IP Law Daily, COPYRIGHT NEWS—Online publisher wikiHow accuses OpenAI of mass copyright infringement to build and operate ChatGPT, (Aug 26, 2026)

By Saurabh Kashyap, B.A., M.A., LL.B., LL.M.

OpenAI copied thousands of wikiHow’s ‘how-to’ articles without permission to train and operate ChatGPT, which then reproduced protected content while diverting readers from its website, the complaint asserts.

Online instructional publisher wikiHow, Inc. has filed a lawsuit against OpenAI entities in federal district court in New York, alleging that they copied thousands of copyrighted articles without authorization to train large language models (LLMs), used the articles through retrieval-augmented generation (RAG) systems, and generated ChatGPT outputs reproducing protected expression. The complaint asserted direct and vicarious copyright infringement and alleged that OpenAI violated the Digital Millennium Copyright Act (DMCA) by intentionally removing copyright management information (CMI). wikiHow sought damages, disgorgement of profits, statutory damages, attorney fees, and an injunction barring OpenAI from operating models trained on its copyrighted works without authorization (wikiHow, Inc. v. OpenAI, Inc., No. 1:26-cv-07171 (S.D.N.Y. Aug. 21, 2026)).

Founded in 2005, wikiHow operates an online instructional publishing platform containing step-by-step articles covering subjects ranging from everyday activities to specialized skills. The company alleged that it employs writers and editors and works with subject-matter reviewers to research, prepare, and maintain its content. Its business depends principally on advertising generated by visitors to its websites and fees from licensing its articles.

WikiHow sued nine affiliated OpenAI entities that allegedly operate together to develop, control, market and monetize generative artificial intelligence products. According to the complaint, OpenAI's flagship ChatGPT product uses LLMs trained on large quantities of human-written material and supplements those models through internet-connected functionality. wikiHow alleged that ChatGPT directly competes with its instructional articles by answering users' how-to questions without requiring them to visit wikiHow's webpages.

Copyrighted works. wikiHow asserted 1,211 copyright registrations covering 11,211 articles identified in an exhibit to the complaint. It alleged that each article contains protectable expression reflected in the selection, wording, and ordering of steps, methods, tips, and warnings. The company also claimed copyrights in more than 300,000 articles overall, although its infringement counts concerned the 11,211 registered articles.

The complaint identified three alleged forms of infringement. First, OpenAI allegedly copied wikiHow articles on a massive scale for inclusion in datasets used to train its LLMs. Second, OpenAI allegedly retrieved and copied the articles through RAG systems that provide external information to ChatGPT when generating answers. Third, ChatGPT allegedly produced outputs containing verbatim or near-verbatim portions of wikiHow articles and reproduced their distinctive selection, sequencing and organization.

Training copies. wikiHow alleged that OpenAI populated its training datasets by scraping its websites directly or through intermediary datasets. It pointed to OpenAI's disclosed use of WebText, WebText2 and Common Crawl in developing earlier GPT models and alleged that similar datasets containing wikiHow material were used for later models.

According to the complaint, wikiHow's website received more than 185,000 visits from OpenAI's published IP addresses between March and May 2025. It further alleged that OpenAI bots visited and scraped the website 148,529 times between May 1 and July 21, 2026, despite robots.txt directives blocking OpenAI crawlers. wikiHow alleged that it had expressly prohibited scraping its content for AI training without written permission and had notified OpenAI of the alleged unauthorized use in December 2024.

ChatGPT outputs. The publisher alleged that ChatGPT had "memorized" portions of its articles and could reproduce them on request. The complaint included examples comparing GPT-4 outputs with wikiHow articles concerning boundaries with a mother-in-law after a baby's birth, interpretations of wedding dreams, differences between March and April Aries, and whether "soul ties" are real. The accompanying comparisons on pages 20-25 highlighted passages that wikiHow alleged ChatGPT reproduced verbatim or nearly verbatim.

WikiHow further alleged that infringement occurred during ordinary ChatGPT use because RAG systems retrieved its articles and supplied their contents as context for answers. Thus, according to wikiHow, users could receive the substance of its instructional content without requesting or even identifying a wikiHow article.

Fair use. Anticipating a fair-use defense, wikiHow alleged that OpenAI's copying was commercial and non-transformative. ChatGPT allegedly repackaged the same expression, meaning, and instructional purpose rather than creating a different purpose or message. Moreover, wikiHow contended that training required copying entire articles and that OpenAI's products supplied substitutes for those articles.

The publisher alleged particularly significant market harm because its articles are freely accessible and its business depends on readers visiting its webpages. ChatGPT answers allegedly diverted those visits and consequently reduced advertising and licensing revenue. wikiHow also alleged that it approached OpenAI about licensing its content beginning in spring 2024, but OpenAI continued using the material without reaching a licensing agreement.

Copyright information. The DMCA claim alleged that OpenAI intentionally stripped CMI—including article titles, authors' names, publication information and terms of use—from wikiHow content. According to the complaint, OpenAI used content-extraction tools designed to separate article text from webpage elements such as bylines, footers and copyright notices.

WikiHow alleged that removing this information concealed the source and ownership of material incorporated into training datasets and subsequently reproduced to users. It sought statutory damages of $2,500 to $25,000 for each DMCA violation, in addition to copyright damages and other relief.

Relief sought. wikiHow asserted direct infringement for copying articles into LLM training and RAG systems, direct infringement based on ChatGPT outputs, vicarious infringement against OpenAI entities that allegedly controlled and financially benefited from the conduct, and CMI removal under 17 U.S.C. § 1202(b)(1). It requested actual damages and OpenAI's attributable profits or available statutory damages, including enhanced damages for alleged willful infringement. It also sought a permanent injunction preventing OpenAI from operating models trained on wikiHow's copyrighted works without authorization and demanded a jury trial.

The Case is No. 1:26-cv-07171.

Judge: NA.

Attorneys: Dustin Fire (Susman Godfrey L.L.P.) for wikiHow, Inc.

Companies: wikiHow, Inc.; OpenAI, Inc.

News: Copyright NewYorkNews