What we know about the filing
IPWatchdog reported on August 25 that wikiHow has filed a lawsuit against OpenAI over ChatGPT's use of its how-to library. The case joins a long queue of publisher-vs-AI disputes, but the subject matter is new: online instruction guides.
The full complaint is still emerging and the precise wording of the claims will be parsed for weeks. The underlying allegation is obvious, though: wikiHow hosts an enormous archive of step-by-step guides, ChatGPT answers the same questions with similar steps, and someone's traffic and licensing revenue is the casualty.
Previous lawsuits against OpenAI focused on news articles and books. How-to guides are a different beast. They sit at the intersection of creative expression and functional facts, which makes this a sharper legal test than yet another dispute over recycled news articles. A court will have to decide how much originality survives in a format that is, by definition, standardized: "First, gather your tools."
In the US, everything depends on fair use
OpenAI's likely answer in any US copyright case is fair use — the doctrine that certain copying is transformative and therefore lawful. The problem for OpenAI is that recent American case law has moved in the opposite direction.
In Thomson Reuters v. Ross Intelligence (February 2025), a Delaware court ruled that a legal AI startup infringed copyright when it trained its model on Westlaw's judicial headnotes. The parties settled before a damages trial, but the ruling remains the strongest US precedent to date against the "training is fair use" theory. Meanwhile, the US Copyright Office continues its multi-part study of AI and authorship, and publishers keep lobbying for a statutory licensing regime. The terrain is unsettled, and every new filing matters.
Europe plays by different rules
For European readers, the wikiHow case is not just an American legal soap opera. The EU relies on a completely different framework.
Directive (EU) 2019/790 introduced a text-and-data mining (TDM) exception in its Article 4. AI companies may use lawfully accessible online content for training, unless the rightsholder expressly reserves that right — typically through a machine-readable statement, metadata or terms of service.
That is a legal instrument US publishers simply do not have. wikiHow — or any European publisher — can say "no AI training" in a form the law recognizes. If a provider ignores that opt-out, it commits copyright infringement under the directive's national implementations. No open-ended fair-use balancing test required.
The AI Act adds a second layer. Its Article 53 obliges providers of general-purpose AI models such as OpenAI to implement a copyright compliance policy, to respect CDSM opt-outs, and to publish a sufficiently detailed summary of the content used for training. These obligations have been binding since August 2025; the regulation as a whole became fully applicable in August 2026. And they are not aspirational: violations can cost up to €15 million or 3% of global annual turnover — whichever is higher.
The transparency summary matters in practice. European rightsholders gain a public document they can check to see whether their content was used for training. In the US, a rightsholder usually finds out only through litigation and discovery. In the EU, the answer is supposed to be published.
What we see from our own workflows
At ai-jarvis.eu we run production AI services — article pipelines, transcription, TTS — and we benchmark models on our own rig. The substitution effect behind this lawsuit is familiar: ask a capable model "how do I fix a squeaky door?" and you get a perfectly structured step-by-step list. It reads like a wikiHow entry without the wikiHow branding.
That is the economic harm at the core of the case: every answer is a pageview that never reaches the original site. European publishers creating tutorials, recipes and repair guides face exactly the same squeeze — and they face it with American platforms on one side and European remedies on the other.
Practical steps for European creators
- Block AI crawlers in robots.txt. A simple technical fix that limits future crawling, though it cannot undo content already used for training.
- Make your opt-out machine-readable. A human-readable "no AI training" banner is not enough under Article 4(3) of the CDSM Directive. Express the reservation in metadata and in the site's terms of service.
- Use the AI Act transparency summaries. For EU-based rightsholders, the published training summaries are a new way to determine whether their work was included.
- Consider licensing. OpenAI has signed licensing deals with major European publishers such as Axel Springer and Le Monde. A licensing agreement is often faster and cheaper than years of litigation.
What happens next
The wikiHow case will take years, and the odds of an out-of-court settlement are good — most large copyright fights end there. But before any settlement, the court will have to confront the substantive question: can a how-to article be protected as creative expression, or is it just a compilation of unprotectable facts and methods?
If the court rules for wikiHow, OpenAI faces pressure to license how-to content at scale — not just for wikiHow, but for every tutorial site. If it rules for OpenAI, instructional content becomes a fair-use training resource, with obvious consequences for the European publishers who produce much of that content in local languages. Service availability is unaffected — ChatGPT remains fully available in the EU — but the outcome will shape how it is trained and how content owners are compensated.
We will follow the case from Prague, and we will keep an eye on how European courts apply the same questions under their own rules.
Could the wikiHow lawsuit make ChatGPT unavailable in the EU?
No. A US copyright judgment does not directly remove a product from the EU market. But if OpenAI were ordered to change its training data or licensing practices, the practical impact would be felt globally, including by European users.
Does blocking AI crawlers with robots.txt fully protect my content?
Not fully. It prevents some future crawling, but content may already be in training data, and archived copies or other crawlers can bypass the block. In the EU, the stronger tool is the statutory opt-out under Article 4(3) of the CDSM Directive.
Are step-by-step instructions even protected by copyright?
The expression of the instructions is protected, but the underlying ideas, methods and facts are not. That is precisely the gray zone this lawsuit will test.
