Paste the post you are replying under, then the comments, one per block with a blank line between them. Each comment is classified — a question, a shared experience, a disagreement, or someone who may be a potential customer — and each type is handled differently, because a reply that answers a question is not the same shape as one that responds to a challenge. The tag is shown so you can see what it decided, and you can change it.
There is no URL box, and that is deliberate rather than an omission. LinkedIn and Quora return a login page or a bot challenge to an anonymous request, so a fetch would produce nothing for exactly the platforms people want to use it on. Pasting works everywhere, keeps the whole thing in your browser, and means the site never makes a request on your behalf. It also makes the accuracy rule structural: one draft per comment you pasted, so a comment that is not on screen cannot produce a reply.
Using it without embarrassing yourself
-
Read every draft before you post it:
the model answers from the post and your notes. Where neither contains the answer it writes around the gap, and a confident wrong answer under your own name is worse than a slow reply. These are drafts, not an outbox.
-
The notes field is what makes replies accurate:
a comment asking how long something took cannot be answered from a post that does not say. Put the figure in the notes and the reply is specific; leave it out and the reply is vague, which is the honest alternative to inventing one.
-
Nothing is posted for you:
automated replying is against the terms of most platforms and gets accounts removed. You copy each reply and post it yourself, which is also the point at which you read it.
-
Comments are other people’s writing:
pasting them sends them to the AI provider you chose, under their terms. Worth a thought before pasting a thread from a private group or a client’s community.
Why there is no box for a link
The obvious design for a tool like this is a URL field: paste a link to the post, and let the site fetch the comments itself. That field does not exist here, and the reason is not an oversight.
LinkedIn and Quora — the two places people most want this for — hand an anonymous request a login wall or a bot challenge rather than the page itself. A server-side fetch would return nothing useful for exactly the platforms that matter most, which makes a URL box worse than no box at all: it would look like it works and quietly fail.
Pasting sidesteps that, and it buys two things beyond working reliably. The [msm_tool slug="comment-reply-assistant"] classifies and drafts entirely from what is on your screen, so the AI provider you connect sees the text directly from your browser rather than through this site. And because a reply can only be generated for a comment that was actually pasted in, a comment that never made it into the box cannot produce a hallucinated response — the accuracy limit is structural, not a promise.
How a comment gets classified, and where the guess breaks
Each pasted comment is scanned for one of four patterns, in a fixed order: disagreement first, then a question, then a possible prospect, and anything left over is treated as a shared observation. The order matters more than it looks like it should. A comment that both disagrees and asks a question — "that's not right, surely you tested this?" — gets tagged disagreement, because that check runs first and stops the search once it has an answer.
The tag decides which instruction reaches the model: a question is told to answer directly in its first sentence; a disagreement is told to concede whatever part is fair before defending the rest; a possible prospect is told to stay conversational and not pitch anything unasked. Each card shows the guess as an editable dropdown, because tone is often the only real signal separating a challenge from a question, and tone is exactly what a keyword pattern cannot read.
What the notes field actually changes
A comment asking how long a migration took cannot be answered from a post that never states a duration. Leave the notes field empty and the reply is honestly vague about it, because the model is told not to invent a figure it was not given. Put the number in notes and the same comment gets a specific answer. The notes field is not optional polish — it is the difference between a draft worth posting and one that reads as evasive.
What it will not do for you
Nothing gets posted automatically. Every draft is copied out and pasted back onto the platform by hand, partly because most platforms' terms prohibit automated replying and enforce it by removing accounts, and partly because that copy step is the last point at which a person actually reads what is about to go out under their name.
The tells it is built to avoid
Certain openers — "Absolutely," "Great question," "Couldn't agree more," "Thanks for sharing" — are excluded from every draft on purpose, along with pitching a product the comment never asked about. Avoiding the obvious tells does not make a reply good on its own; reading it before posting is what does that.
What this site sees
Nothing. The post, the comments and your notes stay in the browser tab; the AI key you supply is used to talk to your chosen provider directly, and neither the text nor the key passes through this site or gets stored anywhere on it.