The opposite would have surprised more than one reader, but it is now written in black and white: Reporterre, an online media outlet dedicated to ecology, bans the use of generative AI in its newsroom, motivated by a
The media thus stakes out a position "against the grain," in a profession that seems to have "accepted, sometimes willingly, to go along with it," laments the editorial team in its statement. The picture is admittedly mixed and murky: while few media outlets openly acknowledge resorting to generative AI, these tools are quietly but steadily slipping into even the most prestigious newsrooms, despite editorial charters or codes of conduct whose precision sometimes leaves much to be desired. Other media outlets are developing tools based on these generative systems (chatbots, automatic summaries, etc.) and are even forging opaque partnerships with the companies that develop them. In this context, it's hard for readers not to get lost.
While Reporterre is part of a broader trend that isn't unique to it (Splann, Bon Pote and others have also chosen to exclude these tools from their editorial process), the outlet stands out through a refusal that spans all production stages: from writing to communications, including even the preliminary research for articles. A commitment that follows logically from the outlet's editorial stance, yet proves harder to implement in the face of growing ubiquity.
A manifesto, not a charter
For journalist Erwan Manac'h, who regularly covers the environmental and social impact of AI for Reporterre, the text won't spark an internal revolution, describing a pre-existing «tacit consensus» on the use of generative tools. The publication wasn't even the subject of major internal debate, and stems according to him from journalistic practice that would anyway make the place of these tools marginal: «We do quite a bit of reporting and work with direct sources whose information isn't heavily digitized», the journalist explains. The commitment, whose arguments were also explained in a video on social media, thus amounts more to a «manifesto», an opportunity for the media to take a stand on a societal issue, than to a charter aimed at governing the use of AI that had become urgent to curb.
But while there was a "tacit consensus" on the subject within the newsroom, the risk of it being challenged as these tools are progressively deployed and their use becomes increasingly widespread in the profession convinced the newsroom to take a stand. The various arguments put forward to support this decision draw on several articles published by the newsroom or other outlets (data-center pollution, impact on memory and critical thinking, degradation of working conditions, unauthorized use of content, etc.), giving it more the character of an opinion piece addressed to readers, of a direction the newsroom commits to following as it moves forward in a context fraught with uncertainty.
But the text provides no verification system, no list of banned tools, and certainly no sanctions in case of breach. Internally, the editorial team relies on trust rather than surveillance through AI detection tools, whose effectiveness is contested anyway. "Just as we won't police each other, we won't police our freelancers," clarifies Erwan Manac'h… which doesn't prevent the editorial team from remaining vigilant. The standard editorial workflow at Reporterre means articles go through the hands of copy editors, tasked with double-checking information. To date, one article containing hyperlinks revealing ChatGPT usage was detected by the editorial team: the freelancer was contacted again and asked to provide direct sources. For more extreme cases, the procedure to follow if there is suspicion of massive AI use in the production of an article has not yet been defined. "The question has never come up yet," admits the online media journalist. It remains to be seen whether this trust agreement will be enough for readers who, beyond good intentions, might well demand guarantees.
A trust contract serving its business model
Clarification was essential for a media outlet whose financial health depends directly on its readers. The publication of this commitment coincides, within a few days, with the publication of their annual accounts, showing more than 54,000 donors who represent 95.4% of its budget.
Marine Slavitch
Yet while other media outlets have addressed AI use through various positions, charters and codes of conduct, the boundaries and limits sometimes remain unclear. What does editorial work "assisted" by AI under a journalist's "control" actually mean? What about research? Correcting spelling mistakes? LinkedIn posts? As the web increasingly appears to be artificialised, giving rise to theories about a "dead web" where AIs would react to and share content written by AIs in an endless loop, authenticity and craft have become rare values to which some internet users express strong attachment—even a need, as Erwan Manac'h observes: "They really want (…) handmade journalism." A transparency and valorisation of the "artisanal" nature of an article that could, for instance, take the form of a label alerting readers to the absence of AI at the end of a publication. The outlet is considering this, though nothing has been finalised at this stage.
Accepted limits and constraints
Reporterre is well aware of this: reading the text will undoubtedly spark plenty of "yes, but..." responses, including and perhaps especially from within the profession itself. The outlet anticipates this by immediately clarifying one point: the commitment targets generative AI specifically, making a distinction from "traditional" AI. While the term is now spontaneously associated with large language models (LLMs) like ChatGPT or Claude, as well as image generation tools, AI actually refers to a much broader computing field that emerged in the mid-twentieth century. "We'll keep using our GPS to get to our reporting locations," notes Erwan Manac'h. Among the footnotes provided at the end of the text, the editorial team specifies that they may use "digital tools using a form of AI whose editorial use became standard before the advent of generative AI (translation software, tools for transcribing audio interviews into text, or audio mixing software notably)", but that they "will favor the most ethical, free, sovereign and environmentally friendly software available". Their commitment focuses above all on any editorial contribution that would substitute for a human, the journalist clarifies, regardless of task scope: "We could certainly imagine that when a 300-page UN report in English arrives, the brief could be written in a fraction of a second by artificial intelligence. That's precisely what we refuse."
Owen Huchon
A refusal that extends to upstream research into articles, in the name of environmental and social issues documented in the manifesto on one hand, but also a matter of method and journalistic rigor. For Erwan Manac'h, a query via a conventional search engine offers more context and clarity on the source and path leading to the information… but this distinction could quickly become obsolete. Google recently deployed AI-generated summaries directly in its search results, or even an entirely generative mode where traditional blue links have given way to an AI-produced web page. « Given the speed and imposed nature of these technologies, the more time goes on, the more a simple Google search will become complicated to conduct », admits the journalist.
Technology transparency that is essential
Faced with a rapidly evolving landscape, the media doesn't rule out having to reconsider certain choices. "This is going to be a question that will come up every time," acknowledges the journalist, for whom the statement will serve as a "prod" to facilitate decision-making in internal debates, much like the choice already made to stop flying for reporting. It wouldn't be the first time that Reporterre has publicly reversed a position already shared with its community: in December 2024, the editorial team published "Why Reporterre isn't leaving X" to justify its continued presence on the platform despite its flaws... before publishing, a month later, "Reporterre leaves X". Such a reversal on generative AI seems highly unlikely, but the editorial team intends to communicate about the evolution of its approach if necessary: "We will act transparently if there are new developments," assures Erwan Manac'h.
Owen Huchon
If using donations to finance plane tickets would be an obvious inconsistency that wouldn't escape its supporters, the same logic could apply to digital tools: what share of their donations ends up in the coffers of Google, Microsoft or Adobe? At a time when technological independence is entering the public debate, a newsroom's choice of tools is no longer merely an internal matter. Erwan Manac'h doesn't rule out the possibility that the newsroom, which already applies choices favoring technological independence and sustainability in its internal tools, might account for them more openly in the future. Could it go so far as to publish, as it does with its finances, a review of its technological choices, and specify concretely what those famous "most ethical, free, sovereign and ecological software" it commits to prioritizing are? "There would be things to say, that's clear, but for now, it's not an idea that's floating around," the journalist replies.
