Text Simplifier

Most tools that promise to simplify text send it to a large language model and hope for the best. This one does not. Every change comes from a rule you can see in the change list, and every rule is a like-for-like swap. The meaning stays put because nothing is generated.

The trade-off is range. A model can merge two paragraphs or turn a list of clauses into a table. Rules cannot. What you get is the mechanical 70% of plain English done for you in a second, with the last 30% left to a human who knows what the text is for.

The dictionary holds 410pairs: single words (commence → start, ascertain → find out, sufficient → enough), padded phrases (prior to → before, at this point in time → now, in the event that → if) and verb phrases (make a decision → decide). Verbs are stored as base forms and inflected on the fly, so “utilized” becomes “used” and “demonstrating” becomes “showing”. UK spellings ending in -ise are caught too.

Capitals are kept: “Prior to” at the start of a sentence becomes “Before”. The article is fixed when the new word needs it, so “an alteration” becomes “a change”, and after has, was or been it picks the right participle (“has been shown”, not “has been showed”).

Any sentence longer than the slider value is cut at a semicolon, “, and” or “, but”, and the new sentence gets a capital letter. Both halves must keep at least five words, and a comma list like “forms, receipts, and letters” is left alone. It deliberately does not split at “which” or “who”, because doing that safely means knowing what the pronoun points to.

The filler pass deletes intensifiers and stock openers such as “very”, “really”, “basically”, “actually” and “in my opinion”, then tidies the commas and capital letter left behind. “Very much” and “very well” are skipped, since the sentence breaks without them.

Passive voice is flagged with a wavy underline, never rewritten, because the fix needs the name of whoever did the action. Contractions (do not → don't) are off by default: they read friendlier but some formal documents do not allow them.

Flesch reading ease is 206.835 − 1.015 × (words per sentence) − 84.6 × (syllables per word). Higher is easier; 60 to 70 is plain English, and below 30 is hard going. The Flesch-Kincaid grade uses the same two inputs to estimate a US school grade. Syllables are counted with a vowel-group rule, which is right for most words and off by one on a few, so compare the before and after figures rather than quoting one to a decimal place. For a fuller breakdown use the reading level checker.

Plain English is writing your reader can understand the first time through. The usual checklist:

• Short sentences. An average of 15 to 20 words, with nothing much past 25.
• Common words. Use, start, end, help, before. The words you would say to a friend.
• Active voice.“We will refund you”, not “a refund will be issued”.
• One idea per sentence. If a sentence has two, it is two sentences.
• You and we. Talk to the reader directly and say who you are.

None of this removes information. A plain version of a tax letter still has every deadline and figure in it. It just stops making the reader work to find them.

Some writing is dense on purpose, and a word swap can change what it means:

• Contracts and regulations.Terms like “terminate”, “shall” and “notwithstanding” can carry a defined legal meaning. Use the legalese translator to understand a contract, not to rewrite it.
• Medical and safety instructions. Splitting a dosage sentence can detach a condition from the step it belongs to. Keep the original next to the simple version and check them line by line.
• Quotes and citations.You cannot simplify someone else's words and keep the quote marks.
• Technical terms your reader already knows. An engineer reading a spec wants the precise word, not the friendly one.

1. 1Say the point first. Write the one sentence you would say if the reader gave you ten seconds. Put it at the top.
2. 2Cut sentences at the joins. Look for "and", "but", "which" and semicolons in any sentence past 25 words. Most of them are two sentences glued together.
3. 3Swap Latin for Anglo-Saxon. Commence, terminate, utilize, ascertain. Each has a short everyday word that means the same thing: start, end, use, find out.
4. 4Name who does what. Turn "the form must be signed" into "sign the form". Readers follow people doing things more easily than things being done.
5. 5Delete the throat-clearing. Openers like "in my opinion" and intensifiers like "very" and "really" rarely carry meaning. Read the sentence without them.
6. 6Read it out loud. If you run out of breath or stumble, the reader will too. Rewrite that line.

The tool above does steps 2, 3 and 5 for you and points at step 4. Steps 1 and 6 need a person.

A sensible workflow uses both: run the rules first so the easy fixes are guaranteed safe, then decide whether the rest needs a human or a model. If the text came out of a chatbot in the first place, the AI text cleaner strips the tells before you simplify it.

• Reading level checker scores your text on several readability formulas at once.
• Passive voice checker highlights every passive phrase and weasel word, with a percentage.
• Legalese translator explains contract language in everyday terms.
• AI text cleaner removes the stock phrases and odd characters that chatbots leave behind.

A simplifier that hands back a new paragraph asks you to trust it. This one shows its working. Every swap, split and cut is a chip you can click to undo, so a wrong guess costs you one click instead of a careful side-by-side read. The before and after numbers update as you go, which makes it easy to see which changes actually moved the score.