Introducing the WriteHuman Chrome Extension: Humanizing and Grammar in Every Text Box

The WriteHuman Chrome extension is out today. It checks your spelling and grammar as you type, and it rewrites AI drafts in the box you are already writing in, whether that is Gmail, Notion, LinkedIn or anything else with a text field.

Here is what it replaces. You get a draft out of ChatGPT or Claude, and it is decent, but it reads like a model wrote it. So you open a humanizer in a second tab, paste the draft in, wait for it, copy the result, switch back to where you were, and paste it in. Somewhere in that loop you spot a typo, and fixing that properly means a third tool.

That loop is far and away the most common way a humanizer gets used, and every step of it happens somewhere other than the page the writing is actually going to end up on. It is not a hard problem. It is a tedious one, repeated several times a day, and it has been sitting in front of us the whole time.

No second tab. No pasting. No losing your place.

Two jobs that only look like one

Writing extensions tend to pick a side. The humanizers rewrite AI text and never once look at your spelling. The grammar checkers fix your typos and leave an AI draft sounding exactly as synthetic as they found it. Both are perfectly good products. Neither one covers an ordinary afternoon of writing, because an ordinary afternoon has both problems in it, usually in the same paragraph.

The first half is the familiar one. Misspellings, disagreements and stiff phrasing pick up a coloured underline where you wrote them, and the correction is sitting there when you hover over the word. None of that is new and we are not going to pretend we invented it.

The second half is the one that took real work, because a grammar checker is no help whatsoever with AI writing. This is the thing people miss. AI writing is not incorrect. It is even. The sentences arrive at a similar length, the paragraphs land at a similar weight, the transitions come off a short and familiar list, and not one word of it breaks a rule that a checker knows anything about. Run it through a spell checker and the spell checker will tell you it is perfect, because by its own standards it is.

You can usually feel it before you can name it. Every sentence lands at the same confident middle length. Nothing runs on too long, nothing gets clipped short for effect. There are no asides, no half-thoughts, no moments where the writer got interested in something and followed it for a line before coming back. Real writing has a pulse, because a real person gets bored, or emphatic, or runs out of breath in the middle of a thought. Generated writing is flat in a way that is genuinely hard to point at and very easy to notice.

Here is the whole problem in a single line. "It is imperative that we leverage our core competencies" is spotless. Spelling, grammar, agreement, punctuation, all clean. It is also a sentence that no human being has ever said out loud to another human being. What you actually wanted was something closer to "the thing that matters is playing to what we are already good at." Getting from the first to the second is not a correction. It is a rewrite, and no amount of fixing errors will ever get you there, because there were never any errors.

That is the second job. It is why the two of them belong in the same place, and why we stopped treating them as separate products.

In practice it behaves the way you would hope. Highlight a paragraph and it rewrites that paragraph. Highlight nothing and it takes the whole box. What comes back carries a human score so you can see how it reads before you commit to it, and a single click drops it into the page you were already working in.

The practical version: you are in Gmail, you have pasted in something a model wrote for you, and it reads like a press release nobody asked for. You select it, you humanize it, and what comes back sounds like you on a reasonably good day. One click and it is in the message. The whole thing takes about as long as it took to read this paragraph.

What it costs

Everything in this category calls itself free, so it is worth being plain about what that word covers here.

Installing costs nothing, and the spelling and grammar checking runs without an account at all. Underlines show up the first time you type into a box. Taking a correction does mean signing in, but a free WriteHuman account covers it, and there is no upgrade prompt lurking behind that step.

Humanizing is the part that runs on your plan, for the straightforward reason that it costs us real money every time it runs. It draws on the same account and the same word allowance as the website, so there is no second subscription and no separate balance to keep topped up. The free plan includes three humanize runs a month at up to 250 words each, which is enough to work out whether you like it and nowhere near enough to actually work on. Paid plans start at $18 a month, or $12 a month if you pay for the year.

Where it works

Wherever you type, which is the honest answer and also a slightly useless one, so here is the specific version.

Gmail and Outlook on the web, where it also quiets the duplicate squiggles those two draw themselves, so you are not reading two sets of underlines over the same unfortunate word. LinkedIn posts, comments and the connection notes nobody enjoys writing. Notion pages and database notes. Slack messages and canvases. X, where a stiff cadence gives itself away fastest, because there is so little text for it to hide in. WordPress and most other content systems, old editors and new ones alike.

That is a description of where people write, though, not a list of what we support. Underneath it the extension attaches to ordinary text boxes and to the rich editors most modern web apps are built on. Support is the default state, not something we add one site at a time, so a tool you use every day and cannot find named here is very probably covered already.

The one place it does not work

Google Docs. Not today, and not next month either.

This one is not us dragging our feet. Google moved Docs onto a canvas renderer a few years ago, which means the text on your screen is drawn as a picture instead of laid out as something a browser can point at. There is no word there for an extension to underline and nothing for it to swap out. We block Docs outright instead of attaching to it and failing in some confusing half-way state that you would then have to diagnose.

A few other extensions do reach Docs, through a different Google framework built for exactly this, and we have not done that work yet. It is on the list. Until it ships the answer is no, and you should hear that from us now rather than discover it yourself at four in the afternoon with something due.

What it can and cannot see

Something that reads everything you type owes you a straight answer about what it touches. The text in the box you are working in goes to WriteHuman over HTTPS to be checked or rewritten. That is the entire list.

It cannot read your password, and not because we keep a careful list of fields to stay away from. It only ever attaches to multi-line boxes and rich editors. Passwords, card numbers and search bars are single-line inputs, so they were never in scope to begin with, which is a much better guarantee than a list somebody has to remember to maintain.

You can also turn it off completely, turn it off for one site, or pause it for a stretch, which is what you want when you are pasting code into a text box and could not care less what a grammar checker makes of it. Words you add to your dictionary, the names and jargon and deliberate misspellings, stay on your own machine and ride Chrome's own sync over to your other ones. They never reach us at all.

What it does not do yet

It is a first release and there are real gaps, so here they are rather than buried in a support article six months from now.

There is no right-click option and no keyboard shortcut, so you open it from the floating button or from the toolbar. The website gives you tone settings and several versions of every rewrite, while the extension keeps one setting and hands back one version. That was a deliberate decision about how much machinery belongs in a panel docked to the side of somebody's inbox, and it is the one we are most likely to revisit. There is no detector built in either, so the AI detector on the site is still where you go to check a piece of writing on its own terms. The checking is English only for now, although it works out what language you are in and stays quiet elsewhere instead of confidently underlining perfectly good French. And it is a Chromium extension, so Chrome, Edge, Brave and Arc, with nothing for Safari or Firefox.

That paragraph is more or less the roadmap, roughly in order.

Getting it

It is on the Chrome Web Store. Add it, pin it so the icon stays where you can find it, and carry on writing, because there is nothing to set up. It starts working in the tabs you already have open, so you do not need to go around reloading everything first. If you already have a WriteHuman account it picks up your plan and your allowance the first time you sign in, and there is no key to generate or token to paste anywhere.

Why we built it

The humanizer started on the website. Then it went to iPhone and Android, then behind an API, then inside AI assistants. Every one of those was the same argument made again in a different place: the tool should be where the writing is, not somewhere you have to travel to and come back from.

The browser was the last big gap and easily the strangest one, because the browser is where nearly all of this writing actually happens. Nobody drafts in a humanizer. People draft in Gmail, in Notion, in a CMS, in a Slack message they have already rewritten twice and will rewrite once more before they send it. Asking all of them to leave, paste, wait, copy and come back was a toll we had somehow decided to charge on our own product.

There is a version of this that only humanizes, and it would have been considerably faster to ship. We already had the engine, the extension around it is mostly plumbing, and the category is full of tools that do that one thing and stop. What killed it is that we use this all day ourselves, and every time somebody tested a humanize-only build they quietly installed a second extension next to it to catch their typos. That is less a product decision than an admission.

The grammar checking came along for a related reason. Once we were living inside the text box anyway, catching a misspelling was a short step from there, and leaving it out to keep the product tidy would only have meant you still needed a second extension for the other half of the same job. We would rather you needed one.

It is out now, and it is a first version. If you install it and something is broken or missing, tell us. At this stage a handful of people telling us what is wrong moves it faster than anything else we could possibly do on our own.