A text expander that tells you what it invented
Turning a few bullet points into a proper paragraph or email is exactly the kind of writing AI is good at — and exactly where it is tempted to make things up. This text expander turns your notes into paragraphs or an email of about 80, 150 or 250 words, in a neutral, formal, friendly or persuasive tone, and then checks the result against your notes.
The check is done by code, not by the AI. A “Facts I provided” list ticks off each of your notes that made it into the text and says what is missing from the others. Any name, number, date, time or price in the text that is not in your notes is highlighted in amber, so an invented lunch time or a made-up person can’t slip through. The model is told to write [placeholders] when it needs a detail you didn’t give; those are highlighted too.
Everything runs on your device: Chrome’s built-in AI where available, otherwise IBM’s open Granite 4.0 model (Apache-2.0, run by transformers.js), downloaded once after you agree. Results vary from run to run, so read the text before you use it.
How to expand notes into text
- Type or paste your notes — one point per line works best. Bullets, numbers or plain lines are all fine.
- Optionally say what it is: “an email inviting the team”, “a product description”, “the introduction of a report”.
- Choose the Length, Tone and Format (paragraphs or an email with greeting and sign-off).
- Press Expand notes. If a model must be downloaded, the page asks first. The text appears as it is written; Stop ends it at once.
- Go through Facts I provided, the amber highlights and the placeholders. Fix or remove anything that isn’t right, then copy or download the text.
Text Expander features
Notes checklist
Each note counts as used when all its names, numbers and dates appear and at least half its main words do; unused notes show what was not found.
Invented details flagged
Names, numbers, prices, dates, times, emails and links that are not in your notes are highlighted where they appear and listed for you to check.
Placeholders, not guesses
The model is told to write [square-bracket placeholders] for missing details such as a time or a name, and they are listed so you fill them in.
Length, tone and format
About 80, 150 or 250 words; neutral, formal, friendly or persuasive; paragraphs or a complete email.
Long notes in parts
Long lists are expanded a group of lines at a time so a slow on-device model keeps showing progress, with the email greeting in the first part and the sign-off in the last.
Why AI invents details — and how this page catches them
Language models write by predicting likely next words. When notes say “lunch provided”, a likely continuation is “lunch will be served at 1 pm” — plausible, specific and possibly wrong. The more you ask a model to expand, the more of these fillers appear. Instructions help (this page tells the model to use only your facts and to write placeholders instead), but small on-device models follow instructions loosely.
So the page compares the facts in your notes with the facts in the result. It recognises equivalent forms — “24 October” and “October 24th”, “₹4,500” and “4500 rupees”, “12” and “twelve” — and flags everything else that is new. It cannot catch an invented claim without a name or number in it (“the venue has a pool”), so read the text through once before sending it.
Tips for notes that expand well
- Put one fact per line: who, what, when, where, how much. The checklist works line by line.
- Include the details you care about, even if they seem obvious — a date, a deadline, the person to contact. Anything missing becomes a placeholder or, worse, a guess.
- Say what the text is for in the “What is it?” box; it changes the structure more than the tone does.
- Ask for less rather than more. 150 words from five notes reads naturally; 250 words from two notes means padding.
- Need it shorter or in a different voice afterwards? Use the Text Shortener or the Tone Changer, and check the length with the Word Counter.
Text Expander FAQ
How do I turn bullet points into a paragraph?
Paste the bullet points into this text expander, choose a length and tone, and press Expand notes. Then check the “Facts I provided” list and the amber highlights before you use the text.
Is my text sent to a server?
No. Your notes are expanded on your device by your browser’s built-in AI or the open Granite model, and the checks are plain code in the page. Only the model files are downloaded, once, after you agree.
Why is a number highlighted in amber?
Because it is not in your notes. The model may have invented it, or written one of your numbers in a different form. Check it, and delete it if it isn’t true.
What are the words in [square brackets]?
Placeholders. The model is told to write one when it needs a detail you didn’t give — a time, a name, a link — instead of making it up. Replace each one before you send the text.
Can it write a whole essay from my notes?
It is built for paragraphs and emails up to about 250 words. Expanding notes into a long essay mostly adds padding, and for coursework you should follow your institution’s rules on AI use.
Why is the text shorter or longer than I asked?
Models estimate length loosely, especially small ones. The note under the result shows the actual word count next to your target.