Using AI to Rewrite Training Content in Plain Language
You've been handed a 40-page procedure written by engineers for engineers, and the course has to work for new hires on their second week. Every sentence has three clauses and two acronyms. You could rewrite it by hand, line by line, for days, or you could let plain language AI tools do the first pass while you focus on accuracy. Can AI really rewrite training content in plain language without changing what it means?
Yes, AI tools are very good at a first-pass plain language rewrite, as long as you give them a clear audience, firm rules about what must not change, and a person who checks the result against the source. Tools like ChatGPT, Claude, Gemini and Microsoft Copilot can shorten sentences, swap jargon for everyday words and turn long paragraphs into steps in seconds. Where they need help is meaning, because they sometimes soften a requirement, drop a condition or "simplify" a regulated term that has to stay exact. Treat the output as a strong draft, have a subject matter expert (SME) confirm the facts, and you can cut rewrite time sharply.
What Counts as Plain Language in Training?
Plain language is writing your audience can find, understand and act on the first time they read it. It isn't the same as dumbing down content. A pipefitter and a pharmacist both deserve clear sentences, even though their vocabularies differ.
The U.S. government's plain language guidance on Digital.gov is a solid reference. The core moves are simple:
- Write for one clear audience.
- Use common words and short sentences.
- Prefer active voice: "Close the valve," not "The valve should be closed."
- Put the main point first.
- Use lists and steps for sequences.
- Define the technical terms you must keep.
In training, there's one more rule: the rewrite has to support the learning objective. If learners must recognize a specific term on the job, keep the term and explain it rather than replacing it.
A Plain Language AI Prompt That Works
The difference between a mushy rewrite and a useful one is almost always the prompt. Vague requests ("make this simpler") get vague results. Here is a prompt pattern I use, adapted to the content each time:
You are an editor who writes plain language training content. Rewrite the text below for new warehouse employees with a high school reading level who are learning this procedure for the first time. Rules: keep every requirement, number, limit and safety warning exactly as written. Keep these terms unchanged and add a short definition the first time each appears: lockout/tagout, PPE. Use active voice, short sentences and numbered steps for any sequence. Do not add new information. After the rewrite, list every place where you were unsure the meaning stayed the same.
That last line pulls a lot of weight, because asking the tool to flag its own uncertain spots gives your reviewer a short list to check first. It doesn't replace a full review, but it points attention where it matters.
Before and After: What a Good Rewrite Looks Like
Here's a typical example of the kind of sentence that shows up in procedures, followed by the kind of rewrite a good prompt produces.
Before: "Prior to the initiation of any maintenance activities on powered equipment, it is required that all energy sources be isolated and verified as being in a zero-energy state by the authorized employee in accordance with the facility lockout/tagout program."
After: "Before you work on any powered equipment, you must shut off and lock out every energy source. Then check that the equipment has zero energy. Only an authorized employee can do this. Follow your facility's lockout/tagout program (the steps for locking and tagging equipment so it can't start by accident)."
The rewrite is longer in words but much faster to read. Each sentence carries one idea. Notice that "must," "every," "zero energy" and "authorized employee" all survived, and those are the words a reviewer checks first.
Where AI Rewrites Go Wrong
AI rewrites go wrong in predictable places, which is good news, because predictable problems are easy to check. Watch for these:
| Problem | What it looks like | How to catch it |
|---|---|---|
| Softened requirements | "Must" becomes "should" or "try to" | Search the output for should, may, try, consider |
| Dropped conditions | "Unless the line is depressurized" disappears | Compare every if, unless and except to the source |
| Changed numbers or limits | A range becomes one number | Highlight every number in both versions and match them |
| Swapped regulated terms | A defined legal or safety term gets a friendly synonym | Give the tool a list of locked terms in the prompt |
| Added content | A helpful tip that isn't in your policy | Ask the reviewer to strike anything not in the source |
None of these mean the tool is a bad fit; they mean the review step is part of the job, the same as a second editor would be.
A Review Routine Your SMEs Will Actually Do
SMEs are busy, so make their review fast and focused. Don't send them the whole rewritten module with "please review." Send a side-by-side document: source on the left, rewrite on the right, with the AI's own "unsure" list and your highlighted numbers at the top.
Ask them three questions only:
- Is any requirement weaker or stronger than the source?
- Is anything missing or added?
- Would a new employee do the right thing after reading this?
Most SMEs get through a module this way much faster than they would with a cold read. Many of them will also tell you the rewrite is clearer than the original, which tends to make the next review go even faster.
How Do You Handle a Long Manual?
Handle a long manual in sections, not all at once. Paste one procedure or chapter at a time, so the tool keeps its attention on a manageable chunk and your reviewer can match each rewrite to its source page.
Keep a running list of locked terms and approved definitions, and paste it into every request. That keeps "isolate," "verify" and "authorized employee" consistent from chapter one to chapter twelve. Note the source document's version number at the top of each rewrite, too, so that if the procedure changes mid-project, you'll know exactly which sections need another pass. And because procedures often contain internal details, use a tool your organization has approved for that kind of content.
Should You Trust Readability Scores?
Readability scores are a useful gauge but a poor goal. Formulas like Flesch-Kincaid count sentence length and syllables, but they can't tell whether a sentence makes sense or whether the steps are in the right order.
Use a score to compare before and after versions and to catch a section that drifted back into long sentences. Then judge the rewrite the way it will be used: hand it to someone from the target audience and watch whether they can do the task. That five-minute test beats any number.
Use Plain Language AI for the First Pass, Then Check Every Requirement
Plain language AI tools can take the most painful part of the job, the first rewrite of dense source material, and finish it in minutes. Your job shifts to setting clear rules, checking meaning and confirming with an SME. That's a better use of an instructional designer's time anyway.
Pick the densest procedure on your update list. Build a prompt with your audience, your locked terms and the "flag what you're unsure of" line. Then run the side-by-side review with your SME. Once it works, the same draft-and-check habit carries over to other accessibility work, like using AI alt text to describe course images.