How to Prompt AI to Rewrite Something in Plain English
Plain English is not one instruction, it is about five. The prompt that names each transformation, and the output checks that catch softened obligations.
How to Prompt AI to Rewrite Something in Plain English
Asking for "plain English" produces shorter jargon. The model swaps long words for slightly shorter ones and leaves the sentence structure, the passive voice and the missing subject exactly where they were. What works is naming the specific transformations you want, because plain English is not one instruction, it is about five.
How to prompt AI to rewrite in plain English
Rewrite the text below in plain English.
Apply these specifically:
- Put a person or thing as the subject of every sentence. No "it is required that".
- Use active voice. Say who does the thing.
- Split any sentence over 25 words into two.
- Replace nominalisations with verbs: "make an application" becomes "apply".
- Keep every number, date, name and obligation exactly as written.
- Do not add anything that was not in the original.
Then list anything you were unsure about, rather than guessing.
TEXT:
[paste here]The last two lines do most of the work. Rewriting tends to drift into inventing, and an explicit instruction to flag uncertainty instead of resolving it is the difference between a rewrite you can trust and one you have to check line by line.
Why each rule is there
Rule | The failure it prevents |
|---|---|
Name the subject | Bureaucratic writing hides who must act. "Payment must be received" never says by whom. |
Active voice | Passive constructions are the single largest source of vagueness in official text |
Split long sentences | Models will happily produce a shorter version of a 60-word sentence that is still one sentence |
Verbs, not nominalisations | "Conduct an assessment" to "assess" removes a word and adds clarity at once |
Preserve numbers and obligations | This is where rewriting causes real harm. A changed deadline is worse than jargon. |
Add a reading level only if you mean it
Asking for a specific reading age works, but it is a blunt instrument. Requesting something very simple on a technical document produces text that is clear and slightly wrong, because the simplification has to drop qualifiers that were load-bearing. A useful compromise:
Aim for the reading level of a quality newspaper. Where a technical term has no plain equivalent, keep the term and define it once in a short parenthesis.
That instruction preserves precision where it matters and simplifies everything around it, which is usually what people actually want when they ask for plain English.
Check the output in one pass
Three quick checks catch most problems:
Search the rewrite for every number and date in the original. All present, all unchanged.
Read only the verbs. If they are mostly forms of "to be", the passive voice survived and the rewrite is cosmetic.
Check the obligations. "Must", "may" and "should" mean different things and models soften "must" into "should" surprisingly often.
That third one is worth taking seriously on anything contractual or regulatory. Softening a requirement into a suggestion is the most consequential error this task produces, and it is invisible unless you look for it.
When not to use this
Do not use a rewrite as the operative version of anything binding. Publish it alongside the original, clearly marked as a plain-English summary, with the original remaining the authority. That is standard practice for a reason: if the summary and the source disagree, somebody has to lose, and it should not be the person who trusted the summary.
If the output still comes back bland and hedged rather than clear, that is a different problem with a different fix, covered in avoiding generic AI output. And if the prompt above is not producing what you expect at all, work through diagnosing a prompt that is not working before adding more rules to it.
Frequently asked questions
Why does asking for plain English not work on its own?
Because it is a goal, not an instruction. The model has no specific edit to perform, so it substitutes shorter words and leaves the structure that made the text hard to read.
Can AI rewrite legal text safely?
It can produce a useful plain-English companion, but the original must stay the operative version. Check every obligation word and every number against the source.
Should I ask for a specific reading age?
Only when the audience genuinely needs it, and expect to lose qualifiers. A newspaper reading level with defined technical terms is usually the better target.
How is this different from summarising?
Rewriting keeps everything and changes how it is expressed. Summarising drops material on purpose. Mixing them is how detail disappears without anyone deciding it should. For the other task, see summarising a long document, and the general grounding is in the full prompt engineering guide.
How did this land?
About the author
Developer Advocate
Steve builds something with Swarmz every week and writes up what worked, what broke, and what he'd do differently. Tutorials and hands-on guides are his lane.
