Automate TikTok Comment Collection for Sentiment Analysis

A product-launch video gets most of its comments in the first 48 hours, then keeps collecting a slower trickle for weeks. Reading the comments once, the day after posting, catches the initial reaction and misses everything after — the comment that shows up a week later after someone actually tried the product, the complaint that starts spreading once enough people hit the same problem. Catching that means pulling the same video's comments repeatedly, not once.

If you're an agent (or building one) reading this rather than a human, the full machine-readable schema for every endpoint below lives at mindcase.co/skills.md.

What does automating comment collection actually mean?

Three things have to happen on a schedule, not just once: the pull, the storage, and the comparison.

• The pull. The same video (or the same creator's recent videos), checked the same way, on a fixed cadence — daily makes sense for the first couple of weeks after a launch, when comment volume is still changing fast.
• The storage. Each day's comments need to land somewhere that keeps yesterday's alongside today's, not overwrite them. A flat file, a spreadsheet, or a database table all work — the point is that day 14 can still be compared against day 1.
• The comparison. The value isn't "here are today's comments," it's "how many new ones came in, and did the tone of what people are saying shift" — and that only means something once there's a prior day to diff against.

How do you pull comments for a video?

TikTok Comments API takes a single video URL, so tracking a video means calling it on the same URL repeatedly rather than looping a list — this is a one-video-per-call endpoint, not a batch one.

curl -X POST "https://api.mindcase.co/v1/data/tiktok/comments/run?wait=true" \
-H "Authorization: Bearer $MINDCASE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "params": {
    "postURLs": "https://www.tiktok.com/@yourbrand/video/7123456789012345678",
    "commentsPerPost": 200
  }
}'

# $0.0005 per comment returned

Each row back includes commentId, comment, likes, and postedDate — the fields that matter for tracking, since a comment only becomes a useful signal once it's tied to when it landed and how much it resonated.

How do you store and diff comments day over day?

This is the part a one-off read skips entirely: appending each day's pull to a running history, then comparing today's set against yesterday's by commentId.

Any commentId in today's pull that wasn't in yesterday's is a new comment — that's the count that actually matters for "is this still getting attention," more than the running total. For comments that already existed, compare likes day over day: a comment gaining likes three days after posting is still resonating; one that stalled at its Day 1 count has stopped mattering to new viewers.

An illustrative two-day snapshot for one video makes the shape of this concrete:

Two things are visible here that a single read on Day 2 alone would blur together: c_9182 is still gaining traction (12 → 41 likes), and c_9204 is a genuinely new comment, not a continuation of an existing thread. A tracker that only stores the latest snapshot loses both signals — it would show 2 comments on Day 2 with no way to tell which one is new or which one is picking up steam.

How do you schedule the pull?

Nothing about the API itself is schedule-aware — it answers one request at a time. The loop above runs wherever you already run scheduled code: a cron job, a serverless function on a timer, or whatever workflow tool your team already uses. Daily makes sense in the first couple of weeks after a launch; once comment volume on a video has settled to near zero, there's little left to track.

What can't you get from Mindcase?

Two cases.

A built-in scheduler or a place to store history. The API answers the request you send it — it doesn't remember yesterday's pull or run itself on a timer. Both are yours to build, on whatever schedule and storage you're already using.

Comments from before you started tracking. Every request reads the video's current comment thread at call time. There's no historical archive of a video's comments as they looked last month — tracking only covers the window from when you begin pulling forward.

Which endpoint should you use for which job?

FAQ