The Guardian’s Peter Lewis certified his columns as human-written

Columnist Peter Lewis from The Guardian publicly confirmed that he writes his texts himself, without AI-generated paragraphs. The occasion was sparked by the growth of AI-generated texts and the launch of the Proudly Human initiative, which offers authors the opportunity to certify human authorship.

Why This Became a Topic

Lewis writes that editorial pages have begun to be filled with what he calls "slop-inion" — columns and essays where the author's voice is replaced by machine-like smoothness. As symptoms, he cites recent cases in Australian media: Crikey had to remove a series of materials on leadership, and the editor of Capital Brief complained that 80–90% of submitted texts look like AI generation.

For the author, the problem no longer boils down to simple plagiarism, when a person simply passes off a ChatGPT response as their own work. In his view, the real risk zone lies much earlier — at the stage of conception, research, and argument assembly. Hence the main question of the text: if the model doesn't just correct spelling, but helps find factual material, suggests an angle of presentation, and points out logical gaps, where does human authorship end?

Lewis calls this not a technical problem but a cultural one. Checking the final text for "bot-style" is insufficient, because the final column can look lively even if important decisions were already suggested to the author by the system. This is why he, despite long-standing skepticism toward generative AI, specifically spent a month with Claude from Anthropic to understand the boundary in practice.

How Certification Works

In this experiment, Lewis relied on the approach of former chief scientist of Australia Alan Finkel, who launched the international Proudly Human initiative. Its goal is not simply to say "I wrote this myself," but to establish clear rules after which the text is still considered human-authored.

The principle de minimis is at its core: such auxiliary actions are permissible that do not undermine the author's right to consider the work their own. In other words, it's not about a complete ban on AI, but about a ban on its substantive co-authorship.

• Spelling and grammar checking are permitted
• Generating ideas at a general level is acceptable
• Automating the basic design of the story and material structure is undesirable
• Assigning text fragments to be written by the model is not allowed
• Using content that significantly affects the final work is not allowed

For Lewis, the key word here is provenance — the verifiable origin of the text. The reader should understand that they are being addressed by a specific author, not a well-tuned assistant hidden behind a personal brand. In the article, he compares this to practices of verifying the origin of indigenous art, where the question of authorship is directly related to protection against appropriation and forgeries. In his logic, such a label becomes both an author's promise and a new mechanism of trust for the audience.

"I don't want to make my work easier at all."

What His Experience Showed

Lewis describes in detail his usual cycle of working on a column: two weeks before the deadline, he reviews the political agenda, formulates questions for studying public sentiment, searches for cultural associations, and then gradually builds an argument. Then come early poll results, drafts, discussion with close readers and editors, final proofreading, and submission to the editorial office.

In this scheme, AI can indeed be useful: quickly test a thesis for soundness, find connections in a data array, highlight logical gaps, or suggest a non-obvious, albeit somewhat weak, metaphor. But the result of the experiment for him turned out to be rather negative. Yes, such support speeds up the process and reduces cognitive load, but at the same time, what he considers the essence of writing disappears: doubt, trying out variants, agonizing assembly of structure, rejection of beloved but unnecessary phrases, several failed versions before a successful one.

Lewis writes that it is precisely this "intellectual friction" that makes text come alive. If the difficult part is removed, the column may become more efficient in production, but poorer in tone and thought. Therefore, human certification for him is not only a public promise to the reader, but also protection from the temptation to choose too easy a path.

What This Means

Lewis's story shows that the next debate around AI in media will not only be about detectors and formal plagiarism, but about the origin of idea, argument, and voice. For editorial offices, this could result in new rules for disclosing AI participation, and for readers — the appearance of labels that function as a trust signal. If such practices take hold, the question will sound not "did AI touch this text at all," but "at what point did the machine become a co-author."