# How do I find out what viewers complain about in YouTube comments with AI?

Connect SocialToAI to your AI client over MCP, then ask in plain words. In this recorded run an AI client answered “What do viewers complain about in the comments of new Kindle reviews on YouTube?” with 1 request on YouTube: 184 items read in 28.6 seconds for 50 credits ($0.17). Below are the calls it made, the answer with its sources, and the prompt and API calls to run it yourself. Recorded 2026-10-18.

- Page: https://socialtoai.com/docs/guides/analyse-youtube-comments/
- Last updated: 2026-10-18

## The question

> What do viewers complain about in the comments of new Kindle reviews on YouTube?

Review videos collect hundreds of comments, and YouTube shows the top ones first, which are often jokes or praise. The complaints are spread through the rest. Reading what this run returned by hand takes about 19 minutes (4,561 words at 240 a minute).

## What the AI did

It chose the tools itself: 1 request on YouTube. Each row is one tool call as the client showed it, with what it sent, how much came back and what it cost.

| Platform | Tool | Sent | Requests | Returned | Credits |
| --- | --- | --- | --- | --- | --- |
| YouTube | [`comment_insights`](https://socialtoai.com/tools/comment-insights.md) | `targets: ["https://www.youtube.com/watch?v=f13Y9H7zooU","https://www.youtube.com/watch?v=GW-FpfCx0Js"] · focus: "what people think of the Kindle" · scale: "small"` | 1 | 184 comments | 50 |
| **Total** |  |  | 1 | 184 items | **50** |

## The answer

What the client answered, from what it had read. The line under each finding is a post it read, or a figure from those posts.

Recorded 2026-10-18

I read 184 comments under 2 review videos; 77 of the 181 that could be judged complain about something. The complaints that come up most:

1. **Prices that feel like a rip-off** (13%, 24 comments): [“this new one feels like a robbery.”](https://www.youtube.com/watch?v=GW-FpfCx0Js)
2. **No warm light** (10%, 19 comments): [“No warm light is a crime”](https://www.youtube.com/watch?v=GW-FpfCx0Js)
3. **Amazon's business practices** (7%, 12 comments): [“they locked their ecosystem and stole my books”](https://www.youtube.com/watch?v=GW-FpfCx0Js)

Used 50 credits · $0.17. Every claim links to its source.

### The posts it rests on

A selection of what was read; the marked words are what a finding rests on. An item without a link says so.

- YouTube · comment: “this new one feels like a robbery.” ([Open the source](https://www.youtube.com/watch?v=GW-FpfCx0Js))
- YouTube · comment: “No warm light is a crime” ([Open the source](https://www.youtube.com/watch?v=GW-FpfCx0Js))
- YouTube · comment: “t my business the moment they pulled the rug under me, when they locked their ecosystem and stole my books.” ([Open the source](https://www.youtube.com/watch?v=GW-FpfCx0Js))

## What it cost

1 request, 50 credits in total: $0.17 at the $10 rate. Every answer of the API states the credits it charged and the balance left, so your AI can tell you what a question cost. A call that fails costs nothing.

| Pack | Credits per dollar | This answer |
| --- | --- | --- |
| $10 = 3,000 credits | 300 | $0.17 |
| $50 = 20,000 credits | 400 | $0.13 |
| $200 = 100,000 credits | 500 | $0.1 |

## Ask it yourself

Connect SocialToAI to your AI client once ([the steps for each client](https://socialtoai.com/mcp.md)), then paste the question. You do not name tools or parameters: the client picks them.

**Prompt**

```
What do viewers complain about in the comments of new Kindle reviews on YouTube?
```

The same in code: the calls this run made, as requests to the API.

**tool.comment_insights**

```
curl https://api.socialtoai.com/v1/tools/comment_insights \
  -H "Authorization: Bearer $SOCIALTOAI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "targets": ["https://www.youtube.com/watch?v=f13Y9H7zooU","https://www.youtube.com/watch?v=GW-FpfCx0Js"],
    "focus": "what people think of the Kindle",
    "scale": "small"
  }'
```

## What to try next

- Ask next: “Quote the questions people ask”
- Ask next: “Which comments want to buy?”
- Ask which comments want to buy, and what holds them back.
- Run the same on a competitor's review.
- Run it as one request with the ready-made tool [Comment analysis](https://socialtoai.com/tools/comment-insights.md): What the comments under your posts are about: the topics and their shares, the questions asked, who wants to buy and what is complained about, with the comments quoted.

- [YouTube MCP server](https://socialtoai.com/mcp/youtube.md): What your AI can read on YouTube, and how to connect it.
- [YouTube API](https://socialtoai.com/platforms/youtube.md): Every call on YouTube, with its credits and a request you can copy.