People searching for how to cite Claude usually trip on the same three things: who the author is, which model name to write down, and what to do about a URL when the conversation is private. This page answers those for Claude specifically, then gives the formats style by style. For the rules that apply to every AI tool, see the main guide to citing AI tools.
A note on the examples. Your conversation with Claude is private to you, so we cannot show a "real" one. Each section therefore gives the format as a template with [bracketed] slots for you to fill in, followed by the style authority's own published example where one exists, quoted and linked so you can check it.
What is specific to Claude
The author is Anthropic. Claude is the product; Anthropic is the company that makes it. Every style below uses Anthropic wherever it needs an author, publisher or developer. Do not write "Claude AI" as the author, and do not list Claude as a co-author of your paper. COPE and ICMJE both say an AI tool cannot be listed as an author, and ACM requires every author to be an identifiable human being.
Model names have three parts. Claude models are named by family, tier and number: Claude Opus, Claude Sonnet and Claude Haiku, each with a version number. Claude Sonnet 4.6, for example, was announced on 17 February 2026. Copy the name from the model picker as written. "Claude 4" on its own is ambiguous because Opus and Sonnet are different models.
Share links are snapshots, and public sharing depends on your plan. Anthropic's help center describes sharing as switching a chat from Private to Public to create a link, and says users on the Free, Pro and Max plans can review their shared chats under Settings > Privacy. Only a public link is useful in a citation, because a reader has to be able to open it. According to the same page, the link shows the chat as it was when you shared it, messages sent afterwards stay private unless you unshare and reshare to update the snapshot, and attached files are not included. That last point matters: if your prompt was "summarise this PDF", the reader will see Claude's answer but not the PDF. Say what the file was in your text or appendix.
Work and school accounts cannot share publicly. Anthropic's help center says Team and Enterprise users can share a chat only with members of the same organisation. For a reader outside it the chat is non-retrievable, so use the fallback formats below.
Claude inside other products. If you reached a Claude model through another company's app, cite the product you actually used and name the model in the description or in your methods section. If you called the API in a study, report the exact model identifier, the dates and your parameters in the methods section; that is reporting a research instrument, not citing a source.
APA 7
APA's guidance (9 September 2025) gives two templates. Cite a specific chat when you can share it and the reference will help readers. Cite the tool in general when citing the chat would be unhelpful or inappropriate, or when you used Claude for something like editing, translating or refining code.
A specific chat with a public link:
Anthropic. ([year, month day]). [Title of the chat] [Generative AI chat]. [Claude, or the model name]. [URL of the shared chat]
APA's own published example of a Claude chat, which you can open:
Anthropic. (2025, May 20). Essential grammar topics for high school graduates [Generative AI chat]. Claude Sonnet 4. https://claude.ai/share/329173b2-ec93-4663-ac68-4f65ea4f166d
The tool in general:
Anthropic. ([year]). [Claude, or the model name] [Large language model]. https://claude.ai
APA's own example of this form is: Anthropic. (2025). Claude 4 Sonnet [Large language model]. https://claude.ai/new
In text: (Anthropic, 2025) or Anthropic (2025). The italic title in the first format is the title of the conversation, shown at the top of the chat and in the sidebar; APA suggests editing it into something descriptive before you share. The date of a chat reference is the day the chat took place or ended. For the tool in general, APA uses the year the tool was last updated. If you cite two chats from the same year, APA's usual rule applies: 2026a and 2026b.
APA does not put the prompt in the reference. Describe it in your text instead, and cite the tool there; APA's sample Method sentence names the model and cites it as (Anthropic, 2025). Long exchanges can go in an appendix or supplemental material if your instructor or editor agrees.
MLA 9
MLA's update (13 August 2025) keeps its 2023 logic: no author, a description of what was generated (usually your prompt) as the title, the tool as the container, the model as the version, the company as publisher. It prefers a stable share link and falls back to the tool's general address when there is none.
"[Your prompt, or a description of what Claude generated]" prompt. Claude, model [model name], Anthropic, [Day Mon. Year], [claude.ai/share/... or claude.ai].
MLA's worked examples all use ChatGPT. This is one, for comparison:
"Describe the theme of nature in Jane Austen's Mansfield Park" prompt. ChatGPT, model GPT-4o, OpenAI, 23 Sept. 2024, chatgpt.com/share/66f1b0a0-d704-8000-be9a-85f53c850607.
In text, use a shortened form of the title in quotation marks, as you would for any work with no author.
Chicago (notes-bibliography and author-date)
Chicago's Q&A, which relies on the 18th edition, says that for most writing you can simply name the tool in your text. For a student paper or research article it gives a note. University guides to the 18th edition add that a chat is treated like a personal communication and is not usually listed in the bibliography.
- Text generated by [Claude, with the model name], Anthropic, [Month Day, Year], [URL of the shared chat].
If the prompt is not in your sentence, put it in the note:
- [Claude, with the model name], response to "[your prompt]," Anthropic, [Month Day, Year], [URL].
Chicago's own examples, from the Q&A: "1. Text generated by ChatGPT, OpenAI, January 10, 2026, URL." and "1. ChatGPT, response to 'Explain how to make pizza dough from common household ingredients,' OpenAI, January 10, 2026, URL."
If you do include a bibliography entry, which library guides allow only when you have a publicly available link, file it under the developer: Anthropic. Response to "[your prompt]." [Claude, with the model name], [Month Day, Year]. [URL]. McMaster University Library's example of this form is: OpenAI. Response to "Tell me how to fix a flat bicycle tire." ChatGPT-4, September 30, 2024, followed by the share link. In author-date, library guides show a parenthetical such as (ChatGPT, March 7, 2023) in place of the note. If you edited the generated text, Chicago says to say so in the text or at the end of the note ("edited for style and content").
Harvard (Cite Them Right)
Harvard has no single authority, so check your university's version first. Cite Them Right is widely adopted in the UK; the templates below follow Robert Gordon University Library's reproduction of it, which distinguishes output you can share from output you cannot.
With a public link:
Anthropic Claude ([year]) [The query you submitted]. Claude response to [Your Name]. Available at: [URL of the shared chat] (Accessed: [day month year]).
Without one, it is a personal communication:
Anthropic Claude ([year]) Claude response to [Your Name], [day month].
In text: (Anthropic Claude, [year]). The published Cite Them Right example of the second form, as reproduced by RGU, predates Claude's wide use and names a Google tool: Google Bard (2023) Bard response to Amanda Grayson, 14 June. Ulster University's Harvard guide follows the same company-plus-tool pattern and gives (Anthropic Claude AI, 2023) as an in-text example.
IEEE
The IEEE Reference Guide has no format for generative AI. What IEEE does have is a disclosure rule: AI-generated content in a submission, including text, figures, images and code, must be disclosed in the acknowledgments, identifying the AI system and the sections that use it, with a brief explanation of the level of use. IEEE says that using AI for editing and grammar is generally outside that policy.
If an instructor still wants a numbered reference, the nearest fit is the Reference Guide's software format. The entry below is our adaptation of it, not an IEEE rule:
[1] Anthropic. Claude. ([year]). Accessed: [Mon. Day, Year]. [Online]. Available: https://claude.ai
Follow your instructor's version if they give one.
Vancouver and AMA
Neither puts a chat in the numbered reference list. University guides to the AMA Manual of Style (section 3.15.13) say not to cite generative AI as a reference and to describe its use in the running text or acknowledgments instead: how you used it, the tool name with version and manufacturer, the dates of use and, for research use, the prompts. Vancouver has no official guidance on AI, so libraries such as University College Dublin recommend treating a chat like a personal communication, mentioned in the text only. ICMJE adds two rules for journal submissions: describe the use in the cover letter and in the manuscript, and do not reference AI-generated material as the primary source.
Disclosing Claude use
A citation covers words and ideas you took from Claude. It does not cover help that leaves no quotable trace, such as restructuring a draft or checking code. For that, write a disclosure. A pattern to adapt, with your own details in the brackets:
I used [Claude model] (Anthropic) on [dates] to [what you used it for, for example suggesting counterarguments to my thesis]. I evaluated every suggestion, wrote all of the final text myself, and verified all sources in the original publications. Prompts and outputs are in [appendix].
Only say what is true of your own work. Where the statement goes depends on the venue: a methods section for empirical work (ACM's policy, updated 14 May 2026, requires this when AI was used to conduct the research), acknowledgments for IEEE, a declaration form for many undergraduate courses. Course policies vary widely and override everything here. If your module bans generative AI, citing Claude does not make the use acceptable.
Verify before you cite anything Claude told you
Claude can state things that are wrong and can describe papers that do not exist. If Claude points you to a study, find the study, read it, and cite the study. A reference to Claude only ever supports the claim "this is what Claude said". It cannot support a factual claim about the world. Our guide to citing sources you found through AI tools covers that workflow.
Sources
All pages were checked on 21 September 2026.
- American Psychological Association, Citing generative AI in APA Style: Part 1, reference formats (9 September 2025)
- MLA Style Center, How do I cite generative AI in MLA style? (Updated and Revised) (13 August 2025)
- The Chicago Manual of Style Online, Q&A: citing content developed or generated by AI (the Q&A itself gives no section number and states that it relies on the 18th edition). The 18th edition's section on AI-generated content is paywalled, and the library guides number it differently: McMaster University Library cites 14.111 and Monash University Library cites 14.112. We also followed Columbia College Library and Western Technical College Library
- Cite Them Right is paywalled; we followed Robert Gordon University Library and Ulster University Library
- IEEE Author Center, Submission and peer review policies: AI-generated content and the IEEE Reference Guide (software format)
- The AMA Manual of Style is paywalled; we followed Himmelfarb Library, George Washington University and Duquesne University Library. For Vancouver, UCD Library
- ICMJE, Recommendations: artificial intelligence, use by authors
- ACM, Policy on Authorship (updated 14 May 2026)
- COPE, Authorship and AI tools position statement
- Anthropic, Introducing Claude Sonnet 4.6 (17 February 2026)
- Claude Help Center, Share and unshare chats
