How does Claude show citations?

Across the consumer surfaces (claude.ai web app, Claude Desktop, mobile apps), Claude renders citations as small inline source chips or pill-shaped tags showing the domain name, attached to the portion of the response they support. Clicking a chip opens the source URL in a new browser tab. In Claude Code (the CLI agent), citations appear as URLs in the terminal output. In the Anthropic API, citations are returned as structured `web_search_result_location` objects in the response, with `url`, `title`, `encrypted_index`, and `cited_text` (up to 150 characters of the cited content) fields. Anthropic's documentation states verbatim that 'Citations are always enabled for web search,' so Messages API responses always carry citation data when the web search tool runs.

Which search engine does Claude use?

Anthropic does not say which provider serves a given answer. Its subprocessor list names three subprocessors for web search (Brave Search, TurboPuffer and Parallel Web Systems), its help center says image search is powered by Bing, and its own crawler, Claude-SearchBot, indexes pages to improve search results. Allowing Claude-SearchBot and Claude-User in robots.txt, staying indexable on Brave, and not blocking Parallel's crawler, ShapBot, are the steps the public documentation supports; none of them guarantees a citation.

Is Claude citation different from ChatGPT or Perplexity citation?

Yes. Each engine appears to use different retrieval and citation behavior, and the same domain can be cited heavily by one engine and absent on another. Perplexity is web-search-first with numbered ranked citations; ChatGPT Search uses OpenAI's web search and retrieval systems; Claude's web search tool ranges from inline chips in the consumer app to structured `web_search_result_location` objects in the API. Engine-specific measurement is safest; cross-engine aggregation should only be used after preserving per-engine breakdowns.

Which Claude surfaces have web search and citations?

As of October 2026, web search is available in the consumer apps (claude.ai web app, Claude Desktop, mobile apps for iOS and Android), in Claude Code, and to developers through Anthropic's API. Every API tool version runs on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry deployments hosted on Anthropic; Google-managed Google Cloud deployments and Foundry deployments hosted on Azure get only the basic version, and Anthropic's Amazon Bedrock documentation lists web search among unsupported features. Initial launch was 2025-03-20 (US paid users feature preview); global rollout to all plans completed 2025-05-27.

What is the difference between the web search tool versions?

Anthropic documents three API versions. `web_search_20250305` performs basic real-time web search. `web_search_20260209` adds dynamic filtering: Claude writes and runs code that filters search results before they enter the context window, keeping relevant information and discarding the rest; it is available with Claude 4.6 and later models and Claude Mythos Preview, and the API provisions the code execution it needs. `web_search_20260318` adds a response-inclusion setting that can leave search results consumed by that code out of the API response. All three return citations through the same `web_search_result_location` schema, but cited URL sets for matched queries may differ and should be tested separately.

/terms/claude-citation · 9 min read · intermediate

Claude citation

Claude citations are the source attributions Anthropic's Claude produces when its web search tool returns real-time web content for grounding. Citations appear inline as source chips in consumer surfaces (claude.ai web app, mobile, Claude Desktop) and as structured web_search_result_location fields in the Anthropic API response. Distinct measurement target from Perplexity, Microsoft Copilot, AI Overview, and AI Mode.

Citation status

ChatGPT·Perplexity·Claude0×Copilot0×Gemini0×

Last checked 2026-11-09

Claude citations are the source attributions produced by Anthropic's Claude when its web search tool returns real-time web content for grounding answers. Citations appear inline as source chips or pill-shaped tags in the consumer surfaces (claude.ai web app, Claude Desktop, mobile apps for iOS and Android), as URLs in Claude Code, and as structured web_search_result_location objects in responses from the Anthropic API¹. The feature was announced 2025-03-20 in feature preview for US paid users and rolled out globally to all Claude plans on 2025-05-27².

Claude citations are a distinct measurement target from other engines' citation surfaces: the same domain can be cited heavily by one engine while absent on another, so presence in Claude's source pool has to be measured on Claude.

At a glance


Operator	Anthropic
Index source	Not documented per answer. Anthropic's subprocessor list names three subprocessors for web search: Brave Search, TurboPuffer and Parallel Web Systems³. Image search "is powered by Bing"⁴
Search-indexing crawler	`Claude-SearchBot`, which "navigates the web to improve search result quality for users"⁵
Training crawler	`ClaudeBot`, which collects web content that "could potentially contribute to" model training
Answer-time fetcher	`Claude-User`: "When individuals ask questions to Claude, it may access websites using a Claude-User agent"
Default citation rendering	Source chips or pill-shaped tags inline in consumer surfaces; structured `web_search_result_location` objects in the Anthropic API
Surfaces	claude.ai web app, Claude Desktop (macOS / Windows), mobile (iOS / Android), Claude Code (CLI), Anthropic API and the cloud platforms listed under Status in 2026
API tool versions	`web_search_20250305` (basic), `web_search_20260209` (adds dynamic filtering), `web_search_20260318` (adds response-inclusion control)
Load-bearing fact for publishers	Claude is a chat product, not a search-first product; citation only happens when Claude retrieves web content (a search, or a fetch of a URL the user supplies), not on every answer. The API documentation says Claude answers "established facts" without searching. Anthropic publishes its crawler user agents and says they honor robots.txt, contrasting with Grok where user-agent rules are unlikely to be honored.

Status in 2026

Anthropic's documentation describes the web search tool verbatim as giving "Claude direct access to real-time web content, allowing it to answer questions with up-to-date information beyond its knowledge cutoff," with the statement that "the response includes citations for sources drawn from search results"¹. Four groups of surfaces matter for citation tracking:

Consumer apps (claude.ai web, Claude Desktop for macOS / Windows, mobile for iOS / Android): citations render as small domain-labeled chips attached to the relevant portion of the answer; clicking or tapping opens the source URL. Anthropic's help center says of web search: "Every response includes citations, so you can easily verify sources yourself"⁴. Anthropic is merging Claude's chat and Claude Cowork into one experience, "rolling out gradually, starting with Pro and Max plans on web, desktop, and mobile"; in it, "There's no web search toggle. Claude searches the web when it helps"⁶. Accounts on the earlier experience still switch web search on or off from the "+" button in the chat window⁴.
Claude Code: Anthropic's CLI agent. In Claude Code sessions where web search is available, cited sources may appear as URLs in terminal output.
Anthropic API: developers add the web search tool to a request, in one of three documented versions (the FAQ below compares them). Citations come back as web_search_result_location objects with url, title, encrypted_index, and cited_text (up to 150 characters) fields, and "Citations are always enabled for web search." The documentation also says when Claude searches: when a request depends on information that is "current, changing, or outside its training data", while "Established facts, math, science fundamentals, or coding concepts" are answered directly. Developers can steer this through the system prompt.
Cloud deployments: every tool version runs on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry hosted on Anthropic; Google-managed and Azure-hosted deployments get only the basic version. Anthropic's Amazon Bedrock documentation lists web search among unsupported features.

Detection methodology by surface:

Surface	Referrer-based detection
claude.ai web app	✅ Works; citation clicks send referrer host `claude.ai`
Claude Desktop (macOS / Windows)	❌ Does not work; citation clicks open in system browser without Referer header (standard Electron / native-app behavior)
Mobile app (iOS / Android)	⚠️ Varies; depends on whether clicks open in an in-app webview or external browser
Claude Code	❌ Terminal-rendered URLs; clicks open in system browser without Referer header
Anthropic API (used by downstream apps)	❌ Server-to-server; downstream apps using the web search tool produce no Anthropic-attributable referrer in publisher logs

Referrer filtering captures only the claude.ai web subset reliably; the other four surfaces require active probing or downstream-app cooperation.

Anthropic charges the web search tool at $10 per 1,000 searches in the API, plus standard token costs; citation fields do not count toward token usage.

How to apply

Three layers of practitioner action for inclusion in Claude's web search source pool:

Be reachable by the documented paths into Claude's search: Anthropic does not say which provider serves a given answer. It lists Brave Search, TurboPuffer and Parallel Web Systems as subprocessors for web search and runs its own search crawler, Claude-SearchBot; disabling that crawler, Anthropic says, "may reduce your site's visibility and accuracy in user search results"⁵, and disabling Claude-User stops retrieval "in response to a user query". The practitioner action is to allow Claude-SearchBot and Claude-User in robots.txt (the AI crawler bots entry has the allow-list), stay indexable on Brave (Brave Search citation covers that side), avoid blocking ShapBot, the crawler Parallel says builds its search index⁷, and keep content readable without JavaScript. Anthropic says blocking ClaudeBot, its training crawler, excludes a site's future materials from training datasets; it does not say whether that block affects search.
Optimize for citation-friendly content shape: clean answer blocks, cite-ability, single-question-single-answer paragraphs, source-supported claims, and primary-source attribution make a page easier to retrieve, summarize, and cite; Anthropic does not say whether its retrieval weights them. For apps built on the API, Anthropic's documentation says that "When displaying API outputs directly to end users, citations must be included to the original source"; for modified or combined outputs it asks developers to display citations "as appropriate based on consultation with your legal team", so a citation to your page is not guaranteed to survive downstream.
Track Claude separately, and read its search step: Claude's citation patterns can diverge from Perplexity, ChatGPT, Microsoft Copilot, and Google AI surfaces, so a blended "AI citation rate" hides per-engine signal; the attribution rate entry covers the decomposition. When probing the consumer app, expand the searches Claude shows in its answer: the query it ran and the results listed for it suggest where a page dropped out. A reworded query (a respelled term, a definition turned into a how-to) can lose the page before retrieval; a page missing from the results of a faithful query most likely dropped out at retrieval; a page listed but not cited dropped out at selection. Anthropic does not document what this panel shows, so treat it as the visible part of the search step, not necessarily the full candidate set. In server logs, keep Anthropic's three user agents apart: Claude-User for answer-time fetches, ClaudeBot for training, Claude-SearchBot for search indexing.

What to skip:

Paying for Claude-specific citation tracker tools in month 1. Anthropic documents no publisher-side dashboard; for small sites, manual probes of the consumer app may be enough at first.
Submitting your domain directly to Anthropic. Anthropic documents no submission portal; Claude reaches public content through its crawlers and its listed search subprocessors.
Treating Anthropic API tool-use citations as a publisher-visible analytics signal. Server-to-server API calls produce no identifiable referrer; the measurement gap is structural.

What remains contested or unverified

How Claude's search sources combine. Anthropic lists three subprocessors for web search and runs its own search crawler, but does not document which source serves a given query, how results are merged, or whether the mix differs by surface, plan, or query type.
How much each crawler matters for inclusion. Anthropic says disabling Claude-SearchBot "may reduce" search visibility, but not whether a page that crawler has never fetched can still be cited through a provider's index.
Whether the consumer app and the API tool search the same way. The app chooses its own query wording, and Anthropic does not document whether it shares the API tool's providers, ranking, or filtering.
Whether dynamic filtering changes which URLs get cited. The tool versions return the same citation schema; their citation outcomes for matched queries have not been systematically compared.
The boundary between web search citations and Anthropic's separate Citations feature for user-provided documents: consumer answers that draw on both may blend them without a visible distinction.

How it relates to other concepts

Parallel surface category to Perplexity citation, Microsoft Copilot citations, AI Overview citation, AI Mode, and AI dev tool citations: each is a distinct citation-surface family with its own measurement story.
Crawl-side dependency on AI crawler bots: Anthropic runs three documented user agents with different jobs, and its search also draws on listed third-party subprocessors, so no single robots.txt rule decides whether Claude's search reaches a page.
Upstream link to Brave Search citation: Brave Search is one of the subprocessors Anthropic lists for web search, so a page's standing on Brave is one documented path into Claude's pool, with an undisclosed weight.
Measurement input to attribution rate and citation-share frames: Claude is one of the five major engines those frames track, with its own per-query citation pattern.

Anthropic's web search tool documentation, platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool (the older docs.anthropic.com path redirects here). Verbatim: "The web search tool gives Claude direct access to real-time web content, allowing it to answer questions with up-to-date information beyond its knowledge cutoff. The response includes citations for sources drawn from search results." Also: "Citations are always enabled for web search." Documented versions: web_search_20250305 (basic), web_search_20260209 (dynamic filtering, "available with Claude 4.6 and later models and Claude Mythos Preview") and web_search_20260318 (response inclusion control). Pricing: $10 per 1,000 searches plus standard token costs; citation fields (cited_text, title, url) do not count toward token usage. Checked 2026-10-29. ↩ ↩²
Anthropic's launch announcement, claude.com/resources/articles/web-search (the earlier claude.com/blog/web-search address redirects here). Original feature preview announced 2025-03-20 for paid Claude users in the United States; global rollout to all Claude plans completed 2025-05-27. Verbatim launch statement on citations: "When Claude incorporates information from the web into its responses, it provides direct citations so you can easily fact check sources." ↩
Anthropic Trust Center, subprocessor list, trust.anthropic.com/subprocessors (the list renders in a browser). It names Brave Search, TurboPuffer and Parallel Web Systems as web-search subprocessors; it does not say which provider serves a given answer. Checked 2026-10-29. ↩
Anthropic Help Center, "Enable and use web search", support.claude.com/en/articles/10684626-enable-and-use-web-search. Verbatim: "Every response includes citations, so you can easily verify sources yourself." and "Image search is powered by Bing". The article names no text-search provider. Checked 2026-10-29. ↩ ↩² ↩³
Anthropic Help Center article on Anthropic's crawlers, support.claude.com/en/articles/8896518. Verbatim: ClaudeBot "helps enhance the utility and safety of our generative AI models by collecting web content that could potentially contribute to their training"; Claude-User "supports Claude AI users. When individuals ask questions to Claude, it may access websites using a Claude-User agent"; Claude-SearchBot "navigates the web to improve search result quality for users." On blocking ClaudeBot: "When a site restricts ClaudeBot access, it signals that the site's future materials should be excluded from our AI model training datasets." On disabling: Claude-User, "prevents our system from retrieving your content in response to a user query, which may reduce your site's visibility for user-directed web search"; Claude-SearchBot, "prevents our system from indexing your content for search optimization, which may reduce your site's visibility and accuracy in user search results." Checked 2026-10-29. ↩ ↩²
Anthropic Help Center, "Claude Cowork and chat are one Claude", support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude. Verbatim: "There's no web search toggle. Claude searches the web when it helps." Rollout: "rolling out gradually, starting with Pro and Max plans on web, desktop, and mobile." Checked 2026-10-29. ↩
Parallel Web Systems, "Overview of Parallel Web Systems' Bots", parallel.ai/parallel-web-systems-bots. Verbatim: "ShapBot collects publicly available web content to build and maintain Parallel's search index" and "Restricting ShapBot prevents our system from indexing your content, which may reduce your site's visibility and discoverability through Parallel's web APIs and search features." Checked 2026-10-29. ↩

Part of Citation surfaces· editorial cluster, not a semantic link

Also in this cluster: AI dev tool citations · AI Mode · AI Overview citation · Brave Search AI citation · ChatGPT search citation · +6 more

Mentioned in· auto-generated from other terms' related lists

FAQ

How does Claude show citations?: Across the consumer surfaces (claude.ai web app, Claude Desktop, mobile apps), Claude renders citations as small inline source chips or pill-shaped tags showing the domain name, attached to the portion of the response they support. Clicking a chip opens the source URL in a new browser tab. In Claude Code (the CLI agent), citations appear as URLs in the terminal output. In the Anthropic API, citations are returned as structured `web_search_result_location` objects in the response, with `url`, `title`, `encrypted_index`, and `cited_text` (up to 150 characters of the cited content) fields. Anthropic's documentation states verbatim that 'Citations are always enabled for web search,' so Messages API responses always carry citation data when the web search tool runs.
Which search engine does Claude use?: Anthropic does not say which provider serves a given answer. Its subprocessor list names three subprocessors for web search (Brave Search, TurboPuffer and Parallel Web Systems), its help center says image search is powered by Bing, and its own crawler, Claude-SearchBot, indexes pages to improve search results. Allowing Claude-SearchBot and Claude-User in robots.txt, staying indexable on Brave, and not blocking Parallel's crawler, ShapBot, are the steps the public documentation supports; none of them guarantees a citation.
Is Claude citation different from ChatGPT or Perplexity citation?: Yes. Each engine appears to use different retrieval and citation behavior, and the same domain can be cited heavily by one engine and absent on another. Perplexity is web-search-first with numbered ranked citations; ChatGPT Search uses OpenAI's web search and retrieval systems; Claude's web search tool ranges from inline chips in the consumer app to structured `web_search_result_location` objects in the API. Engine-specific measurement is safest; cross-engine aggregation should only be used after preserving per-engine breakdowns.
Which Claude surfaces have web search and citations?: As of October 2026, web search is available in the consumer apps (claude.ai web app, Claude Desktop, mobile apps for iOS and Android), in Claude Code, and to developers through Anthropic's API. Every API tool version runs on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry deployments hosted on Anthropic; Google-managed Google Cloud deployments and Foundry deployments hosted on Azure get only the basic version, and Anthropic's Amazon Bedrock documentation lists web search among unsupported features. Initial launch was 2025-03-20 (US paid users feature preview); global rollout to all plans completed 2025-05-27.
What is the difference between the web search tool versions?: Anthropic documents three API versions. `web_search_20250305` performs basic real-time web search. `web_search_20260209` adds dynamic filtering: Claude writes and runs code that filters search results before they enter the context window, keeping relevant information and discarding the rest; it is available with Claude 4.6 and later models and Claude Mythos Preview, and the API provisions the code execution it needs. `web_search_20260318` adds a response-inclusion setting that can leave search results consumed by that code out of the API response. All three return citations through the same `web_search_result_location` schema, but cited URL sets for matched queries may differ and should be tested separately.

Sources & further reading

New terms shipped that week, plus one observation from the AI-citation tracker.

More about what you'll get

Last fact-checked 2026-10-29. Spotted an error or stale claim? See editorial methodology.

Changelog (4 entries)

2026-10-29: Corrected against Anthropic's current documentation: Brave Search is one of three subprocessors Anthropic lists for web search, Claude-User covers fetches made while Claude answers a question, and the robots.txt advice now separates the search crawler (Claude-SearchBot) from the training crawler (ClaudeBot) and covers Parallel's search crawler. Added the third API tool version, current platform availability, when the API documentation says Claude searches, the merged chat and Cowork app (no web-search toggle), and a link to our dispatch on a Claude citation drop.
2026-05-28: Added an At-a-glance summary table at the top of the entry, matching the section structure used by sibling citation-surface entries. Surfaces ClaudeBot vs Claude-SearchBot vs Claude-User as three first-party documented Anthropic UAs with distinct roles. Load-bearing fact: Claude is a chat product, not search-first; citation only happens when the web search tool is invoked. Anthropic publishes first-party crawler documentation and respects robots.txt, contrasting with Grok (no contract). Metadata dates synced to 2026-05-28.
2026-05-26: Same-day revision. Dates re-verified against the Anthropic blog post (March 20 2025 + May 27 2025 both confirmed). Footnote 1 URL updated to the canonical docs.anthropic.com path. Six overconfident sentences hedged (Desktop / mobile rendering, Claude Code appearance, two-tool-version equivalence, month-1 tool-spend tradeoff, cross-engine divergence, ClaudeBot purpose disclosure). Contested bullet on ClaudeBot necessity expanded with two-scenario contrast (partner-index vs proprietary-index dominance). Wikipedia source removed from the Sources block.
2026-05-26: Initial publish. Claude citations are the source attributions produced by Claude's web search tool across consumer surfaces (claude.ai, Desktop, mobile, Claude Code) and the Anthropic API. Web search announced 2025-03-20 with global rollout 2025-05-27. Entry covers per-surface citation behavior, the API citation schema (web_search_result_location with url / title / cited_text), how to optimize, what to skip, and the measurement gap. Sibling to perplexity-citation / microsoft-copilot-citations / ai-overview-citation / ai-mode / ai-dev-tool-citations in the citation-surface cluster.