A while back, we were writing blog posts for a client. Every time I delivered one, they'd paste it into an AI tool to "check my work." And every time, the AI told them the same thing: this looks like it was written by AI.
I'll be honest: on some of those posts, I did use AI. It helped with the first draft, and then I edited it the way I'd edit any reporter's copy. I checked the facts, cut the filler, fixed the structure, and made sure it actually answered the question. The tool flagged them.
So I got frustrated and wrote the next one completely from scratch. Every word, no AI anywhere near it.
The tool flagged that one too.
It especially didn't like sentences built like "it's not this, it's that." The client got frustrated, and for a while, that score became the conversation. Not whether the posts answered the right questions. Not whether the facts were solid. Not whether they'd bring in traffic.
Here's the thing: I learned to write in a newsroom. Journalism school at Humber College, then years of reporting on technology and business. And it turns out that's exactly why AI detectors can't tell my writing from a machine's.
Where did AI learn to write?
Large language models like ChatGPT learned to write by reading enormous amounts of text from the internet. A lot of that text is journalism.
When the Washington Post dug into C4, a dataset of roughly 15 million websites used to train AI models from Google and Meta, it found the New York Times was the fourth-largest source of text in the whole thing. Ahead of it were only Google Patents, Wikipedia and Scribd. And when the Times sued OpenAI and Microsoft in 2023, its lawsuit claimed millions of its articles had been used to train their chatbots.
So AI writes a bit like a decent reporter because, in a real sense, it studied reporters. Clear sentences. The main point up top. Short paragraphs. Those are newsroom habits, and AI picked them up from us.
Why does clear writing get flagged as AI?
AI detectors don't actually know who wrote something. They make a guess based on patterns. One of the big ones is predictability: if your word choices are expected and your sentences are clean and simple, the detector gets suspicious.
That's a problem, because predictable, clean and simple is exactly what editors spend years beating into young journalists. Say what you mean. Cut the extra words. Don't make the reader work.
Some phrases also get flagged because AI overuses them now. "It's not X, it's Y" is a perfectly good way to make a point. Writers used it long before AI came along. AI just borrowed it and wore it out.
Do AI detectors actually work?
Not very well. And that's not just my opinion.
- OpenAI shut down its own detector. In 2023, the company behind ChatGPT pulled its AI text classifier because of its "low rate of accuracy." In its own testing, it correctly caught only 26% of AI-written text and wrongly flagged 9% of human writing.
- They're biased against non-native English speakers. A Stanford study found that detectors flagged essays by non-native English writers as AI 61% of the time. They almost never made that mistake with native speakers. The reason: simpler vocabulary looks "predictable."
- They flag history. One popular detector rated the preamble of the Declaration of Independence as almost 98% AI-generated. Thomas Jefferson did not have ChatGPT.
A detector score tells you very little about who wrote something. It tells you even less about whether it's any good.
Does Google penalize AI content?
This is the question that actually matters for your business, and Google has answered it directly.
In its official guidance on AI-generated content, Google says it rewards high-quality content "however it is produced." It also says, plainly, that using AI "doesn't give content any special gains. It's just content."
What Google does go after is content made mainly to game search results. In March 2024 it introduced a spam policy against "scaled content abuse": pumping out lots of low-value pages just to rank. And Google is clear that the policy applies whether those pages were made by automation, by humans, or both.
So Google isn't asking "was this written by a person or a machine?" It's asking "does this actually help the person who searched?" That's the right question, and it's the one my client's AI checker never asked.
Why does journalistic writing work so well for search?
Because the habits newsrooms teach line up almost perfectly with what search engines reward.
Answer first. Journalists use the "inverted pyramid": the most important information goes in the first sentence or two. Google's featured snippets and AI Overviews do the same thing. They pull a short, direct answer from the top of a section. If your answer is buried in paragraph six, someone else's gets pulled instead.
Headlines that say what's inside. A good subhead tells the reader exactly what the next section covers. That also tells Google what the page is about.
One idea per paragraph. It makes a page easy to skim on a phone, which is how most people will read it.
Show your sources. Reporters attribute everything. Linking to credible sources builds trust with readers, and trust is the "T" in E-E-A-T, the framework Google uses to describe helpful, reliable content: experience, expertise, authoritativeness and trustworthiness.
Check the facts. Boring, essential, and the step most often skipped.
What can't AI do?
AI is a tool. It's like a hammer: you can build a house with it, or you can knock someone out with it. It depends on who's holding it.
That's how we use it at Qonda. AI can help get a first draft on the page. Then a person does the part that matters: adding what we actually know, checking every fact, and making sure the page answers the question someone typed into Google.
What AI can't do is report. It can't interview your customers. It can't visit a job site. It can't run an audit and tell you what it found. It can only rearrange what's already been written.
That's why Google added "Experience" to its quality framework. First-hand knowledge is the one thing a machine can't fake well. In our last post, the most useful numbers weren't the industry stats. They were the results from audits we ran ourselves, because nobody else had them.
And when AI is left alone without a human checking the work, it gets things wrong. Back in late 2022, the tech site CNET quietly started publishing AI-written finance articles. Of the first 77, it ended up issuing corrections on 41.
Why should a blog post answer questions?
Because that's what people type into Google. They don't search "dental implants." They search "how much do dental implants cost in Tulum" or "is it safe to get implants in Mexico."
A good blog post works like an extended FAQ. Each section answers one real question, clearly, near the top. That helps with Google's "People also ask" box, with AI Overviews, and with AI assistants like ChatGPT, which all look for direct answers to direct questions.
The best source of those questions is your own business. What do customers ask on the phone? What lands in your WhatsApp every week? Those are your blog topics.
One note for anyone who's been around SEO a while: Google used to show FAQ "rich results," those expandable questions under a search listing. It limited them in 2023, and by May 2026 they'd stopped appearing in search entirely. The special FAQ code won't do much anymore. Writing that answers questions still does.
What's the difference between SEO content and brand storytelling?
This is where my client and I were talking past each other.
The posts we were writing had a specific job: answer the questions their customers were searching for, and bring that traffic to the website. They weren't meant to be the company's origin story or its big-picture brand voice. That's valuable work too, and we do it, but it's a different job with a different scope.
When you judge SEO content by whether a tool thinks it "sounds human," you're measuring the wrong thing. The better questions are: Does it answer what people are actually searching? Is it accurate? Is it bringing in visitors?
Where to start
If you're writing content for your own website, with or without AI:
- Answer the question in the first two sentences. Then explain.
- Use questions as your subheads. The ones your customers actually ask.
- Cite your sources. Link to the original study, not a blog that summarized it.
- Add something only you know. A real example, a real number, a real photo from a real job.
- Have a human check every fact. Especially if AI helped write it.
- Ignore the AI detector. Ask whether the page is useful instead.
Let's look at your content
At Qonda Digital, we write SEO content in English and Spanish, built to answer the questions your customers are asking and to bring them to your site. We use the same habits I learned in newsrooms: get to the point, back it up, and check everything.
If you're not sure whether your website's content is working, we'll walk you through a free 30-minute audit. We'll show you what questions people are searching, where you show up, and where you don't. You keep the findings either way.
Email me directly at ty@qondadigital.com to arrange a free call.