How to Use ChatGPT for Instructional Design: Real Workflows

ChatGPT works best for instructional designers as a drafting and critique partner across five core tasks: learning objectives, course outlines, storyboards, scenarios, and analysis work. This guide gives you a working prompt for each task, plus the specific way that ChatGPT tends to fail at it, so you can catch the problems before they reach a learner.

I've used ChatGPT in real instructional design work since early 2023, and I'll tell you plainly where it falls short. Everything here works as of August 2026, written against the GPT-5.5 and GPT-5.6 generation of ChatGPT models.

What can ChatGPT actually do for instructional designers?

ChatGPT reliably speeds up drafting, ideation, and critique across the design phase of your process. It is least reliable wherever factual accuracy, organizational context, or live learner interaction come into play. If you want the wider picture of where AI fits across the whole field, start with my AI in instructional design guide; this page covers ChatGPT specifically, in detail.

My stance on the tool is on the record. When I picked my top three eLearning tools in late 2025, the list was Storyline, ChatGPT, and Freepik (since rebranded to Magnific), and ChatGPT earned its spot by being useful at nearly every stage of the design process. The models have kept improving since then.

AI has gotten much better at instructional design work. It's on par with a professional instructional designer now, especially if you have some prompting skills.

Prompting skill is the difference between output you can build on and output you delete and try again. Here's my read on where ChatGPT fits, task by task, after three years of using it in real projects:

ID task	What ChatGPT does well	Watch out for	Prompt in this guide
Learning objectives	Measurable objectives with conditions and criteria (once you demand them)	Vague verbs (understand, know, learn) on lazy prompts	Objectives prompt
Course outlines and ideation	Fast first-drafts and engagement ideas	Generic sequences that are not the steps you would actually teach	Ideation prompt
Storyboards	Screen-by-screen drafts in your own storyboard format	Bloated voiceover and click-next filler interactions	Storyboard prompt
Scenario writing	Decision points, dialogue, and plausible-wrong distractors	Melodrama and giveaway correct answers	Scenario prompt
Critique of your work	Rubric-based review where you are the accuracy check	Flattery, unless you ask for a demanding reviewer	Critique prompt
Analysis and research	Summarizing intake docs, drafting SME questions, deep research reports	Invented sources in standard chat	Analysis workflow
Conversational practice delivery	Not a fit. It can write the scenario document, but it cannot play a character inside your course and score the conversation	Needs a purpose-built simulation layer	Where the boundary sits

I recorded a full walkthrough of my ChatGPT workflow back in 2023. The model it shows is several generations old now, but the prompting principles in it still hold:

How do you write ChatGPT prompts that produce usable ID work?

Give ChatGPT a role, a task, and context every single time. This 30-to-60-second investment is the highest-return habit in this entire guide.

In early 2025 I ran a side-by-side test. First, a bare request: "write an outline for a five minute eLearning course about artificial intelligence for instructional designers." The outline was fine. Generic, but fine. Then I rebuilt the same prompt with a role, a task, and context. It took me an extra 30 to 60 seconds to write, and the output was dramatically better: sharper objectives, a stronger outline, and choices that reflected the audience instead of a template. FIU's Center for the Advancement of Teaching teaches the same pattern to faculty: use fully worked prompts that spell out the role, audience, and output format.

Here's the framework every prompt in this article follows:

1. Assign a role

   Tell it who to be: "You are a senior instructional designer at a healthcare company who designs scenario-based eLearning for frontline nurses." The role sets the vocabulary, the assumptions, and the quality bar.

2. State the task precisely

   Name the deliverable, the length, and the format. "Draft three terminal objectives" beats "help me with objectives."

3. Give it context

   Paste in the business goal, the audience details, the constraints, and any source material. ChatGPT knows nothing about your organization unless you tell it.

4. Specify the output format, then invite questions

   Ask for a table, a bulleted outline, or your exact storyboard columns. End with "Ask me up to 5 clarifying questions before you start." The questions it asks will often expose gaps in your own thinking.

The master template, ready to lift:

You are a senior instructional designer at [company type] who designs
[modality] for [audience].

Task: [the deliverable you need, including length and scope].

Context: [business goal, audience details, constraints, source material].

Output format: [table, bulleted outline, screen-by-screen storyboard, etc.].

Ask me up to 5 clarifying questions before you start.

How do you use ChatGPT to write learning objectives?

Never accept ChatGPT's first draft of a learning objective. Left unconstrained, it reaches for exactly the vague verbs that ID 101 teaches you to avoid: understand, know, learn, be aware of.

This was glaring back in 2023 when I first tested it. The default output leaned on "understand this" and "know that" in almost every draft, and Luke Hobson, who was publishing his own ChatGPT experiments around the same time, hit similar walls. The models are far better now. As I said above, current ChatGPT is on par with a professional instructional designer when you prompt it well. But the operative phrase is still "when you prompt it well": your prompt has to force measurable verbs, conditions, and criteria.

And it delivers when you do. In a live workshop last summer, I had ChatGPT draft a terminal performance objective for training instructional designers themselves. It produced something close to this: "given a project kickoff document or intake form describing a stakeholder's training request, the instructional designer will identify the underlying business goal the training is intended to support, with at least 90% accuracy as judged by a senior ID or training manager." It includes the condition, observable performance, and criterion. We then spent the session stress-testing whether the objective actually held up, which is exactly the posture to take with any AI draft.

The prompt that forces this structure:

You are a senior instructional designer who writes measurable learning
objectives. Write [N] terminal objectives for [course topic] aimed at
[audience].

Requirements:
- Each objective includes a condition ("Given a..."), an observable
  performance verb aligned to Bloom's taxonomy, and a criterion for success.
- Banned verbs: understand, know, learn, be aware of, appreciate.
- The performance must be something a reviewer could watch or score.

The business goal this training supports: [goal].

If your understanding of learning objectives is shaky, fix this first. My guide on how to write learning objectives covers the conditions-and-criteria structure that these prompts enforce.

How do you use ChatGPT for course outlines and idea generation?

Ideation is ChatGPT's most forgiving use case. Treat the output as brainstorming for you to judge (not something for you to copy and paste).

Back in early 2023, I asked ChatGPT to outline a YouTube video, "five steps to becoming an effective instructional designer." The outline was decent. It was not the five steps I would give, but it gave me something concrete to push against, and pushing against a draft is faster than staring at a blank page.

The engagement-idea variation is the one I recommend even to anyone getting started with AI. Back in 2023 I said that if you use ChatGPT for nothing else, use it for this, and the outputs impressed me even on those early models. Today's models are even better. Feed it your objectives and ask for ideas:

Here are the learning objectives for a course I'm designing:
[paste objectives]

The audience: [who they are, what they already know, and how they actually
feel about this topic].

Give me 10 ideas for making this course more engaging without changing the
objectives. For each idea, name the format (scenario, challenge, game
mechanic, story frame, simulation) and describe the first screen in one line.

Expect three or four duds, a few decent options, and one or two ideas worth developing. Ten ideas in thirty seconds means you evaluate instead of generate, and evaluating is where your expertise pays off.

How do you use ChatGPT for storyboarding?

ChatGPT drafts screen-by-screen storyboards fast if you hand it three things: the objectives, the source content, and your exact column structure. You still own the instructional decisions, but it does the typing.

It's predictable how it will fall short. Unconstrained, it writes voiceover that reads like an essay, piles text onto every screen, and fills the course with click-to-continue interactions that test nothing. So constrain it:

You are an eLearning storyboard writer. Using the learning objectives and
source content below, draft a screen-by-screen storyboard as a table with
these columns: Screen #, On-Screen Text, Voiceover Script, Visual/Media
Notes, Programming Notes.

Constraints:
- Maximum 40 words of on-screen text per screen.
- Voiceover is conversational, second person, no jargon.
- Every 3 to 4 screens, include an interaction that requires a decision
  with consequences, not a click-to-reveal.

Objectives: [paste]
Source content: [paste]

Then run a second pass on the narration alone, because first-draft voiceover always runs long:

Rewrite the voiceover script below so each screen is 60 words or fewer,
conversational, and written for the ear rather than the eye. Keep the
meaning, cut the filler. If cutting a screen's script would lose something
important, flag it instead of silently keeping it long.

Review the draft against your own storyboard standards before it goes anywhere near a stakeholder. If you're still building those standards, my guide to creating a storyboard for eLearning covers the components a strong one needs. And for a full worked example from a different corner of the field, Natalie Berkman's case study documents a real ChatGPT-assisted design project in higher ed, prompts and caveats included.

How do you use ChatGPT to write scenarios and dialogue?

ChatGPT is strong at drafting decision points, realistic-mistake distractors, and character dialogue. It is predictably bad at subtlety. Unconstrained, it writes melodrama, and it telegraphs the right answer so hard that no learner has to think.

The fix is to anchor every decision point in a real mistake people make on the job, and to demand distractors drawn from what people actually do:

I'm designing a scenario-based learning experience for [audience]. A real
mistake people make on the job: [describe the mistake and its consequence].

Draft one decision point:
- Set the scene in 2 to 3 sentences of concrete workplace detail.
- Write the character dialogue naturally. No melodrama.
- Give 3 response options: the best action plus two plausible wrong options
  based on what people actually do. No obviously dumb answers.
- For each option, write the realistic consequence. Show it through what
  happens next, not through a lecture.

Iterate hard on the distractors. If you can spot the right answer without reading the scene, so can your learners. For the design thinking behind all of this, see my scenario-based learning guide.

There's also a boundary that the prompt cannot cross. Everything above produces text that you paste into your authoring tool. A written branching scenario cannot listen to a learner, respond in character, adapt to what they actually said, and score the conversation inside your course. That kind of live, evaluated practice is a different category, and it's what AI role-play simulations exist for.

Full disclosure: this category is where my own company lives. devlin.ai is my AI simulation builder. You describe a scenario in plain English and get a working voice or text simulation that embeds in Storyline, Rise, or any LMS, with AI evaluation, transcripts, and scores reporting back to the course. Describe a conversation your learners need to practice and try the simulation it builds, right here in this article:

Can ChatGPT give useful feedback on your own work?

Yes. Critique is one of ChatGPT's highest-value, lowest-risk uses, because you are the accuracy check. It cannot hallucinate your storyboard; the storyboard is right there in the prompt.

Three critique workflows earn their keep:

1. Rubric review. Paste your storyboard and ask for a review against named criteria, including the multimedia principles from Richard Mayer's Multimedia Learning.
2. Objective audit. Paste your objectives and ask it to flag every unmeasurable verb and every missing condition or criterion. Yes, this is the same tool that once wrote those vague verbs itself. It is much better at catching them than avoiding using them.
3. Stakeholder role-play. Ask it to play a skeptical operations director reviewing your design document. The objections it raises are easy rehearsal for the real meeting.

The rubric-review prompt:

Act as a demanding instructional design reviewer. Review the storyboard
below against these criteria:

1. Alignment: does every screen serve a stated learning objective?
2. Multimedia principles: flag violations of Mayer's principles, such as
   redundant on-screen text plus identical narration, extraneous content,
   and split attention.
3. Practice quality: are the interactions decisions with consequences, or
   click-to-reveal filler?
4. Tone: does the voiceover sound like a person or a policy manual?

Be specific: quote the screen, name the problem, suggest the fix. Then
give me the three highest-impact changes.

[paste storyboard]

Ask ChatGPT for feedback without it and you'll get compliments. Ask for a demanding reviewer and you'll get notes that are actually worth applying.

Can ChatGPT help with analysis and needs assessment?

Yes, in three ways: summarizing intake documents, drafting SME interview questions, and running deep research. The rule across all three: verify every factual claim before it gets anywhere near a course.

For intake work, paste the kickoff document or intake form and ask it to extract the stated request, the implied business goal, the audience, and every unanswered question you should raise with the stakeholder. Then have it draft SME interview questions targeting those gaps. The output is a starting agenda, not a finished analysis, but it turns a blank page into a working document in minutes.

Deep research is the standout feature. Per OpenAI's deep research documentation, you describe the outcome you need, choose which sources it can draw on (websites, your uploaded files, connected apps), review and edit its proposed research plan, and get back a documented report. Controlling the source list is what separates this from regular chat, where source quality is a coin flip.

This feature is the part of ChatGPT I still use religiously, even as other tools have taken over parts of my daily drafting. I run a deep research pass every quarter or two on where instructional design and L&D are heading. It ensures I don't miss anything about where the field is going, and I don't have 10+ hours per quarter to do that depth of research by hand.

One warning, and it's the big one: standard chat (not deep research) will still invent citations. It will name plausible-sounding studies, attach real researchers to papers they never wrote, and do it all with total confidence. Any factual or research claim from a regular chat session should get verified at the source before you repeat it. No exceptions.

Where does ChatGPT fall short for instructional design?

Four failures account for nearly every ChatGPT problem I've seen in ID work: fabricated sources, generic tone, zero organizational context, and data privacy.

Fabricated sources. Covered above, and worth repeating: in standard chat, citations are unverified until proven otherwise.

Generic tone. Unprompted output sounds like everyone else's unprompted output. If your course voiceover could belong to any company in any industry, learners notice. The fix is context and examples of your actual voice in the prompt, plus your own editing pass.

Zero organizational context. Dr. Philippa Hardman's evaluation of Claude, ChatGPT, and Gemini at instructional design tasks makes the sharpest version of this argument: general-purpose models produce output that looks competent while missing learning science and the local constraints a real designer works within. I agree with the diagnosis. The prompting patterns in this guide are the working answer: you supply the theory, the constraints, and the context, and the model supplies drafting speed. What you cannot do is skip that supplying step.

Data privacy. When I talked about ChatGPT's future in early 2024, I said it could become one of the most important ID tools in the years ahead, but the resistance around copyright, ethics, and data privacy was real and reasonable: companies did not want their content pumped into the base ChatGPT. This concern still shapes policy today, and OpenAI's own documentation draws the line clearly. Per OpenAI's data usage policy, consumer-plan content can be used for model training unless you opt out, while business-tier customers are excluded from training by default.

Should you build a custom GPT for your ID team?

Yes, whenever a task repeats. A custom GPT encodes your prompt engineering once, so you (and your whole team) get consistent output without rebuilding the role-task-context scaffolding every session.

Learning objectives were my proof case. After fighting the vague-verb defaults through 2023, I built the Learning Objective Writer for Instructional Design, a custom GPT with the measurability rules, the banned verbs, and the eLearning-assessability constraint baked into its instructions. The gap between its output and the base model's default output made the case for custom GPTs better than any argument could. I also maintain a Become an Instructional Designer GPT for people exploring the career. Both are free to use.

I documented the whole build process, including the prompt-engineering tricks that make custom GPT instructions stick, in this video from when the feature launched:

The interface has evolved since I recorded it, but the instruction-writing principles transfer directly. Access has actually gotten cheaper: per OpenAI's pricing page, anyone can use custom GPTs on the free plan, and creating and sharing your own requires any paid plan, starting with Go. Here's the current lineup for individuals, from OpenAI's pricing page as of August 2026:

Plan	Price	What an ID gets
Free	$0	Default models, limited deep research, use (but not create) custom GPTs
Go	$8/month	Higher usage limits, create and share custom GPTs
Plus	$20/month	GPT-5.6 Sol access, expanded deep research, projects and scheduled tasks, custom GPTs
Pro	$100 or $200/month	5x or 20x Plus usage, GPT-5.6 Sol Pro, maximum deep research

There's a career angle here too. In the portfolio reviews I've done this year, most candidates blend in with tools-list portfolios: I can use Storyline, I can use Camtasia, I can use ChatGPT. Business context is rare, decisions go unexplained, and AI shows up superficially at best. A documented custom GPT, with the problem it solves, the instructions you engineered, and the before-and-after output, is the kind of demonstrated AI judgment almost nobody is showing yet. This is why students at Peck Academy, my career school, build an AI-integrated flagship project in Storyline 360 as the centerpiece of their portfolios.

When do you need more than ChatGPT?

You need purpose-built tools whenever the output must be a running learning experience rather than a document. ChatGPT is the universal drafting layer; delivery belongs to specialized tools.

The pattern holds across categories. ChatGPT writes a storyboard, but Storyline builds the course. It drafts a scenario, but a simulation platform delivers evaluated conversational practice. It describes the image you need, but a generation tool produces it. Expecting one chat window to do all of this is how teams end up disappointed with a tool that was doing its actual job fine.

For the full map of which specialized tools are worth your time across authoring, media, simulations, and analysis, see my AI tools for instructional designers roundup.

Frequently asked questions

Which ChatGPT plan do instructional designers need?

Start free. The free tier handles drafting, critique, and limited deep research, which is plenty for testing the workflows in this guide. Upgrade to Go ($8/month) if you want to create custom GPTs cheaply, or to Plus ($20/month) for the flagship model, expanded deep research, and projects. Check current details on OpenAI's pricing page, since plans and limits change often.

Is using ChatGPT for instructional design work cheating?

No. It's a drafting tool, and the instructional decisions remain yours: you set the objectives, judge the distractors, and own what ships. Two obligations come with it, though. Follow your organization's disclosure and AI-use policy, and never paste confidential content into a consumer plan without checking your org's data rules first.

Will ChatGPT replace instructional designers?

It replaces tasks, not the role: the drafting shrinks while the analysis, judgment, and stakeholder work grow in importance. I've written a full breakdown of whether AI will replace instructional designers and which skills hold their value.