Generative AI has invaded everyday writing, but the quality of the texts it produces is under scrutiny. As Proton announces full AI integration across its services and Reddit prepares to move to logged-in-only access to fend off bots, the public debate focuses on an uncomfortable question: is AI actually improving writing or flattening it? The issue is central to Human Driven AI and to anyone working on ethical AI.
What happened
In recent weeks, converging signals have emerged. Proton, the Swiss encrypted-services provider, announced widespread adoption of generative AI across its applications, triggering mixed reactions in the privacy community. In parallel, Reddit said it will make login mandatory to access content, a defensive move against AI crawlers eroding search traffic. The platform has seen organic visits drop as language models consume its data without returning users.
On the editorial front, The Atlantic published a podcast describing AI's entry into writing as a "post-shame" era: people no longer feel embarrassed about using automated tools, but the average quality of texts is declining. The Economist doubles down: AI often makes writing worse even when it seems to make it better, because it homogenizes style and thought. The Weighty Thoughts blog asks why AI is still poor at creative and argumentative writing, pointing to structural limits in models trained on mediocre texts.
Why it matters
The stakes are not merely aesthetic. If professional writing — journalism, communication, technical documentation, corporate email — is delegated to models that produce averagely correct but voiceless texts, the cost falls on readers and on the people who work in those fields. Newsrooms cut freelancers, companies cut copywriters, and the overall quality of information declines. The phenomenon touches the AI future of work: this is not outright replacement, but a gradual erosion of skills and critical capacity.
Then there is the data question. Reddit closing access is the symptom of an unsustainable economic model: platforms produce content that AI models use for free to train, returning texts that cannibalize their traffic. This is an AI governance problem before it is a technical one. Without clear rules on attribution and compensation, the quality of original sources — and therefore of the models themselves — is bound to degrade.
The HDAI perspective
Writing is a human activity before it is an output. AI can assist, but it cannot replace the responsibility of the person writing. The Proton case shows that even companies with a strong ethical identity struggle to balance innovation and transparency: integrating AI without explaining how, on what data and with what guarantees fuels distrust. The Reddit case, for its part, signals that the sustainability of the information ecosystem requires fair agreements, not walls.
The Human Driven AI perspective is clear: responsible AI is not measured by the amount of text generated, but by its ability to leave room for human judgment. Tools that propose, not impose; that assist, not flatten. It is the same philosophy that inspires HDAI Summit 2026 in Pompeii, where these themes will be at the center of the conversation among researchers, companies and institutions.
What to watch
In the coming months, watch two indicators: whether Proton will publish a detailed policy on data use for training, and whether Reddit will reach a licensing agreement with major AI providers. Both will signal whether the industry is moving toward ethical AI or toward further concentration of power. The quality of writing, in the end, depends on who controls the data and the rules.