Lessons 77.1 through 77.3 put ideas into the world. This lesson listens to what comes back. Comments are not a chore beneath creation — they are the cheapest research, objection-handling, and relationship system a small builder has. Every reply either earns the next viewer or quietly teaches the feed that your post leads nowhere.

Listening is a loop, not inbox zero

Social listening means systematically reading what people ask, doubt, correct, and request around your posts and your topic — then feeding it back into product, proof, and content. The loop has five steps:

• Collect: pull ten to twenty recent comments, quotes, and DMs across your primary platform plus one discovery surface (search, Reddit, TikTok comments on nearby topics).
• Categorize: sort each into recurring question, objection, correction, request, praise, or spam/noise. Most value lives in the first three.
• Answer: reply genuinely to the ones that help future readers, in public, in your voice.
• Library: save reusable answers as a response library — approved wording for price, timing, limits, method, and links.
• Escalate: mark what needs a private or human path — angry customers, privacy, safety, legal claims, misinformation about your product — and move it off the public thread fast.

A research publication learns quickly this way. "Which filings does this cover?" asked four times is a content brief (a coverage page). "Does this invent quotes?" asked twice is an objection to answer with the verification demo, then to pin. "You missed footnote 14" is a correction to thank, fix, and credit. Praise is fuel, not signal — log it, do not chase it.

What AI may do, and what only you may do

Connect to Parts VI and VIII: models are excellent sorters and drafters, terrible judges. Give AI the mechanical half:

• Group fifty comments into themes and count repeats.
• Draft two reply options from approved facts only — your promise wording, your evidence links, your response library.
• Surface unanswered themes, sentiment shifts, and questions your library cannot answer.

Keep for a human: the actual reply decision, tone, corrections, and every sensitive case. Never let a model invent availability, pricing, coverage, or a customer story to sound helpful. Never let it argue with a critic, dismiss a correction, or promise a fix the team has not agreed to. A fluent wrong reply costs more than a slow right one — screenshots outlive feeds.

A safe pattern mirrors Class 70: original comment preserved → AI grouping and draft from approved sources → human review and edit → public reply with link back to evidence → log entry with outcome. If the draft adds a fact you cannot point to in your library, delete the sentence, not just the word.

The response library and the escalation rule

Build the library before you need it. Five cards cover most threads:

• What it is / who it fits: one sentence plus boundary ("fits solo analysts; not built for real-time terminal data").
• Proof: link to the sample, demo timestamp, or booking page — never "trust me."
• Price/timing/condition: exact words, updated when reality changes.
• Limits: what you do not do, stated plainly. Limits stated early prevent the angriest threads.
• Next step: the one action per topic, matched to Lesson 77.2's CTA discipline.

Escalation is the other half of professionalism. Move to DM, email, or a human owner when you see: personal data in a comment, a charge or order dispute, a safety or harassment thread, a claim that your product does X when it does not, or any reply where a public back-and-forth would humiliate someone. Reply publicly once — "sending you a DM now so we can fix this with your details" — then go private and log the resolution. Speed matters less than visible responsibility.

Exercise: ten comments into one decision

Create SOCIAL-LISTENING-LOG.md. Work from real comments on your posts or, if you are new, from comments on three nearby creators plus your own first replies.

# SOCIAL-LISTENING-LOG.md — Week of [date], platforms: ___

## Collected (10)
1. "[quote/paraphrase]" — source/link — category: Q / objection / correction / request / praise / noise
2. ... (through 10)

## Answered (3 genuine public replies)
- Comment #__ → reply summary + evidence linked:
- Comment #__ → reply summary + evidence linked:
- Comment #__ → reply summary + evidence linked:

## Library update
- New / edited card: [title + approved wording + link]

## Content brief (1 from the log)
- Question/objection → asset idea → format → platform → promise:

## Escalation (1 marked)
- Comment #__ → why private/human → action taken → follow-up date:

Worked mini-example — tailor's rescue video: ten comments include "price for this?" ×3 (library card + carousel with ranges), "do you do leather?" (objection/boundary → pinned reply + highlight), "that stitch will fail" (correction → thank, close-up follow-up video = content brief), one angry pickup-time complaint (escalation → DM, remake log). One week, one follow-up asset, one calmer critic.

Finish line: a SOCIAL-LISTENING-LOG.md with ten categorized comments, three genuine public answers, one library update, one content brief, and one marked escalation.

Verify quickly: read your three public replies as a stranger would. Do they each teach something a future viewer needs, with proof linked? If they only say "thanks!" or "DM us," they are activity, not distribution.

Common failure mode: automating replies end to end — identical "Great question! Link in bio!" under every comment. Feeds read it as spam, humans read it as indifference, and the learning that should have become your next video evaporates.

Check your understanding

1. What are the five steps of the listening loop, and what does each produce? 2. What may AI do with comments, and what must a human keep? 3. When must a thread move private, and what should the last public line say?

Next

You can hear the audience. Lesson 77.5 removes the mythology around why some posts travel — recommendation systems, satisfaction signals, and the controllable choices that matter more than posting-time superstitions.