Readers rarely think "this was written by AI." They just stop reading. The tells are concrete: every paragraph the same length, sentences that restate the heading, transitions doing no work. Each has a fix — specifics, varied rhythm, and details only someone who was actually there could add.
The tells are a checklist, not a vibe
Machine-flavored text announces itself in four repeatable ways. Paragraphs of nearly identical length. Sentences that restate the claim before them instead of advancing it. Long comma-spliced chains that keep qualifying one thought. And stock transitions, the moreovers and furthermores, stapled between paragraphs that do not actually connect. None of this requires a detection tool to spot. Read a page aloud once and you will hear all four.
Uniform word counts read like a metronome
Open a filler article and squint. Every section: one heading, one 95-word paragraph. The shape never changes because nothing in the content is pushing back on it. Real subject knowledge is lumpy. Some questions deserve four paragraphs and a table; some deserve two sentences. Let the material set the length. If every section of a draft weighs the same, that is not discipline, it is the absence of judgment.
Sentences that restate instead of advance
The heading says clones ship on Mondays. The first sentence says shipping happens at the start of the week. The second adds that this weekly cadence is when orders go out. Three sentences, one fact, and the reader feels the wheels spinning. The repair rule is blunt: every sentence must add something checkable. A number, a name, a condition, a consequence. If a sentence only rephrases its neighbor, cut it and watch the page tighten.
Stock transitions and comma-splice chains
"Moreover" and "additionally" are load-bearing nowhere. They appear when paragraphs were generated separately and needed glue. The same goes for chains like "quality matters, consistency is important, readers notice these things, trust develops over time" — four claims spliced together, none developed. Replace the glue with an actual relationship: because, but, which is why. If no real relationship exists between two paragraphs, the problem is the outline, not the transition.
The fix is information, not attitude
Bolting on slang or a jokey tone does not make text read as human-written. Information does. Prices from your own register. The strain that sells out by Friday. What the delivery driver gets asked at the door. Google's position backs the same standard: AI-assisted drafting is not against its policies, but scaled content that adds no value is spam, whoever produced it. The page earns its place by containing something the other ten results do not, and first-hand detail is the cheapest way to get there for an operator and the most expensive thing for a competitor to fake.
A worked example: one paragraph, before and after
Before: "Blue Dream is a popular strain that many consumers enjoy. It offers a balanced experience that appeals to a wide variety of users. Its effects are often described as uplifting." Every sentence restates the one before it, and nothing is checkable. After: "Blue Dream is the strain our budtenders hand to nervous first-timers. Sativa-leaning but gentle, berry-forward, and it sells out most Fridays." Two sentences, one long, one short. A specific use case from the counter, a flavor note, an inventory fact. That paragraph could only come from a shop, and readers can tell.
Key takeaways
- The four tells: uniform paragraph lengths, restate-the-claim sentences, comma-splice chains, stock transitions.
- Cut every sentence that repeats its neighbor; each line must add a checkable fact.
- Let content set section length. Identical weights signal no judgment was applied.
- First-hand specifics, like prices, sell-out days, and counter questions, beat tone tricks.
- Google does not ban AI drafting; it targets scaled low-value pages. Added information is the bar.
Frequently asked questions
Will Google penalize AI-written pages?
Not for the tool. Google's guidance is that AI-assisted content is fine and low-value content produced at scale is spam, whoever typed it. The test is whether the page adds information a searcher could not get from the ten results already ranking.
Which tell should I fix first?
Restated claims. Delete every sentence that repeats the heading or the sentence before it in different words. Pages routinely lose a fifth of their length this way and read better immediately, before you touch voice or rhythm at all.
Do AI-detection tools settle the question?
No. They flag careful human writers and clear lightly edited machine text, in both directions, too often to act on. Reading the page aloud and checking it against a short list of tells catches more with less collateral damage.
Can a machine draft become a good page?
Yes. Treat the draft as scaffolding. Cut the restatements, break the uniform paragraph lengths, and replace every vague claim with a specific only your operation knows, like a price, a strain that sells out, a thing a customer said. The edit is the work.
Want this done for your brand?
We build cannabis & CBD search visibility with full transparency, for dispensaries, brands, and MSOs.
Get a free cannabis SEO audit ›