The Best Writing Style? Human apparently.

How AI turned authenticity into something we have to perform.

Early in my career, I managed content teams. I had worked as a content producer myself before that, so editing and directing other people’s work came naturally. Most of my feedback focused on whether the content was on-brand and built for search.

Now, I produce and edit content so that it sounds more human.

That often means rewriting a sentence until it sounds like something I would say out loud, which contradicts almost everything I was taught about business writing.

This is strange, because the same things I used to edit towards now make readers suspicious and less trusting. Business writers have spent years learning how to build an argument and structure paragraphs that work for the reader and for algorithms. Now that AI has absorbed those conventions, the same work can read as detached, even “un-human.”

So these days I edit for “presence”. I write as though I am talking to a peer, adding personal detail around a point and letting more of my own voice into the sentence.

---

A small preprint released in April suggests our suspicion of AI-generated content may already be changing how we write. Researcher Daniel Tabach asked 21 people to write short opinion pieces with access to the same AI platform, but warned some of them that an AI detector would analyse their work.

When 251 judges compared all the work, they were slightly more likely to read the warned writers’ work as human, at about 54% vs 46% of those who didn’t get the warning.

The study is small and no one has peer reviewed it yet, so take this with a grain of salt. But it is worth noting that the participants who expected a detector to check their work spent longer writing and exchanged more messages with the tool, and even some stopped using it altogether.

---

Online content writers already serve more than one audience. We write for the person we want to reach, and we write for the algorithms that decide whether the work reaches them at all.

We learnt to think about keywords, headlines, meta descriptions and where to place an answer so Google could understand the page. Some of that improved online writing and made useful information easier to find, but also produced millions of articles designed around what people searched for rather than what a writer wanted to say.

AI engine optimisation asks writers to think about whether a model can understand their work well enough to cite or surface it.

---

This is all becoming difficult because how I structure for presence, can also become a formula. AI can write conversationally, adding an anecdote or dropping a slightly awkward phrase into the middle of a paragraph so the writing feels less polished. You can prompt it to include doubt, humour and a personal detail or two.

Once we define human writing as a set of stylistic traits, those traits become available to anyone with a prompt.

If polished sentences no longer prove a person wrote them, and conversational sentences no longer prove someone lived them, what becomes the proof?

---

Perhaps this is why the question of whether something was “written by AI” increasingly feels inadequate. Professional writing will involve AI somewhere in the process, whether it helps with research, editing, translation or the first draft.

Drawing a clean line between human and machine production will become harder, and probably less useful.

---

I will probably continue deleting sentences because they sound like AI. I will keep reading my work aloud and changing phrases until I can hear myself in them.

But I also want to remember why I started writing in the first place.

I write to understand what I think, and to connect that with what someone else is thinking. Search engines and AI systems may help the writing travel, but the reason for sending it stays human.

Someone could still be reading it on the other end.

All the Zest 🍋 Cien

Cien Solon is the CEO and co-founder of LaunchLemonade, building governed AI agents for regulated industries.

References and other reading

Daniel Tabach, “Can Humans Detect AI? Mining Textual Signals of AI-Assisted Writing Under Varying Scrutiny Conditions,” April 2026. https://arxiv.org/abs/2604.23471

Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, “GEO: Generative Engine Optimization,” November 2023. https://arxiv.org/abs/2311.09735

Google Search Central, “AI features and your website.” https://developers.google.com/search/docs/appearance/ai-features

Google Search Central, “Creating helpful, reliable, people-first content.” https://developers.google.com/search/docs/fundamentals/creating-helpful-content

Edward Chen, “’Humanizer’ tool can erase signs of AI-written text,” Nature, July 2026. https://www.nature.com/articles/d41586-026-02105-3

Nature, “Universities are relying on AI-detection software to catch cheating. How well do the programs work?,” July 2026. https://www.nature.com/articles/d41586-026-01358-2

Jasper Roe, Mike Perkins, Peter Bannister, Leon Furze and James Wood, “Dramaturgies of Deception: AI Humanizers and the Performance of Legitimacy in Higher Education Assessment,” May 2026. https://arxiv.org/abs/2605.02649

Utsav Paneru, “Please Make it Sound like Human,” April 2026. https://arxiv.org/abs/2604.11687

Yongtong Gu, Songze Li and Xia Hu, “MASH: Evading Black-Box AI-Generated Text Detectors via Style Humanization,” January 2026. https://arxiv.org/abs/2601.08564