What the Writing Tone Analyzer does
This analyzer reads the tone signals in a draft - how formal or casual it sounds, how confident or hedged, how warm or cool, how direct or indirect, and how much jargon it uses - and shows the exact words behind every reading, sentence by sentence.
It works from published word and phrase lists, not from a trained model, so each result can be checked: a letter reads as cool because of "unfortunately", "rejected" and "penalty", and the page highlights those words. It runs in your browser and the draft is not uploaded.
How to use it
- Paste an email, letter, post or reply. The readings update as you type.
- Pick a target tone if you have one - friendly and professional, formal, casual, assertive or an empathetic support reply. Each reading is then compared with a range.
- Read the four meters. Each shows the reading, a score out of 100 and how many signals it is based on. Fewer than 3 signals is reported as not enough evidence rather than guessed.
- Look at the highlighted draft and the sentence table to see which words produced each reading, then change the ones that pull away from the tone you want.
- Copy the summary or download the signals as CSV or JSON to share with a colleague or keep with the draft.
Reading the results
Each dimension has two opposite word lists. The score is 50 + 50 x (signals on the right-hand side - signals on the left-hand side) / all signals, rounded. So 100 means every signal points one way, 0 means every signal points the other way, and 50 means they balance.
Scores of 66 or more take the right-hand label (Formal, Confident, Warm, Direct), 34 or less take the left-hand label (Informal, Hedging, Cool, Indirect), and anything between is Mixed.
Jargon is reported per 100 words rather than as a score, because there is no opposite list: plain words are simply the absence of jargon.
Contractions, exclamation marks and emoji count as informal signals; sentences that begin with an instruction verb (Send, Book, Choose) count as direct; passive voice counts as indirect.
Worked example: a stern payment letter against a friendly-professional target
Load the stern payment letter. Formality has 4 signals, all formal - Dear, required, prior to and Yours sincerely - so it scores 50 + 50 x 4 / 4 = 100, Formal.
Warmth has 5 signals, all cool - Unfortunately, rejected, cannot, immediately and penalty - so it scores 50 + 50 x (0 - 5) / 5 = 0, Cool. Directness has 2 direct signals (your, You) and 4 indirect ones (has been rejected, were not received, It should be noted, be accepted): 50 + 50 x (2 - 4) / 6 = 33.3, shown as 33, Indirect.
Confidence has only one signal, "You must", so it is reported as not enough evidence. Against the friendly-professional target (formality 35-80, warmth 60-100, directness 55-100) the page suggests lowering formality, raising warmth and raising directness - for example by thanking the reader and rewriting the passive sentences as "We did not receive your documents".
Formulas and scoring rules
- Dimension score
round(50 + 50 x (pos - neg) / (pos + neg))Only when pos + neg >= 3; otherwise the page says there is not enough evidence.- Jargon density
jargon terms / words x 100Shown to one decimal place.
Why word lists instead of AI
A model that labels text as "aggressive" or "friendly" cannot show why, and its answer can change from one run to the next. Word lists are cruder, but every reading comes with the words that caused it, the same text always gives the same result, and you can see the complete lists under Every word list used.
The trade-off is that the lists cannot see context. "Great, another delay" counts as warm because of "great". Treat the readings as a checklist of words worth a second look, not as a judgement of the writer.
Limitations: what the result does not prove
- It counts listed words and patterns. It does not understand sarcasm, irony, quotation, negation ("not happy" counts "happy") or the relationship between writer and reader.
- The lists are for English and lean towards workplace writing - emails, letters, support replies and memos. Fiction, poetry and other languages will give misleading readings.
- A short text often has too few signals for a reading. That is reported as not enough evidence rather than as a neutral score.
- Target ranges are A2Z editorial starting points, not a standard. A formal legal notice and a formal thank-you letter can both be right with very different warmth.
Privacy: where your data goes
Everything you paste, type or drop is processed in this browser tab. It is not uploaded, logged, stored or sent to analytics. Session recording and tag-manager scripts are switched off on this page.
Standards and sources
- GOV.UK - Content design: writing for GOV.UK (active voice, plain English)
- plainlanguage.gov - Federal plain language guidelines
- Nielsen Norman Group - The four dimensions of tone of voice
Frequently asked questions
How can I check the tone of an email before sending it?
Paste the email, choose the target tone you want - for most workplace emails, friendly and professional - and read the gap list. It names the dimensions that are off target and highlights the words to change, such as hedges to remove or stern words to soften.
What makes writing sound formal or informal?
Formal signals include connectors such as therefore and regarding, set phrases such as prior to and yours sincerely, and words such as request and ensure. Informal signals include greetings like hey, words like awesome and stuff, contractions, exclamation marks and emoji. The full lists are on the page.
What are hedging words, and should I remove them?
Hedges soften a claim: might, perhaps, I think, seems, could. Keep them where you really are uncertain - they are honest. Remove them where you know the answer, because stacked hedges make a clear message sound unsure.
Why does my text say not enough evidence?
A dimension needs at least 3 matching signals before a score is shown. Short messages often contain one or two, and a score based on one word would be misleading. The words that were found are still highlighted.
Is this an AI tone detector?
No. It uses fixed English word and phrase lists and a simple formula, so it gives the same result every time and shows its evidence. It cannot understand meaning the way a person can.
Does passive voice make writing less direct?
Usually. "Your documents were not received" hides who did what; "We did not receive your documents" says it plainly. Plain-language guidance such as GOV.UK's recommends the active voice, and this page counts passive constructions as indirect signals.
Last reviewed by the A2Z.Tools team against the sources listed above.
