NEW ChatGPT Sites Just Changed Everything (Builds Anything)

What the video teaches (main idea)

• The speaker demonstrates how ChatGPT Sites (via GPT Work) can be used to build and host a working web app in minutes, even with no coding.
• They show a real project (“Social Scrape”) that uses an agent/skill to scrape short-form social media content (e.g., Instagram/TikTok), then stores and displays videos + metadata in a hosted app for a team.

Key product features being showcased

GPT Work / @sites workflow

• You can instruct ChatGPT with prompts like “@sites build a …” and it will create a hosted site.
• The generated site includes:
  • A built-in database (to store scraped stats/transcripts/assets)
  • Hosting on the internet
  • Authentication (team members can sign in)
  • Storage (upload/hold images/videos)
  • Sharing

Deploy & access

• The generated app can be opened in a browser and accessed in Chrome (not just inside ChatGPT).
• Users may need to sign in to access the live site.

Site management / settings

In the left panel under Sites, the creator can:

• Change the domain and add custom domains
• Rename the app
• Add environment variables (e.g., API keys)
• View analytics
• Inspect database contents (video records, captions, avatars, timestamps, etc.)

The “Social Scrape” app (what it does)

Purpose

• A team-facing interface for content/ads based on scraped social video data.

Capabilities described

• Scrapes short-form platforms, including full videos from Instagram/TikTok.
• Retrieves stats (e.g., likes, comments) and the transcript.
• Stores this data in the site’s database/storage.
• Displays results in a UI resembling a short-form feed.
• Includes a download button for videos.

Example usage

• The agent can be prompted to start with specific creators’ videos, such as:
  • “Riley Brown’s best videos from 2025 that are not sponsored.”
• The creator can request additional accounts, such as:
  • “Look up Callaway on Instagram and bring in 10 of his videos.”

How iteration/editing is handled

• The initial generated UI may be “clunky” or contain mistakes (e.g., incorrect UI proportions, profile picture mismatch).
• The creator fixes issues using annotations/comments, instructing the agent to “make these changes.”
• After edits, the site is updated and can be refreshed to reflect improvements (e.g., fullscreen behavior and overlay layout like TikTok/Reels).

Technical mechanism explained (skills/plugins + external API)

• They explain that GPT Work / Codex uses skills/plugins to extend capabilities.
• The app relies on a “scrape creators” skill which calls an external API:
  1. Get an API key for “scrape creators”
  2. Provide the key to ChatGPT Work / Codex so it can create a skill
  3. The agent uses that skill to pull:
     • Video content
     • Likes/comments stats
     • Transcripts
  4. The agent places results directly into the hosted app (instead of only “presenting” data in chat)

The speaker notes this API-key handling approach is “technically not best practice,” but describes it as their workflow.

Review/guide/tutor-style takeaways emphasized

• Tutorial-style outcome: By the end, the creator claims you can open ChatGPT, use Sites, and build a team-shareable app without coding.
• Best-practice highlighted (workflow):
  • Use ChatGPT Sites as the interface layer
  • Scrape data via skills/API
  • Display it in a custom mini app for teammates
• Additional upgrade mentioned: Web MCP
  • Suggests updating/deploying so you can enable Web MCP.
  • Web MCP is described as allowing a site to expose tools that ChatGPT/Codex can discover and use directly on the live page.
  • Mentions a 10-day Web MCP challenge with cash prizes.

Main speakers/sources

• Main speaker: The video narrator/creator (no name explicitly given in the subtitles)
• Sponsored source mentioned: OpenAI
• Tool/Service sources referenced:
  • ChatGPT Work
  • Codex
  • “Scrape Creators” API / Scrape Creators skill
  • Web MCP (and “10-day Web MCP challenge”)