Common AI writing patterns
Unedited chatbot copy has habits you can learn to see: the stock opener, a handful of pet words, sentences cut to one length, paragraphs all the same shape, no facts of its own. On its own, one proves nothing. A page full of them went live without an edit, and ten minutes with a text editor tells you which kind you have.
What are the most common AI writing patterns?
The standard reference is Signs of AI writing, kept by the Wikipedia editors who clean out machine-written text, and it calls each entry a potential sign of a problem, not the problem itself.
Stock phrases and pet words
The typical opener praises today's fast-moving digital world. Further down come garner and interplay, words no one says out loud. One of them is a habit. Six is a template.
One sentence length, one paragraph shape
You write a short line. Then a long one that runs on through a clause or two, since the thought needs the room, and the reader feels the change of pace. Stock copy keeps each sentence in one middle band, with not a single one of 30 words or more. And almost every paragraph has two to four sentences.
Layout habits that give it away
Look for a bold label and a colon in front of every list item, headings split in two by a colon, emoji in place of bullets and a last section called Conclusion that restates the whole page.
Sentence shapes that repeat
Transition words open sentence after sentence while adjectives travel in threes. Then there's the frame that tells you what something isn't before it says what it is.
Facts that aren't there
Someone who knows the trade mentions a price or a street without thinking. Stock copy never does, and readers sense that gap before they spot a single word.
Is an em dash still a sign of AI writing?
Not as much as it was. Newer models use far fewer, and ChatGPT drops them when asked. Wikipedia files heavy use under historical indicators, dated 2022 to September 2026.
Pet words fade the same way. The guide notes that delve was everywhere in ChatGPT text in 2023 and dropped off sharply in 2025.
So judge rhythm and facts first. A model will skip a word on request, but a page with nothing of its own reads the same whatever wrote it.
How to check a page for AI writing patterns by hand
You need the page and a text editor, nothing else.
- Copy the main text into the editor, without the menu and footer.
- Read the first sentence of each section. Would it fit any page on the web just as well?
- Search for stock phrases and pet words and note where they cluster.
- Count the sentences in each paragraph, and look for one sentence of 30 words or more.
- Check the layout against the habits above, starting with the colons.
- Count the facts, from prices and dates to things the business did.
If most steps turn something up, rewrite the page.
One page in 298 read like stock AI copy
On 17 October 2026 we pointed MarketingTracer's AI signals check at corewebvitals.io, our own site: 269 of its 298 pages came out low, 28 medium and one high.
The high one was /core-web-vitals-report. It scored 66 out of 100 and opened with this line:
In the ever-evolving landscape of web development
Its lists put a bold label and a colon in front of every item, and the check flagged buzzwords and lists of three.
Rewritten the same day, it now scores 19.
For calibration, the site's articles scored between 0 and 20 with a median of 14. In the same test, human texts from before 2022 scored between 1 and 10 while unedited ChatGPT copy scored 100.
How to avoid AI writing patterns when you rewrite a page
Put the answer or the offer where the opener was. Say plainly what each stock phrase was hiding, and add what only the business knows, like the price and the lead time. Rhythm comes last. Merge two thin paragraphs, break up a fat one. Don't aim for zero, though. A page that answers the question can carry a few signals and still sound like a person.
Check every page of a client site at once
Checking by hand is fine for one page and hopeless for a client with 300. MarketingTracer's AI content check runs inside the on-page scan, gives each English or Dutch page with enough text a score from 0 to 100 and highlights the sentences that carry a pattern.
FAQ
Does Google penalize AI-written content?
No, not for the AI part. Google's February 2023 guidance says appropriate use of AI or automation is not against its guidelines.
Its spam policies target scaled content abuse instead: many pages generated primarily to manipulate rankings, with little in them for readers, whatever produced them. That is the content farm problem with a faster engine.
Can a detector prove a text was written by AI?
No. OpenAI withdrew its classifier on 20 July 2023, citing its low rate of accuracy: in its own tests it caught 26% of AI-written text and mislabeled 9% of human writing. Pattern counts can't prove authorship either, because edited AI text with real specifics reads like human writing.
Are colons a sign of AI writing?
Not inside a sentence, where a colon does honest work. A colon in every heading on the page is the habit.
Which pages should I check first?
Start where buyers decide, on the home page and the service pages. Then go through the product and category pages, where stock phrases crowd out the sizes and delivery times buyers came for.