Promptwatch Logo

ChatGPT agent

Agent that can use its own browser to perform tasks for user.
OpenAI
AI AssistantAgentUser Initiated

What is ChatGPT agent?

ChatGPT agent is OpenAI's task-performing agent with its own virtual browser. It visits a site because a ChatGPT user assigned a task, such as finding information or completing steps on the web. This is delegated browser activity, not a background crawl.

According to the OpenAI help page, the agent uses screenshots of its browser window to understand pages. It can navigate, click buttons, and fill forms as the task proceeds. A single run can create a sequence of reads and actions that resembles an interactive browser session.

When a login is needed, the agent can pause and ask the user to take over the virtual browser. OpenAI says screenshots are not captured while the user is controlling it, which lets the user enter a password or other sensitive value. Control can then return to the agent so it can try to continue the workflow.

OpenAI also documents confirmations for high-impact actions, prompt-injection monitoring, and supervised watch mode on some sites. Those product safeguards do not replace a website's own authorization or transaction checks. A site should still decide what the signed-in user may read, change, submit, or purchase.

ChatGPT agent is not the OpenAI crawler used to build search results, and its record gives it no model-training role. OAI-SearchBot and GPTBot are separate identities for those functions. Blocking this agent can stop delegated tasks on the site, but it does not by itself remove the site from ChatGPT search or opt content out of training.

Cloudflare's bot directory lists no stable user-agent pattern for ChatGPT agent. It identifies the agent through Web Bot Auth and gives https://chatgpt.com/.well-known/http-message-signatures-directory as the public key directory. Signature validation is the available way to distinguish covered OpenAI agent requests from a client that merely claims an OpenAI name.

Robots.txt behavior is unknown in this bot, while Cloudflare's bot directory marks followsRobotsTxt as false. Without a stable token, a ChatGPT agent robots.txt block would be guesswork. Sites that want to permit or deny this agent should use signed identity at the edge and ordinary application controls inside authenticated sessions.

Relevant for AI searchChatGPT

Is ChatGPT agent relevant for AI search?

Yes. ChatGPT agent feeds ChatGPT, so the pages it can reach shape what those AI products say about you.

ChatGPT agent crawls and indexes pages so the AI search or assistant behind it can retrieve them at answer time. A page it has never fetched cannot be quoted, summarized, or linked in that product's answers, so most sites keep it allowed to stay citable. Blocking it removes your pages from that AI surface and hands those citations to competitors.

How to handle ChatGPT agent

Handle ChatGPT agent as an interactive browser acting for a user. Public pages can use the same policy as other browsers. Login, account changes, form submissions, and purchases should retain normal permissions and any confirmation step your application requires.

Do not create a robots.txt group from a guessed user agent. No stable token is recorded, and Cloudflare says the agent does not follow robots.txt. Validate Web Bot Auth signatures with https://chatgpt.com/.well-known/http-message-signatures-directory, then use the verified identity in a CDN, WAF, or origin rule if you need a specific allow or block.

A valid OpenAI signature authenticates the agent request, not the end user's authority. Keep session checks and action-level authorization in place. If you block the agent, manage OAI-SearchBot and GPTBot separately because this decision does not control search indexing or training.

Examples

  • A user asks ChatGPT agent to find a suitable restaurant, and the site sees the browser move through search results and reservation details for that task.
  • A workflow reaches a login screen, pauses while the user takes over to enter credentials, then resumes under the site's normal session permissions.
  • A publisher blocks verified ChatGPT agent actions at its WAF while leaving its separate OAI-SearchBot policy unchanged.

Frequently asked questions about ChatGPT agent

Learn about AI visibility monitoring and how Promptwatch helps your brand succeed in AI search.

A ChatGPT user has assigned a task that requires the agent's virtual browser. The visit is part of that task, not an autonomous indexing run.

Related bots

ChatGPT-User

Handles user-initiated requests in ChatGPT, accessing external content to provide real-time information; not used for automated crawling or AI training.

AI Assistant

GPTBot

Crawls web content to improve OpenAI's generative AI models and ChatGPT; respects 'robots.txt' directives to exclude sites from training data.

AI Crawler

OAI-AdsBot

Validates the safety of web pages submitted as ads on ChatGPT; data collected is not used to train generative AI foundation models.

Advertising

OAI-SearchBot

Indexes websites for inclusion in ChatGPT's search results; does not crawl content for AI model training.

AI Assistant

Amazon Kendra

Amazon Kendra is a managed information retrieval and intelligent search service that uses natural language processing and advanced deep learning model.

AI Assistant

Be the brand AI recommends

Monitor your brand's visibility across ChatGPT, Claude, Perplexity, and Gemini. Get actionable insights and create content that gets cited by AI search engines.

Promptwatch Dashboard