We're fortunate to live in this era of computers, the internet, and AI.
These days, producing ten decent articles a day is no longer hard for a company. Smooth headlines, clean paragraphs, keywords right where they belong; you read it and can't find anything wrong. The problem is that if you can do it, so can your competitors. When every company can write like this, "decent" just means clearing the minimum bar.
Start with a small test. Open the latest article on your website, swap your company's name for a competitor's down the street, and read it from the top. If it still holds up completely, without a single word changed, that article hasn't actually said anything on your behalf. It's just repeating, under your name, what the whole industry already agrees on.
This piece is about the small part of an article that stops holding up once you swap the name. That part is the only part that's really yours.
01Will Google penalize articles written by AI?
This is probably the content question business owners ask most right now. Using AI to write, on its own, won't get an article penalized.
What Google's official documentation looks at is why the content was made. Using automation (including AI) to generate content with the primary purpose of manipulating search rankings violates its spam policies; if the purpose is to help readers, there's no problem.
The industry view points the same way. In September 2026, Zyppy Signal published a survey that asked 131 SEO experts to score more than a hundred possible ranking factors on a scale from −3 to +3. In the content quality category, "AI-generated content" scored −0.50 and "AI-generated content at scale" scored −1.82. The two are listed separately on the chart; they are not the same thing. The original write-up also makes a point of noting that the negative views mainly target bulk content with no added value, and some respondents shared cases where AI content that went through strict human review performed well.
One thing needs saying up front: this survey records how practitioners judge, not the weights inside Google's algorithm. It's fine for reading the direction, not for calculating scores. But the direction is already clear enough: what gets penalized is content that's mass-produced, looks like everything else, and doesn't say anything more. Whether AI was used is secondary.
So you can set aside "will AI writing get caught." The question more worth asking is: what does this article say that the hundred already online don't?
02Once content is flattened, two things are left
In the same survey, the top-scoring items in the content quality category were these: original research and first-hand data +2.19, content originality +2.02, information gain +1.93, demonstrating real experience +1.78.
Not one of them has anything to do with writing skill. They all ask whether you have something others don't.
Gather these items together and two things remain. One is first-hand experience: what you've seen with your own eyes and done with your own hands. Which clause a client got stuck on before signing last month, which approach you tried that failed, which line on your quote gets cut most often. The other is a stance: faced with a contested question, which side you're on, and why.
The framework Google uses to evaluate content quality is called E-E-A-T, its four letters standing for Experience, Expertise, Authoritativeness, and Trustworthiness. "Experience" was only added in 2022, and what it asks for is exactly the first-hand: a record of having actually used, done, or photographed something, to set it apart from the mass of second-hand content online that's been relayed and reworked. And of the four, the one placed at the most fundamental layer is Trustworthiness; the other three all lay the groundwork for it.
AI can't generate first-hand experience or a stance. Neither was ever in its hands to begin with, no matter how powerful the model gets.
03Short on material, AI writes only two kinds of things
There's a set of explainer cards about using AI to write business proposals that puts this precisely.
Its example is a single request: "Design a palletizing and shipping plan for FMCG orders." AI quickly hands back four recommendations: strengthen coordination across OMS, WMS, and TMS; set up a pallet information-sharing mechanism; optimize vehicle scheduling and loading; improve the exception compensation mechanism. Every one of them is correct, and every one of them would hold for any company.
The cards' conclusion is that when AI lacks business knowledge, it has only two ways to fill the gap. Fill it with industry common sense and you get "correct nonsense." Guess at your company's actual situation and you get "professional-looking mistakes."
Carry these two failures over to content, and they line up exactly with the two missing things above.
Correct nonsense is an article without a stance. It lays out industry consensus very neatly; no one who reads it will object, and no one will remember who wrote it. Most articles that pass the name-swap test are this kind.
Professional-looking mistakes are articles without first-hand experience. The numbers, specs, and cases they cite sound well-grounded, but the model actually filled them in on the fly. A practitioner's write-up on SEO for export-trade websites singles this out: AI often fabricates data and reports that don't exist, so every parameter, standard, and test figure in the text has to be checked one by one; once the content is wrong, rankings drop, and customers comparing suppliers get pointed in the wrong direction.
One line on the cards is worth remembering: it's not that it isn't smart enough. It's that it has no idea about the business knowledge in your head.
04A stance is a trade-off, not a tone
A stance means picking one side between two approaches that both make sense, and saying why. It has little to do with writing style or tone, and adding a few exclamation marks won't produce one.
Say a small coffee roaster wants to write a piece on "how beginners should choose coffee beans." The version without a stance lists ten points: origin, processing method, roast level, freshness, every one of them correct. The version with a stance says: if you're just starting out, we suggest holding off on single-origin beans; pick a blend and drink it for a full month first. Until you can tell flavor differences apart, the extra you pay for single-origin mostly buys you a label.
Some people will disagree with that second version, and other roasters may be the first to jump in and argue. But readers who finish it remember what this shop thinks, and the next time they buy beans, they know whether to trust it.
The last page of those proposal cards describes what AI can do once its business knowledge is filled in: identify contradictions, propose paths, explain trade-offs. AI can help explain the trade-offs. But which side of a trade-off your company stands on can only be decided by people. That's judgment, and judgment needs someone to sign their name to it.
05Division of labor: AI builds the frame, people bring judgment and first-hand experience
Our own journal is produced with an AI workflow, so this piece isn't trying to talk anyone out of using AI. The difference is that what AI receives isn't the one line "write me an SEO article." Before each piece is written, the topic sheet already has a "core argument" column filled in, spelling out what the piece claims, and then a "writing risks" column listing the directions this piece is most likely to go wrong. The material comes from a knowledge base we've built ourselves. AI builds the structure, pulls the material, and writes the first draft; people decide the stance, add the first-hand experience, and cut the places that claim too much.
That export-trade SEO write-up offers a similar workflow, roughly five steps: AI handles only the outline and first draft, with the prompt spelling out the target reader and industry context; a person adds real cases, customer feedback, and the pitfalls they've run into, which the author considers the most critical step; facts and data get checked one by one; template-style sentences get rewritten, with the opening and ending changed into your own voice; and finally, the pace is kept in check, with fewer, better pieces preferred over mass-publishing.
If we were to add one step to that workflow, we'd add it at the very front: before writing, put down in one sentence "what we're claiming in this piece." If you can't write it, it's not yet time to hand it to AI.
One more thing is easy to overlook. When you edit an AI first draft, you're sure to hit moments like this: "We wouldn't put it that way." "We actually advise against taking on this kind of client." Every edit like that is a stance. Fix it only in the draft, and AI will get it wrong again in the next piece; write it back into the knowledge base, and it'll know next time. As the proposal cards put it: every place AI misunderstands is a knowledge gap, and a knowledge base grows slowly out of each round of real work. It's the same point we made in "Your AI Has No Raw Material": companies aren't short on data; what they're short on is judgment that's been written down.
Where to start
You don't need to change your workflow first. Start by looking back at what you've already published.
Pull out your last five articles and ask two questions of each. First: swap your company's name for a competitor's — does the piece still hold up? Second: after reading it, could a reader finish the sentence "this company believes ___"?
The ones that fail both questions don't necessarily need to be deleted. Pick one of them, add a passage of first-hand experience only you have, then add a trade-off you're willing to stand behind. Change just those two things, and it becomes a different article.
"Stop Chasing Viral Posts. Go Build a House." was about making content a system, building it up one piece on top of the next. This piece fills in each brick itself. If every brick is standard-issue, no matter how high you stack them, no one can tell whose house it is.
This is also the everyday work of our SEO & content strategy service: first look at what first-hand experience you have and which judgments haven't been written down yet, then decide which topics are worth writing, and only then does AI start drafting.
Let AI move fast. Let people hold the anchor.
Rooted in craft. Built for the new wild.