SOCIAL

Reply to comments without sounding like a brand

Use when a post has picked up comments and you need replies that continue the conversation rather than thank people.

comment-reply-drafting.md

Download .md

You are replying to comments on a social post as {{PERSONA}} - a real person with a job, not a brand account.

Original post:
"""
{{POST_TEXT}}
"""

Comments:
"""
{{COMMENTS}}
"""

For each comment, output: Comment (first 8 words) | Type | Reply | Reply length.

Classify each comment as one of: genuine question, disagreement worth engaging, agreement with something to add, self-promotion, low-effort praise, hostile, off-topic.

Reply rules by type:
- Genuine question: answer it directly in the first sentence. Add one specific detail. Under 45 words.
- Disagreement worth engaging: concede the strongest part of their point before responding. Never win by volume. Under 60 words.
- Agreement with something to add: add a new fact or example, do not just say thanks. Under 35 words.
- Low-effort praise: a short human reply, under 12 words, or mark SKIP.
- Self-promotion, hostile, off-topic: mark SKIP or IGNORE and say which.

Across all replies:
- Never open two replies with the same word.
- No "Great point!", no "Thanks for sharing", no "Couldn't agree more", no emoji unless the comment used one first.
- Do not promise features, dates, refunds, or results. If a reply would need that, mark ESCALATE and say what to check.
- No em dashes.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{PERSONA}}
{{POST_TEXT}}
{{COMMENTS}}

Getting a better result

The SKIP and ESCALATE marks are the point - most comment threads need fewer replies, not more.
Paste the comments with usernames removed if you are working in a shared doc.
Give it your own last five replies as a voice sample; generic warmth is the default failure mode here.

Questions about this prompt

When should I use this rather than replying by hand as comments arrive?

When a post has more comments than you can answer well. Replying as they land gives the loudest commenter the most of your attention and produces a column of Great point. This classifies every comment first, then decides which deserve a reply at all and how long each one may be.

What do I need before running it?

The original post, the comments as written, and a {{PERSONA}} described as a person with a job rather than a brand account. The tips suggest pasting your own last five replies as a voice sample, since generic warmth is the default failure here. Strip usernames if the drafts will sit in a shared doc.

What comes back, and which part saves me time?

A table classifying each comment and drafting a reply inside a word cap set by type, with SKIP, IGNORE and ESCALATE marks. The marks are the useful output: most threads need fewer replies rather than more, and ESCALATE catches the ones where a reply would promise a date, a refund or a feature.

What is the mistake?

Posting the disagreement replies without reading them properly. Conceding the strongest part of someone's point first is right, but the concession then stands in public under your name, so it has to be one you actually hold. Anything marked ESCALATE goes to whoever owns that answer instead.