AI assistant detection

What is ShieldLabs AI assistant detection?

ShieldLabs AI assistant detection recognizes supported assistants retrieving information from your website. It returns the AI Bot detection, the provider when available and optional metadata about the client name and identity verification. AI assistants use large language models (LLMs) to answer questions, compare information and summarize what they find. To prepare a response, an assistant can request pages from your website. ChatGPT, Claude and Perplexity are examples; the supported client list shows the requests ShieldLabs recognizes. Use the assistant name, provider and available verification results with the Risk Score to distinguish recognized AI traffic from risky automation. For agents that navigate pages and interact with forms, read ShieldLabs AI agent detection.

How AI assistant detection and verification work

ShieldLabs combines recognized client identity with the network evidence available for the identification. Supported traffic can identify itself by its request headers; provider address information can establish additional attribution. Recognition and verification answer different questions:

Result	What it establishes
A recognized identity claim	The request presents a known bot or assistant identity
Verified provider attribution	Available evidence confirms the provider
AI bot detection	ShieldLabs classifies the identification as recognized AI-bot traffic

A recognized name is not proof that a request came from that provider. Check client_identity.verified when your application requires confirmed attribution. For GPTBot and ChatGPT-User, supported verification includes the provider’s published IP ranges. AI assistant detection analyzes activity received through your ShieldLabs integration. Make sure the assistant reaches that integration so its request appears in your results.

Event payload

The following illustrative excerpt shows a recognized ChatGPT-User identification inside a scored webhook’s data object:

{
  "identification_id": "13f84f05-7c2a-4e9b-9f1d-2a6b8c0e4d11",
  "risk_score": 0,
  "signals": [{ "name": "ai_bot", "weight": 0 }],
  "detection_flags": { "ai_bot": true },
  "ai_bot_owner": "OpenAI"
}

These are JSON fields inside data of an identification.scored event. The excerpt omits other fields and false detection flags for readability. Read the result after verifying the webhook signature:

Field inside `data`	Type	Meaning
`detection_flags.ai_bot`	boolean	Recognized AI-bot traffic, including supported assistant fetchers and AI crawlers
`ai_bot_owner`	string, optional	Provider of the active AI Bot detection; omitted when the name is unavailable
`signals`	array	Includes `ai_bot` with weight 0 when the detection is active
`risk_score`	integer	Recognized good-bot traffic scores 0
`client_identity`	object, optional	Identity claims, verified attribution and assessment results

Read the assistant name

When available, client_identity.claims identifies the recognized profile, its provider, the agent name and its purpose. This shortened example shows the claim for ChatGPT-User:

{
  "client_identity": {
    "claims": [{
      "profile_id": "openai-chatgpt-user",
      "provider_id": "openai",
      "provider_name": "OpenAI",
      "agent_name": "ChatGPT-User",
      "client_kind": "fetcher",
      "purpose": "user_fetch",
      "source": "http_ua"
    }]
  }
}

user_fetch describes retrieval for a user request. training describes a content-collection crawler. Search-oriented profiles use search and can appear under the Search bot detection. The claims list records presented identities. It does not replace the separate verified attribution list.

Interpret identity assessments

client_identity.assessments supplies context for recognized claims and verification attempts. Interpret each assessment with its profile and the available attribution:

`status`	Meaning
`verified`	A verification check succeeded within its stated scope
`unverified`	A known identity was recognized without that assessment establishing verification
`unknown`	The available input did not establish a verification result
`conflict`	The identity claims or evidence disagree

A provider can have verified attribution while a name claim remains unverified. Read the scope in client_identity.verified; do not infer a single verification state from one assessment. Missing client_identity means this metadata was not supplied. In a scored webhook, the array is at data.client_identity.assessments. A candidate_profile_id links an entry to a claim’s profile_id; an entry without it reports a general assessment. Empty or unavailable assessments do not establish a human visitor or a verified client. See Identity assessment for the field definitions.

Supported AI bots and assistants

ShieldLabs supports the following AI bots and assistants:

Name	Provider	Activity
GPTBot	OpenAI	AI content crawler
ChatGPT-User	OpenAI	Retrieval for a user request
ClaudeBot	Anthropic	AI content crawler
Claude-User	Anthropic	Retrieval for a user request
Perplexity-User	Perplexity	Retrieval for a user request
MistralAI-User	Mistral AI	Retrieval for a user request
DuckAssistBot	DuckDuckGo	Retrieval for AI-assisted answers

Search-oriented AI crawlers, including OAI-SearchBot, Claude-SearchBot and PerplexityBot, appear in the bot directory. The directory separates product detections from the purpose suggested by a crawler’s name.

Using AI assistant detection results

Use the AI bot flag to separate recognized AI traffic from human activity and other automation. Use verified attribution when access depends on a particular provider.

Result	How to use it
`ai_bot: true`, provider supplied	Attribute recognized AI traffic to that provider
Verified provider attribution	Apply provider-specific access conditions when appropriate
A known name without verification	Keep recognition separate from confirmed identity
Conflicting identity evidence	Review the conflict before granting provider-specific access
No recognized AI bot	Read the other risk signals and Risk Score for this identification

The dashboard groups recognized AI bots under Good bot. Their Risk Score is always 0. Use identity verification and account context when granting access to protected content or actions.

View AI traffic in the dashboard

Open Analytics and inspect an identification’s signals. AI Bot displays the provider when available; Good bot combines AI bots, AI browsers and search bots in aggregate views. For backend automation, read ai_bot and ai_bot_owner from the scored webhook. History can include optional client_identity, but it does not return those two fields. See bot results in webhooks and History.

ShieldLabs

Core concepts

Detection

Bot detection

AI detection

Analytics dashboard

Security and privacy

Resources

Legal

AI assistant detection

What is ShieldLabs AI assistant detection?

How AI assistant detection and verification work

Event payload

Read the assistant name

Interpret identity assessments

Supported AI bots and assistants

Using AI assistant detection results

View AI traffic in the dashboard

Next steps

AI agent detection

Bot directory

​What is ShieldLabs AI assistant detection?

​How AI assistant detection and verification work

​Event payload

​Read the assistant name

​Interpret identity assessments

​Supported AI bots and assistants

​Using AI assistant detection results

​View AI traffic in the dashboard

​Next steps

AI agent detection

Bot directory

What is ShieldLabs AI assistant detection?

How AI assistant detection and verification work

Event payload

Read the assistant name

Interpret identity assessments

Supported AI bots and assistants

Using AI assistant detection results

View AI traffic in the dashboard

Next steps