. The structure of the prompt stays the same. The role, audience, industry, and topics are the four variables you swap. Everything else — the output format, word count constraints, and tone block — transfers directly.
This usually means your word count constraint isn't explicit enough. Add a hard boundary like 'each script must be between 120 and 160 words — do not exceed 160 words.' If the AI still goes long, add: 'Count the words in each script and confirm the count at the end of each lesson.' That accountability instruction forces compliance in most cases.
Specify both the format and the difficulty standard. State: '3 multiple-choice questions per lesson, 4 answer options each, with one clearly correct answer marked and a one-line explanation of why it's correct.' Also include a calibration instruction like: 'Questions should be passable at 80% by someone who read the lesson once but challenge anyone who skimmed it.' This produces assessment-grade questions, not trivia.
Yes, whenever the content will reference real workflows. Naming your actual tools (Salesforce, Notion, Jira) prevents the AI from substituting generic alternatives. You don't need to share confidential policies — just the tool names and any process names your team uses. The more specific you are, the more the output matches day-one reality for your new hires.
Not directly. This prompt generates the content layer — scripts, objectives, and knowledge checks — which you then import into your authoring tool (Articulate, iSpring, or similar). Think of the AI output as a structured first draft. You'll still need your LMS authoring tool to add interactions, branching, and SCORM packaging. The prompt eliminates the blank-page problem, not the build step.
Include a brief description of the topic in the prompt rather than a topic name alone. For example, instead of 'escalation rules,' write: 'escalation rules — when a CSM should loop in their manager vs. resolve independently, with a 2-hour response standard.' That level of context lets the AI write accurate content even without access to your internal documentation.
Move the tone instruction to the top of the prompt, before the numbered list. Tone set early acts as a filter for everything that follows. Also try adding a concrete example: 'Write the way a senior colleague would explain this to someone on their first week — helpful, not academic.' Behavior-based tone descriptors outperform adjective-only instructions like 'friendly and professional.'