AI reply Chrome extensions: the three shapes

Every AI reply extension does roughly the same thing — read the message you are answering, draft a response. What separates them is where the tool lives while you use it. There are three answers to that, and the one you pick decides whether the extension is still installed in a month.

By Rex (Kuan Hsiung Chen) · Updated 30 August 2026

What an AI reply extension actually is

A browser extension that takes the thread in front of you as context and writes a reply into the field you were already typing in. The distinction from a general AI writing assistant is the input, not the model: with a writing assistant you paste the incoming message in and copy the answer out; with a reply extension the incoming message is the input, and you never move text by hand.

That sounds like a small difference and it is the entire difference. Copying a message into a chat window, prompting, reading, copying back and reformatting takes about ninety seconds. A reply that took ninety seconds to produce is a reply you could have typed. The only versions of this that survive contact with a real inbox are the ones that remove the copying.

Shape one: the sidebar

A panel docked to the side of the page, usually with a chat interface in it. This is the most common shape because it is the easiest to build — the extension owns its own surface and never has to understand the page it is sitting next to.

The cost is that you are back to moving text. The sidebar does not know which field your cursor was in, so the output arrives in the panel and you copy it across. It also permanently occupies horizontal space, which on a laptop screen means the thread you are replying to gets narrower every time you open it. Sidebars are good when the task is genuinely a conversation with the AI — researching, summarising a long thread, asking follow-up questions. They are a poor fit for "answer this email and move on".

Shape two: the floating button

An icon that appears over the compose box, or follows your selection around. Better than a sidebar in that it knows roughly where you are working, so it can usually insert its output in the right place.

The cost is that it is always visible, including every single time you did not want it. A button that overlaps the send control, covers the formatting toolbar, or pops up whenever you select a word to correct a typo becomes an irritant faster than it becomes a habit. This is the shape with the highest uninstall rate for a reason that has nothing to do with the quality of the AI: the tool interrupts the ninety-five per cent of typing that needed no help in order to be available for the five per cent that did.

Shape three: the inline command

Nothing on the page at all until you press a key. The extension opens at the caret, in whatever field you were already in, and closes again when you are done.

The cost here is real too, and worth stating: there is nothing to discover. A tool with no visible surface has to be remembered, which means the first week is worse than a floating button's first week — you will forget it exists, and type the reply out by hand. What it buys is that once the keystroke is in your fingers, the tool disappears into the work. Nothing covers the send button, nothing narrows the thread, and nothing appears while you are typing the messages that needed no help.

Six things worth checking before you install one

In rough order of how often they are the thing that actually ruins it:

1. Does it read the thread, or just your prompt? Some extensions labelled "AI reply" are a generic prompt box in a costume — they generate from what you type and never look at the message you are answering. Test it on a real thread before you rely on it: ask for a reply without describing the incoming message and see whether the output makes sense.

2. Where does it work? An extension built as a Gmail integration works in Gmail. One built on standard inputs, textareas, and contenteditable fields works in Gmail, Outlook on the web, LinkedIn, support consoles, forums and most web forms. If your day is spread across more than one product, this is the difference between a tool and a tab.

3. Can you steer the tone without rewriting the prompt? The reply is rarely wrong in content and often wrong in register — too formal, too eager, too long, or agreeing to something you meant to decline. Switching tone should be one control, not a re-prompt.

4. What is the pricing model? Nearly every one of these is a monthly subscription, because generation costs the developer an API call each time. That is honest pricing for heavy use and poor value for ten replies a week. "Free to install" is not "free to use" — read the pricing page, not the install button.

5. What happens to the message text? Generating a reply means sending the relevant message to an AI provider; that is unavoidable and not the thing to worry about. The questions worth asking are whether message content also ends up in analytics, whether drafts are retained after the reply is generated, and whether the extension asks for permission on all sites when it only needs some.

6. Does it also do templates? Not every reply should be generated. Refund policies, booking confirmations, addresses and legal boilerplate need to be identical every time, and paying an AI to re-derive fixed wording is slower, costlier and less reliable than replaying it. A tool that only generates will quietly push you into generating things you should have saved.

Where Keyjot sits

Keyjot is the third shape. Press ⌘ ; on macOS or Ctrl ; on Windows and Linux, and a command window opens beside the caret in the field you were already typing in. Nothing is on the page before that, and nothing stays after.

From that window, number keys 1–9 insert saved templates, and the AI option reads the message you are replying to, takes one line of intent from you, and drafts the reply — with tone switchable between formal, friendly, short and decline. Both jobs sit behind one keystroke, which is the answer to point six above: the fixed wording stays fixed and free, and generation is there for the replies that are genuinely different each time.

It works on standard inputs, textareas and rich-text fields rather than as a single-product integration, so Gmail, Outlook on the web, LinkedIn, support consoles and most web forms are all the same surface to it. Pricing is a one-time $3.99 access pass including 100 AI reply tokens; templates and hotkeys are unlimited and never consume tokens, so they keep working whether or not you have any left.

The honest caveat is the one in shape three: for the first few days you will forget it is there. If you want something that reminds you it exists, a floating-button tool will suit you better.