Answers
Can ChatGPT or Claude watch a Douyin video? (2026)
Short answer
Not from a link: the model gets a page it cannot play, and there is no Douyin transcript endpoint here. What it can read is the text around the video. Through okdata it can fetch the caption, hashtags, author and counts for $0.002 and the comments at $0.002 a page, so one video with two comment pages costs $0.006.
People paste a Douyin link into a chat and ask for a summary. What comes back is often a refusal or, worse, a confident summary guessed from the link text. The model never saw the video. A Douyin page is an app shell that loads the player with scripts, and a tool that fetches the page usually gets only that shell. On the API side, both Anthropic's and OpenAI's vision docs list image formats (JPEG, PNG, WebP, GIF) as visual input, and Claude's notes that only the first frame of an animation is used. Whether a chat app accepts an uploaded video file depends on the product and plan and changes often, so check its help pages; a Douyin link is not a file either way.
What a model can read about a Douyin video
| What | Where it comes from | Field |
|---|---|---|
| Caption and hashtags | Video detail | text, hashtags |
| Author | Video detail | author.name, author.id |
| Likes, comments, shares, saves | Video detail | metrics.* (plays are not public) |
| Posted, length, soundtrack | Video detail | published_at, ext.duration_ms and ext.music with view=full |
| What viewers say | Video comments | text, likes, ip_location, hottest first |
| The pictures of a photo post (图文) | Video detail | images, which a vision model can look at |
That is often enough for the real question behind "watch this": what is the video about, how did it do, and what do people think of it. Comments carry much of the answer, because Douyin captions are short.
What is missing: the spoken words, on-screen text, and anything visual in a video post. okdata has no Douyin transcript endpoint and does not return Douyin video files.
Doing it through MCP
Connect the okdata MCP server (https://api.okdata.online/mcp) to Claude, ChatGPT or another MCP client and ask in plain words: "Look up this Douyin video and its top comments, then tell me what the audience liked and complained about." The model calls video detail and comments itself.
| Request | Calls | Cost |
|---|---|---|
| Video detail only | 1 | $0.002 |
| Detail plus two comment pages | 3 | $0.006 |
| 20 videos, same depth | 60 | $0.12 |
Example: build the context yourself
import os, requests
API, H = "https://api.okdata.online", {"Authorization": f"Bearer {os.environ['OKDATA_API_KEY']}"}
link = "https://v.douyin.com/iRk7pQe/"
v = requests.get(API + "/v1/douyin/videos/get", headers=H, timeout=60,
params={"url": link, "view": "full"}).json()["data"]["item"]
c = requests.get(API + "/v1/douyin/videos/comments/list", headers=H, timeout=60,
params={"id": v["id"], "fields": "text,likes"}).json()["data"]["items"]
context = (f"Caption: {v['text']}\nHashtags: {v['hashtags']}\nAuthor: {v['author']['name']}\n"
f"Likes {v['metrics']['likes']}, comments {v['metrics']['comments']}, saves {v['metrics']['favorites']}\n"
"Top comments:\n" + "\n".join(f"- ({x['likes']}) {x['text']}" for x in c))
print(context) # paste into any model, or send it with your own API call
When you need the spoken words
- Ask for them. Creators and agencies usually have the script.
- If you have the right to the video file, for example your own account's videos, transcribe it locally with an open-source speech model such as Whisper and give the model the text.
- On YouTube the transcript is available directly; see summarising a YouTube video with Claude.
For wiring Douyin data into either assistant, see Claude and Douyin data through MCP and connecting ChatGPT to Douyin data.
Sources
- Claude vision docs: supported formats are JPEG, PNG, GIF and WebP images; animations use the first frame
- OpenAI images and vision guide: image input file types
- Whisper: open-source speech recognition
Where people asked this (read 2026-11-12)
Updated