Check AI content for plain English in two passes. First, run a readability checker such as Word's Document stats or the free Hemingway web app and flag long sentences and a grade level above about 8. Then read it as your least expert customer would, replacing jargon and splitting sentences. Scores find problems; only a human read confirms the fix.
The trap is trusting the score. Readability formulas only count sentence length and syllables, so "Grant admin consent on the tenant" scores as easy while baffling most customers, and a sentence can be short and still vague. AI drafts are especially good at sounding clear while hiding who has to do what, by when.
The routine below has four stages and takes about 25 minutes for a one-page email or notice. It is built around one illustrative business: a software reseller that looks after Microsoft 365 licences, backups and security tools for about 140 small-business customers, and that uses AI to draft customer notices full of technical terms.
Plain English is not the same as short words. The international plain language standard, ISO 24495-1, published in 2023, defines it by outcome: readers get what they need, can find it, can understand it and can use it. Keep that last one in mind throughout. A notice nobody can act on has failed, whatever its score.
What readability scores measure, and the jargon they can't see
Most plain English checkers use one or both of the Flesch formulas, the same ones Microsoft Word reports:
- Flesch Reading Ease = 206.835 − (1.015 × average sentence length) − (84.6 × average syllables per word). Higher is easier. Microsoft suggests 60 to 70 for most standard documents.
- Flesch-Kincaid Grade Level = (0.39 × average sentence length) + (11.8 × average syllables per word) − 15.59. It maps to a school grade. Microsoft recommends around 7.0 to 8.0 for most documents.
Both formulas reward short sentences and short words. Neither knows what a word means. That is why they miss the three problems AI content has most often:
- Short jargon. "SSO", "tenant", "SKU", "DPI", "bleed" are all short, so they score as easy. To a customer they are noise.
- Missing actors. "Consent must be granted prior to migration" scores reasonably and still doesn't say who grants it.
- Missing steps. A draft can be perfectly readable and skip the one step the customer needed, such as where to find the link.
So use the score as a smoke alarm. It tells you where to look, and the later stages decide what to change.
Stage 1: score the draft in Word or Hemingway (five minutes)
You probably already have a plain English checker. In Word for Microsoft 365, open the document, go to the Home tab, choose Editor, then Document stats, and Word shows the reading ease and grade level. In older versions, go to File, Options, Proofing, tick "Check grammar with spelling" and "Show readability statistics", then run Spelling & Grammar; the statistics appear when the check finishes. Microsoft's own page on readability statistics in Word has the steps for each version.
The free Hemingway Editor web app goes further for editing, because it highlights individual problems: sentences that are hard or very hard to read, passive voice, adverbs, and an overall grade level. Its paid Hemingway Editor Plus adds AI rewrites, starting at $100 a year billed annually as of September 2026; for checking, the free version is enough.
One caution before you paste anything into a free website: check its privacy terms, and strip customer names and account details first. Word's statistics run on the document already open on your machine, which is simpler for anything sensitive.
Record five numbers for each draft. These are the thresholds the reseller uses as house rules; only the first two come from Microsoft's guidance, the rest are sensible targets rather than standards.
| Measure | Target for customer-facing copy | Red flag |
|---|---|---|
| Flesch Reading Ease | 60 or higher | Below 50 |
| Flesch-Kincaid Grade | 8 or lower | Above 10 |
| Average sentence length | Under 20 words (house rule) | Over 25 words |
| Longest sentence | Under 30 words (house rule) | Any sentence over 40 words |
| Undefined acronyms | None | Any acronym a customer hasn't seen from you before |
Stage 2: make the AI list its own jargon and long sentences
Before rewriting, get an inventory of problems. The same assistant that wrote the draft is decent at spotting jargon when you ask it to audit rather than rewrite, because the task is narrower. Five minutes, and it stops you fixing things at random.
You are checking a customer email for plain English. Do not rewrite it.
The reader runs a small business and is not technical.
List, in a table with columns Problem | Exact text | Why a customer might struggle:
1. Every term or acronym a non-specialist may not know.
2. Every sentence over 25 words.
3. Every sentence where it is unclear WHO must do something.
4. Every instruction that lacks a deadline, a location (where to click) or a consequence.
Then list any facts a customer would need that the email does not give.
Email:
[paste the draft]This is an illustrative audit of the reseller's AI-drafted notice about moving customers to a new backup service:
| Problem | Exact text | Why a customer might struggle |
|---|---|---|
| Jargon | "cloud-to-cloud backup solution" | Doesn't say what is backed up (email? files?) |
| Jargon | "grants admin consent ... via the OAuth authorisation link" | Most owners won't know what consent or OAuth means or who can give it |
| Long sentence | "In order to facilitate the migration ... continuity of protection." (43 words) | The deadline is buried at the end of a long sentence |
| Unclear actor | "it is required that a Global Administrator ... grants" | Customer may not know who their Global Administrator is |
| Missing consequence | "to ensure continuity of protection" | Doesn't say what happens if nobody acts |
What the audit missed is as instructive as what it found. It did not flag "tenant" in the first sentence, a word the assistant uses so often it treated it as ordinary English, and it did not notice that the email never says whether the change costs anything. So add your own pass: circle every word you use at work but wouldn't use with a family member, and ask "what will the customer ask us after reading this?"
Stage 3: rewrite one problem type at a time
Fixing everything in one pass tends to produce a rewrite that is shorter but still muddled. Work through the problem types in order, because each fix makes the next one easier to see. Here are the four moves with small before-and-after examples from other businesses, all illustrative.
Define jargon you need; delete jargon you don't
Some technical words are unavoidable, and deleting them leaves customers unable to follow instructions from their own software. A print shop can't avoid "bleed" or "CMYK", because design apps use those words. The fix is to define them once, in passing.
Before: Please ensure all artwork is supplied as a print-ready PDF in CMYK colour space at a minimum resolution of 300dpi with 3mm bleed on all sides and a 5mm safe zone.
After: Send your design as a PDF. Three settings make it print well. Set colours to CMYK, not RGB (it's in your design app's document settings). Use images of at least 300dpi, or they may look blurry. Add 3mm of extra background around every edge, called "bleed", because we trim the paper after printing, and keep text 5mm inside the edge so none gets cut off.
Split sentences at every "which", "however" and semicolon
AI drafts chain clauses with "which", "whereby", "however" and semicolons. Each join is a natural place to end one sentence and start another. A sentence with two instructions in it usually gets one of them ignored.
Name the person who acts
Passive voice ("refunds will be processed") hides who does what. Plain English says "we" and "you". This clothing shop returns policy shows the difference.
Before: Returns are accepted within 30 days of receipt provided that items are returned in their original condition with all tags attached, subject to inspection. Refunds will be processed to the original payment method within 5-7 business days following receipt and inspection of returned merchandise.
After: You can return anything within 30 days of it arriving, as long as it's unworn with the tags still on. Once it reaches us, we check it and refund you within 7 working days, to the card you paid with.
Put the action and the deadline first
If the reader has to do something, say so in the subject line and the first two sentences. Background goes after. AI drafts do the opposite, opening with "As part of our ongoing commitment to..." because that is how corporate announcements they were trained on begin.
The reseller's migration email, scored before and after
Here is the worked example in full. The reseller's AI draft went to its first batch of customers unedited, which is how the business learned this lesson.
Before (AI draft, sent to the first 40 customers): As part of our ongoing commitment to delivering best-in-class data protection, we are pleased to inform you that we will be transitioning all customer tenants from our legacy backup platform to a next-generation cloud-to-cloud backup solution, effective 1 November. In order to facilitate the migration, it is required that a Global Administrator within your organisation grants admin consent to the new application via the OAuth authorisation link provided below, which must be completed prior to 25 October to ensure continuity of protection. Please be advised that retention policies will be migrated automatically; however, historical restore points older than 90 days will remain accessible via the legacy portal until the conclusion of your current contract term. Should you have any questions or require further assistance, please do not hesitate to reach out to our support team, who will be more than happy to assist.
After (edited, sent to the remaining 100): Subject: Action needed by 25 October: approve your new backup service. We're moving your email and file backups to a new backup service on 1 November. It does the same job and costs you nothing extra. We need one thing from you. Someone with full admin rights to your Microsoft 365 account must click the link below and press Accept. It takes about two minutes. This gives the new service permission to copy your data. Please do this by 25 October. If nobody approves it, your backups will stop on 1 November. Your settings for how long backups are kept will move across automatically. Backups older than 90 days stay in the old system, and you can still restore from them until your current contract ends. Not sure who your admin is? Reply to this email and we'll check it for you.
Scored with the Flesch formulas (tools count syllables slightly differently, so expect a few points of variation):
| Measure | AI draft | Edited version |
|---|---|---|
| Words (body) | 143 | 131 |
| Sentences | 4 | 12 |
| Average sentence length | about 36 words | about 11 words |
| Longest sentence | 43 words | 22 words |
| Flesch Reading Ease | about 21 | about 82 |
| Flesch-Kincaid Grade | about 19 | about 5 |
The word count barely moved. What changed was the number of sentences, the order (action and deadline first) and four definitions: "admin rights" instead of Global Administrator, "press Accept" instead of grant consent, "how long backups are kept" instead of retention policies, and the consequence spelled out. The edit also added two facts the draft never gave: the cost (nothing extra) and what to do if you don't know who your admin is.
The results here are illustrative: the first 40 customers sent 11 "what do I need to do?" replies and 6 missed the deadline. The next 100 sent 3 questions and 2 missed it. At roughly 10 minutes per support reply and follow-up call, the edit saved the team about two hours in chasing on this one notice, for 20 minutes of editing.
Stage 4: a five-minute test with a real reader
Scores and audits can't tell you whether someone can act on the page. A reader can. Give the edited draft to someone who doesn't know the subject (a colleague in another role, a friend who runs a different kind of business) and ask them three questions, without letting them reread:
- What is this asking you to do?
- By when?
- What happens if you don't?
If any answer is wrong or hesitant, that part of the text is the problem, whatever its score. For the reseller, the first tester answered question 1 with "click a link", which was right but incomplete, because she didn't realise it had to be someone with admin rights. That is why the edited version gives the admin requirement its own sentence and offers to check who the admin is.
Anything important also deserves a read on a phone. Plain English on a laptop can still turn into a wall of text on a small screen, and most of the reseller's customers open email on their phones first.
Plain English edits that quietly changed the meaning
Simplifying is editing, and editing can change facts. These are the realistic ways it goes wrong, including the one that cost the reseller a dispute.
- A commitment became a monthly plan. The reseller asked an assistant to "simplify" its terms page. "Annual commitment, billed monthly" came back as "a simple monthly plan". Two months later a customer asked to cancel, reasonably believing the plain English version, and the business had to honour the rewrite rather than argue about it. The fix is a rule: when simplifying anything about money, length of contract or cancellation, compare every number and condition line by line with the original before publishing. For terms of business themselves, the tutorial on writing terms of business with AI and what needs a lawyer explains where a professional has to be involved.
- A condition fell off. An events company's cancellation clause ("refundable if cancelled more than 90 days before the event, less the deposit") was simplified to "you can cancel for a full refund up to 90 days before". Three words gone, a $500 difference on every cancellation. Assistants drop conditions because they are the complicated part.
- "May" turned into "will". Hedged statements ("delivery may take up to 10 days in busy periods") get firmed up into promises ("delivery takes 10 days"). Check the verbs of every promise.
- Too simple for the audience. A notice to IT managers rewritten at grade 4 reads as patronising and strips out the technical detail they need. Plain English means clear for the actual reader. For a technical audience, keep the terms and fix the sentence structure.
- Score-chasing. Told to "get the reading ease above 70", one assistant chopped sentences into fragments. "Backups. Moving. 1 November." scores beautifully and reads like a telegram. The reader test catches this; the score never will.
If you check translations as well, simplify first: a plain English original translates more reliably, and the tutorial on checking AI translations before customers see them picks up from there.
Writing plain English rules into the prompt
Once you've edited a few drafts, move the lessons upstream so the next draft needs less work. These are the rules the reseller now keeps in its saved instructions, taken directly from its own mistakes:
Write for small-business owners who are not technical.
- Put the action and deadline in the subject line and first two sentences.
- One idea per sentence. Keep sentences under 20 words on average; none over 30.
- Say who acts: "we", "you", "your admin". No passive voice for instructions.
- Define any technical term the first time, in plain words, or leave it out.
- Say what happens if the reader does nothing.
- Never change numbers, prices, dates, contract lengths or conditions when
simplifying. If a condition seems complicated, keep it and explain it.
- End with one way to get help, not a list of options.Rules in the prompt reduce the work but don't remove the check. The reseller still scores every customer notice, runs the audit prompt, and has one person do the three-question read before anything goes to more than ten customers. If you want a second opinion on grammar and clarity tools, the comparison of Grammarly and ChatGPT for checking business writing covers what each catches. And if several people draft customer notices, fold these rules into a house style guide your team and AI both follow, so everyone edits towards the same standard.
Plain English checks: follow-up questions
What reading level should customer emails aim for?
Microsoft suggests a Flesch Reading Ease of 60 to 70 and a grade level around 7 to 8 for most standard documents. For emails asking customers to do something, going easier than that is fine as long as nothing important is dropped. Treat the score as a warning light, then judge the draft by whether a reader can act on it.
Can I just tell the AI to write at a lower reading age?
It helps, and it is worth putting in your prompt, but check the result for meaning. When an assistant simplifies to hit a target it tends to cut conditions, exceptions and numbers, which are often the parts that matter legally or financially. Compare the before and after line by line for any missing condition.
Is it safe to paste customer emails into a free readability website?
Only if you have checked the tool's privacy terms. The simplest habit is to remove names, account numbers and anything identifying before pasting, or to use the readability statistics already built into Word, which run on the document you have open.
Do readability tools work for languages other than English?
The Flesch formulas were designed for English, and applying them to other languages gives misleading numbers. For other languages, rely on a fluent reader and short, direct sentences rather than an English score.
Further reads
- How Much Does Human Editing of AI Content Cost? — What a human editor charges if you want a pro to do the rewrite.
- A Five-Minute Fact-Check Routine for AI Output Before It Goes Out — Pair readability checks with a quick fact check before sending.
- Tone of Voice Examples: AI Copy Before and After Editing — Before-and-after edits that fix tone as well as clarity.
- Email Newsletter QA Checklist Before You Hit Send — A full pre-send checklist for newsletters and customer emails.
- How to Audit Your Website for AI Content That Needs Fixing — Find the AI-written pages on your site that need a plain English pass.
- How to Draft Customer Email Replies With AI That Sound Like You — Prompts for customer replies that start out clear.
- How Accountants Use AI to Explain Tax to Clients in Plain English — A facts-block method for turning tax computations into plain-English client explanations, with prompts, a readability check and a review routine.
- How Coaches Use AI to Build Workbooks, Exercises and Programmes — Turn your coaching method into a programme and workbook with AI doing first drafts, while the method, the edits and the claims stay firmly yours.
- How Funeral Directors Can Use AI for Admin With Care — Where AI can take funeral admin off your desk, where it must never speak for you, and the read-back routine that stops a misspelt name reaching print.
- Can AI Draft a Suitability Report That Passes Compliance Review? — What file checkers test in a suitability report, a prompt that ties each reason to the fact-find, five ways AI drafts fail, and a pre-submission check.
- Explaining Dental Treatment Plans to Patients With AI — A dentist-checked workflow for turning clinical treatment plans into plain-English explanations patients can reread and compare at home.
- Explaining Lens Options and Eye Test Results With AI Help — Prompts and checked templates for explaining prescriptions, lens choices and eye test findings to patients in plain English, with the clinician in charge.
- Creating Home Exercise Programmes With AI for Physio Patients — A physio-led workflow for AI-assisted home exercise programmes: shorthand to patient instructions, pain rules, delivery tools and adherence messages.
- How to Write Your Website Copy With AI: Home, About, and Services — Page-by-page prompts for your Home, About and Services copy, built from a fact sheet, with sample outputs, a claims check and a 20-minute customer test.
- Should You Pay for an AI Humaniser Tool? What It Does and Risks — What AI humaniser tools really do to business copy, why beating detectors is the wrong goal, and the editing routine that fixes robotic drafts instead.
- AI Words to Avoid: A Banned-Phrase List for Your Brand — A ready-made banned-phrase list for AI drafts: the words and sentence patterns that sound generic, what to write instead, and a five-minute check.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Support, Get your document's readability and level statistics (formulas and recommended ranges); Hemingway Editor website (free web app and Hemingway Editor Plus pricing); International Plain Language Federation page on ISO 24495-1 (2023).