Note revised on 25 May 2026. Article originally published in February 2026 — full rewrite. This note absorbs the content of the pieces previously titled "AI reshuffles the creation deck: 5 trends for 2026" and "Top 5 AI tools for writers in 2026", de-indexed and redirected to this page.
The public debate on writing assisted by generative AI reached, in 2026, an equilibrium point it did not have in 2023. The extreme positions — AI will replace writers, or AI will always produce texts inferior to human ones — have settled down. What has installed itself is a more nuanced professional practice, where the writer's value is no longer measured in raw writing time, but in the quality of framing, review and contextualisation they bring to work now shared with generative models.
This note sets out what editorial practice in the firm observes about this shift, and what the Google doctrine published on 15 May 2026 adds to this diagnosis.
What played out between 2023 and 2026
Three years sufficed to stabilise the practice. In 2023, the first uses of AI in writing were dominated by a fascination with the volume produced in little time. Many companies published serially generated content, little reviewed, little contextualised. This content quickly proved visible on reading, weak in search performance, and fragile in the face of the successive Google algorithmic updates that tightened the requirement on actual usefulness to the reader.
In 2024 and 2025, a more disciplined practice settled in structures that accepted to reinvest in editorial review. The model shifted: AI produces a structured raw material, the professional writer takes that material and injects what cannot be generated — the precise example, the brand tone, the cultural nuance, the editorial position, the factual verification. The overall productivity of the process effectively increased, but without the writer's suppression; on the contrary, their role became more precise.
In 2026, the Google doctrine published on 15 May confirmed that the search engine's generative features rely on the central ranking and quality systems of the engine[1]. This confirmation has a consequence for writing: what penalised mediocre content in the classical Google ranking now penalises it in generated responses as well. Selectivity increases. Editorial laziness costs more.
What AI does well, objectively
Practice observed in the firm, cross-checked with an MIT study published in 2023, confirms four tangible contributions of generative models to writing work.
Ideation and clearing the ground. A writer facing a new subject can, in a few minutes of exchange with a model, map a thematic field, identify angles, structure a provisional outline. This work previously demanded several hours of preliminary documentary research. The gain is real, and it frees up time for higher-value work.
Reformulation and adaptation. Adapting existing content to several channels or several audiences demanded substantial rewriting. Models produce correct variants in a few minutes, which the writer then adjusts. For structures that must decline their content into French, German and English — a common situation in Switzerland — this gain is particularly sharp.
The structured first draft. On a subject well framed by a precise brief, a model produces a first version containing the skeleton of the final content: sections, transitions, aggregated facts, correct formulations. The writer's work then consists less in writing than in revising, enriching and correcting.
Cross-verification and synthesis. On a documented subject, models can compare several sources, synthesise divergent positions, signal inconsistencies. This function is useful provided it is controlled — models still regularly invent references, citations or figures, and every factual assertion must be verified against the source.
A study conducted by MIT in 2023 documented, in a controlled experimental framework on realistic professional writing tasks (emails, short reports, structured notes), an average reduction of 37% in writing time on these tasks when assisted by a model, with quality perceived by independent evaluators equivalent or superior to non-assisted texts[2]. This order of magnitude remains a useful reference for framing the possible magnitudes of gain. It does not constitute a guarantee generalisable to all editorial production — long tasks, demanding brand content, or work with a strong strategic dimension present different profiles that are not covered by the study's scope.
What AI does not do
The observed limitations have been stable since 2023, despite the successive improvements of the models. Authentic storytelling, which rests on lived experience or fine understanding of an audience, remains human territory. Models produce correct but rarely memorable narratives. Humour, double meaning and precise cultural references remain difficult to handle — a French-speaking Swiss writer knows when to slip in a local reference or a tone calibrated on cantonal sensibilities; the models, for their part, remain literal or borrow clumsily.
Regulatory and sector nuances specific to the Swiss context — banking vocabulary, FADP requirements, formulations compliant with employment law, legal expressions specific to the Code of Obligations — demand expertise that models do not systematically have. Clear editorial position, which decides a subject rather than balancing it prudently, also remains difficult to obtain from a model without substantial editorial framing — models tend towards consensus and neutrality.
Where the writer's value shifts
This observation outlines a precise shift. A professional writer's value is no longer measured in volume of words written per hour. It is measured against four things.
The quality of the initial brief. A precise brief (audience, tone, position, constraints, length, sources) determines what a model can produce. A vague brief produces vague content. The writer who knows how to formulate a structured brief obtains usable material at first attempt.
The severity of the review. A review that only removes errors lets semantic redundancy, recognisable generic tone, hollow formulas and factual approximations slip through. A review that transforms the text by deleting, rewriting, injecting human expertise is the one that creates the value observable by the reader.
The sector expertise brought outside the text. The writer who knows the sector — fiduciary, watchmaking, medtech, law, finance — brings to the text precisions, examples, nuances that the model cannot correctly invent. This expertise is exactly what Google recognises in its E-E-A-T framework (experience, expertise, authoritativeness, trustworthiness), now structuring for ranking and therefore for AI citability.
Editorial coherence over time. A brand voice is built across dozens of publications coherent with each other, not on an isolated publication. The writer who holds this coherence over time produces cumulative value that models do not reproduce.
What May 2026 adds to this practice
The Google doctrine of May 2026 has two direct consequences for assisted writing practices. First, it confirms that the fight against mediocre generated content goes through the central ranking systems of search, not through a specific mechanism for generative features. Quality content, whether AI-assisted or not, will be treated on merit by search. Lazy content, whether assisted or not, will be too.
Then, it reinforces the argument that assisted writing is neither a shortcut nor a threat, but a tool — whose value depends entirely on the rigour of the practice employing it. A company that invests in a disciplined work chain (structured brief, assisted first draft, rigorous human enrichment, editorial review, factual verification) produces content that is both faster to publish and more solid against the growing selectivity of generative components.
This discipline is not a technological subject. It is a subject of editorial method, whose practice in the firm shows that it now distinguishes structures maintaining a reliable editorial presence from those losing it.
Sources
[1] Google Search Central, Optimizing your website for generative AI features on Google Search, published 15 May 2026. developers.google.com/search/docs/fundamentals/ai-optimization-guide [↩]
[2] Noy, Shakked, and Whitney Zhang. Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. MIT, 2023. economics.mit.edu/sites/default/files/inline-files/Noy_Zhang_1.pdf [↩]
Jérôme Deshaie is CEO of MCVA Consulting SA, a Swiss firm specialising in strategic consulting on artificial intelligence, based in Valais.