10 Writing Tips for Instructional Designers (With Rewrites)
By Devlin Peck · Updated
Part of the Instructional Design Fundamentals guide
Clear writing is the highest-impact, lowest-budget way to improve a learning experience, and it remains a core instructional design skill no matter which tools you use. In 2026 the skill has shifted, though. Most course copy now starts life as an AI draft, so the writer's value is less about producing words and more about editing them: cutting filler, restoring your voice, and making sure that every sentence earns its place.
This guide gives you ten writing tips, and every tip ends with a before-and-after rewrite so you can see the rule applied to a real sentence instead of just reading the rule.
I recorded the original video version of this guide back in 2022, and the fundamentals in it still hold up:
Why does writing matter more in the AI era?
Since AI produces fluent filler by default, the differentiating skill moved from drafting to editing. Any tool can now generate grammatically clean paragraphs about any topic. What the tools produce, left unedited, is what many people call "slop": unnecessary adverbs, empty intensifiers, generic transitions, and rhythmic sentence patterns that scan well but say nothing.
Learners feel this even when they cannot name it. Copy that reads as generated gets skimmed, and skimmed copy does not change behavior. So the tips below are written for the way courses actually get made now: some sentences you draft yourself, many you inherit from an AI agent or a SME document, and everything needs your editorial pass.
What are the best writing tips for instructional designers?
The best writing tips all serve one main idea: write like you talk, and cut everything that does not help the learner act. Here are the ten tips at a glance, then each one in detail with a rewrite you can use.
| Tip | The rule in one line | Example fix |
|---|---|---|
| 1. Write like you talk | Use "you," contractions, and everyday words | "Upon completion, employees will be equipped..." becomes "After this lesson, you'll know how to..." |
| 2. Use the active voice | Make the actor the subject of the sentence | "The report must be submitted by you" becomes "Submit the report" |
| 3. Cut the slop | Keep a banned-word list and enforce it | Delete "delve," "unlock," and "supercharge" on sight |
| 4. One idea per sentence | Split sentences beyond ~25 words | One 40-word storyboard sentence becomes three short ones |
| 5. Show, don't tell | Behavior beats explained emotions | "Maria was frustrated" becomes "Maria clicked Submit for the third time" |
| 6. Write to the objective | Every paragraph serves what the learner will do | A feature tour becomes "here's how you'll log a call in under two minutes" |
| 7. Make dialogue sound human | Contractions, interruptions, subtext | "I would like to state that..." becomes "Do you have a minute?" |
| 8. Narration is not on-screen text | Never read the slide aloud | A voiced paragraph becomes one spoken line plus a four-word key point |
| 9. Read it aloud | If you stumble, the learner stumbles | "Do not hesitate to contact" becomes "Need help? Contact..." |
| 10. Edit to a readability target | Aim for a middle school reading level | Run the draft through a readability check before it ships |
1. Write like you talk (conversational tone)
People learn better when the copy sounds like a person talking to them. This is not a style preference; it is Mayer's personalization principle, one of the most consistently supported findings in multimedia learning research. Use "you," use contractions, and use the words your learners actually say at work.
Conversational does not mean sloppy. It means the sentence would sound natural being said out loud to a colleague.
Before: "Upon completion of this module, employees will be equipped with the knowledge required to ensure compliance with the organization's data handling policies."
After: "By the end of this module, you'll know how to handle customer data safely."
2. Use the active voice
Active voice puts the actor first and cuts words at the same time. Passive voice creeps into compliance and policy content because it feels official, but it hides who is responsible for the action, which is exactly the thing training needs to make clear.
The quick test: if you can add "by zombies" after the verb and the sentence still works ("the report must be submitted... by zombies"), it is passive. Rewrite it.
Before: "The incident report must be submitted by the employee within 24 hours of the event being observed."
After: "Submit the incident report within 24 hours of seeing the incident."
3. Cut the slop: keep a banned-word list
Every published "AI slop word list" bans the same things: "leverage," "utilize," "dive into." Fine, ban those too. But the banned list I actually enforce on devlinpeck.com copy are a little different, because every entry on it earned its place by showing up in real copy and getting caught. The list is not static, and the highest-value bans turned out to be sentence cadences, not individual words.
Here is the real list, with the reason each entry is on it:
| Banned word or pattern | Why it earned its place | Write instead |
|---|---|---|
| Em dashes and en dashes | The strongest single tell of generated text. This is the one ban I enforce in code: my site's content schemas reject the characters outright, so they physically cannot build. You will not find one anywhere in this article. | A period, a comma, a colon, or parentheses |
| "Quietly" | Banned outright. "Quietly shipped" fakes modesty instead of stating a fact. | State the fact |
| "Delve" | The canonical LLM verb. No human says it about their own work. | "Look at," "dig into," or name the specific thing you did |
| "Unlock," "supercharge," "dive in" | Marketing-hype verbs that promise a transformation the sentence has not earned. | Say what the reader will actually be able to do |
| "The future of work" and similar hype labels | Hype vocabulary. These are claims doing work that adjectives shouldn't do. | The specific change, with a number if you have one |
| The "not X, but Y" reversal tic | A sentence shape, not a word: the rhetorical seesaw AI reaches for when it has nothing to say. | One direct claim |
| Staccato negation fragments ("No gate, no email wall, no teaser version") | Fragments stacked for rhythm read as generated. | A real first-person sentence |
| Urgency theater and fake scarcity | Banned from all my lead-gen copy. It manufactures pressure the offer does not have. A trust rule for the site. | The real deadline (or no deadline) |
That staccato-fragment ban came from a real editing pass on August 5, 2026, and the before-and-after is worth showing because the "before" is not obviously bad. It is punchy, it scans well, and it was in a few places on my site.
Before: "Every report from the survey will be free for everyone. No gate, no email wall."
After: "Every report from the survey will be free for everyone. I will never put the results behind an email form."
The fragments have rhythm but no author. The rewrite says the same thing as a sentence a person would actually say, and it is stronger.
"I will never" is a commitment; "no gate" is a slogan.Devlin Peck
Build your own version of this list from your own copy. When a word or pattern shows up in a draft and makes you wince, that is an entry.
4. One idea per sentence
Short sentences with front-loaded verbs are the backbone of plain language. The Federal Plain Language Guidelines make the same recommendation for government writing: short sentences, one idea each, the action up front. If a US federal agency can explain tax law this way, your onboarding course can survive it, too.
Anything much past 25 words is a candidate for splitting. Storyboard copy is where the 40-word monsters hide.
Before: "In the event that a customer requests a refund after the 30-day window has closed, employees should first consult the exceptions matrix, which can be found on the intranet, before escalating the request to their team lead for review."
After: "Sometimes a customer asks for a refund after the 30-day window closes. When this happens, check the exceptions matrix on the intranet first. If the case isn't covered there, escalate it to your team lead."
5. Show, don't tell
Labeled emotions tell the learner what to feel. Shown behavior lets them recognize the situation, which is what transfers to the job. This matters most in scenario copy, where an explained emotion flattens the story and a shown behavior makes it real.
Before: "Maria was frustrated with the new expense system and felt unsupported by her manager."
After: "Maria clicked Submit for the third time. The spinner froze again. She closed her laptop and started drafting a message to her manager: 'I give up.'"
6. Write to the objective, not the content dump
Every paragraph should help the learner do something differently after the course. When you write a content dump, you're usually not helping learners achieve anything. Buthen you write to support an action, you produce training.
I learned this one the draining way. Some of the least enjoyable work I have done as an instructional designer was writing storyboards for dry, info-heavy experiences that some early clients demanded. The copy exhausted me before it ever had the chance to bore a learner, and the root cause was always the same: no clear picture of what the learner needed to do. Start by writing measurable learning objectives, then make every paragraph earn its place against them.
Before: "The CRM was introduced in 2019 and consists of five modules: Contacts, Pipeline, Reporting, Automations, and Admin. Each module contains multiple sub-features, which will be reviewed in the following sections."
After: "By the end of this module, you'll log a customer call in the CRM in under two minutes. Here's the three-step flow you'll use every time."
7. Write scenario dialogue that sounds human
Real people speak in contractions, interrupt each other, and leave things unsaid. Scenario dialogue fails when characters speak in policy language, announce their feelings, or deliver the course content at each other. Give characters subtext: what they want but will not say directly.
The read-aloud test (tip 9) is mandatory here. If you cannot deliver the line without sounding like a robot in a customer service video, neither can your character.
Before: "Hello, David. I have observed that your reports have been submitted late on three occasions this month. It is important that all reports are submitted in a timely manner."
After: "David, got a minute? The last three reports went out late, and I want to figure out what's going on."
There is also a newer version of this skill. In scenario-based learning built with AI simulations, you increasingly write the persona and the evaluation criteria in plain English instead of scripting every branch line: who the character is, what they want, what they will push back on, and what a good response from the learner looks like. It is the same craft (specific, human, objective-driven) applied one level up. Full disclosure: this is exactly the input my own tool takes. devlin.ai turns a plain-English scenario description into a working conversation simulation that embeds in Storyline, Rise, or any LMS, and then AI evaluation, transcripts, and scores reporting back to the course.
Try the writing exercise yourself: describe a conversation your learners need to practice, in plain English, and see what it builds.
8. Write differently for narration than for on-screen text
Narration scripts and on-screen text are different writing jobs, and duplicating them is worse than sticking with one or the other. Narration is written for the ear: conversational, linear, and slightly redundant. On-screen text is written for the eye: scannable, short, and front-loaded. Mayer's redundancy principle covers why reading the slide aloud actively hurts learning: the learner's reading pace and your narration pace fight each other.
Before: On-screen: a 60-word paragraph about phishing. Narration: the same 60 words, read aloud.
After: Narration: "Most breaches start with one rushed click on a link that looked fine." On-screen: "One click starts most breaches."
9. Read it aloud
Reading your copy aloud is the easiest editing pass available, and it catches problems that no AI checker will. It is also the single test that enforces tips 1, 4, and 7 at once: tone, sentence length, and dialogue all fail out loud before they fail on a screen.
Before: "Should you require assistance, please do not hesitate to contact your designated departmental representative."
After: "If you need help, contact your department rep."
10. Edit with a checklist and a readability target
Give yourself a concrete finish line: a middle school reading level, roughly grades 6 to 8, for general workforce audiences. This is the target I recommended in the video above, and it is still the right one. A tool like the Hemingway Editor will grade your draft in seconds and flag the long, dense sentences dragging it up.
My own practice has evolved here. I used to run drafts through checkers routinely; as of 2026 I do not use them much anymore. I have trained AI on my voice profile and focus on keeping everything reader-friendly and free of jargon. But that instinct was built by years of checking, so if you are still building the habit, use the tool. Bake the pass into your storyboarding workflow so copy gets edited before it ever reaches a slide, and if you want to go deeper on the craft, several of the best eLearning and learning design books are really books about clear communication.
Before: "Familiarization with the escalation protocol is essential to the facilitation of timely incident resolution."
After: "Learn the escalation steps so incidents get resolved fast."
Want to see how your own copy holds up against these rules? Paste a passage into the checker below and it will flag banned words, passive voice, over-long sentences, and reading level, with a suggested fix for each.
Paste a paragraph of your own course copy. I check it against the rules in this article: the word-level bans from my Tip 3 banned list, passive voice patterns, sentence length, and an estimated reading level. The sentence-shape bans, like em dashes, the not-X-but-Y tic, and stacked fragments, still need your own read-aloud pass. You get a fix list you can work through in order.
Should you use AI to write course content?
Yes for first drafts, alternatives, and summaries. No for final copy. AI output is fluent but generic by default, and the edit you apply for clarity, brevity, and voice is what turns it into something that actually helps people learn.
I can tell you exactly where I draw the line in my own production, because my site runs on this split. AI drafts my article rewrites and refreshes, but every draft is caged three ways: it has to pull real stories from my experience bank (voice memos, YouTube transcripts, my LinkedIn posts, etc.) and is forbidden from fabricating anything; a linter mechanically enforces structure, and the content schemas reject em dashes at build time, so a slop draft cannot even compile; and I personally review and edit every article before it ships. What I always write by hand: the source material itself, the positioning and brand copy, and the standards documents (including the banned list above) that the AI is then held to.
The same split works for course content. Let AI draft the storyboard, the scenario skeleton, or three alternative phrasings of a tricky explanation. Then edit aggressively, watching for what AI reliably gets wrong:
- Hedging adverbs that fake modesty ("quietly," "simply," "effortlessly")
- The "not X, but Y" reversal tic filling space where a claim should be
- Stacked fragments that have rhythm but no author
- Empty intensifiers and hype verbs from the banned list above
- Throat-clearing openers that delay the point by a full paragraph
How do you defend plain language to SMEs and stakeholders?
Anchor the conversation to the learner's action, not the document's completeness, and offer the SME a home for the detail you cut. Most plain-language pushback is really a fear that cut content means lost content. Remove the fear and the objection usually goes with it.
Three scripts you can reuse:
- The action anchor. "What should someone do differently after this course? Let's protect that. Everything else can live in a linked reference doc, and I'll make sure it's one click away."
- The authority card. "Plain language isn't dumbing it down. US federal agencies are required by the Plain Writing Act to communicate this way, and the government's own plain language guidance exists because even legally mandated content works better in plain English."
- The read-aloud demo. "Can I read you both versions out loud?" Then read the SME's paragraph and your rewrite back to back. The formal version almost never survives being spoken.
The reference doc offer is the most important of the three. SMEs spent years earning that level of detail; give it a respectful home instead of a delete key, and they will let the course copy stay lean.
Frequently asked questions
What reading level should eLearning be written at?
Aim for a middle school reading level, roughly grades 6 to 8, for general workforce audiences. This is not about your learners' intelligence; it is about their attention. They are reading your course between meetings, and plain sentences transfer to the job faster than dense ones. A free readability check like the Hemingway Editor gives you the grade level in seconds. Technical audiences can handle domain vocabulary, but the sentence structure around it should stay simple.
Is a conversational tone unprofessional for compliance topics?
No. Conversational tone changes how you address the learner ("you," contractions, everyday words), not how rigorous the content is. Mayer's personalization principle shows people learn better from a conversational style, and compliance training is precisely where you cannot afford learners tuning out. Keep the facts exact and the legal terms accurate; loosen everything around them.
Do grammar rules like comma usage still matter?
Yes, because sloppy mechanics erode the learner's trust in everything else on the screen. The highest-value comma rule for course copy: when you join two complete sentences with "and," "but," or "so," put a comma before the conjunction, and never splice two complete sentences together with a comma alone. That said, clarity beats pedantry. If following a formal rule makes a sentence harder to say out loud, rewrite the sentence instead.