The signs of AI writing, and the plain fix for each

Wikipedia's editors keep a long, public list of the habits that give machine-written text away. This chapter takes the ones that appear most on business websites and shows how to rewrite each one.

The best known signs of AI writing are a group of overused words (delve, crucial, testament, showcase, vibrant and others), sentences built as "not X, but Y", lists of three added for rhythm, claims that something is pivotal or a lasting legacy, vague sources such as "experts say", and endings that repeat the introduction. The fix is to state each fact once, in plain words, with its real source. Detection tools make errors, so judge the page by its content.

The guide Wikipedia's editors use

Wikipedia has a special reason to care about this. Thousands of volunteers check its articles, and since ChatGPT became public on 30 November 2022 they have had to deal with text pasted from chatbots. A group of editors called WikiProject AI Cleanup keeps a public page, "Signs of AI writing", that lists the habits they meet most often, each with real examples. The page was last edited on 4 November 2026, the day this chapter was written.

Two warnings from that page belong at the start of any discussion of the subject. The first is that the list "is descriptive, not prescriptive": it describes what editors have seen, and it is no rule book. The second is that the signs are "only potential signs of a problem". The deeper problems are wrong facts, invented sources and empty content. Removing the surface signs and leaving those problems in place, the page warns, "could just make detection harder."

So, what follows is useful for editing, and every fix comes back to the same idea: a sentence should carry a fact or an opinion the writer holds. Text that only sounds like writing is the thing to remove.

The overused words

Wikipedia's guide cites several studies showing that chatbots use certain words far more often than people do. Its list of "words to watch" includes: additionally (especially at the start of a sentence), align with, boasts (meaning "has"), bolstered, crucial, deep dive, delve, emphasizing, enduring, enhance, fostering, garner, highlight (as a verb), interplay, intricate, landscape (as an abstract noun), meticulous, pivotal, robust, showcase, tapestry (as an abstract noun), testament, underscore (as a verb), valuable and vibrant.

The list changes with each generation of models. Wikipedia notes that "delve" was "famously overused by ChatGPT in 2023 and early 2024" and then "dropped off sharply in 2025". One or two of these words in a page mean little. Many of them together, used again and again, are what the page calls "one of the strongest tells for AI use."

The fix is to use the ordinary word. "Our team boasts 20 years of experience" becomes "Our team has 20 years of experience". "We delve into the intricacies of kitchen design" becomes "This guide explains how kitchen layouts work". "A vibrant neighbourhood" becomes a fact about the neighbourhood: the market on Saturdays, the school two streets away.

Sentence shapes that give text away

Words are easy to search for. Sentence shapes are harder, because no single word marks them, and they are the habits that survive most often after a quick edit. Wikipedia lists several, and these four appear on business websites again and again.

• Negative parallelism. Sentences such as "It's not just a kitchen, it's a lifestyle" or "This isn't about price; it's about value". The first half denies something nobody claimed, so that the second half sounds larger. Wikipedia lists the forms "Not only ... but", "Not X, but Y" and "Y rather than X". The fix is to state the positive fact once: "The kitchen has room for six people to eat."
• The rule of three. "Fast, reliable and affordable." "Quality, service and value." Wikipedia notes that chatbots "often use this structure to make superficial analyses appear more comprehensive." Keep a list of three only when there are three real items, and give each one a fact.
• Undue emphasis on significance. Phrases that call an ordinary event "a pivotal moment", "a testament to" something, or part of a "broader trend" or "lasting legacy". A bakery opening a second shop is news to its customers; it marks nothing in history. Keep the fact and drop the importance.
• Vague attributions. "Experts say", "studies show", "many believe", "industry reports suggest". If a real source exists, name it and link to it. If it does not, the sentence goes.

Endings, openings and other structures

The shape of the whole page can give it away as well. Wikipedia's editors list "outline-like conclusions about challenges and future prospects": the closing section that says the subject "faces challenges" but "continues to thrive", or that "the future looks bright". On a business blog the same habit appears as a final paragraph headed "Conclusion" that repeats the introduction in different words.

Openings have their own habits. Chatbot drafts often start by announcing what is coming ("In this article, we will explore...") or by stating how important the topic is "in today's fast-paced world". A reader who searched for a question wants the answer, so put the answer in the first paragraph and let the rest of the page explain it.

Then there is text that was meant for the chatbot's user and was pasted by mistake: "Certainly! Here is a blog post about...", "I hope this helps", "Let me know if you would like me to expand on any section". Google's rater guidelines name "As an AI language model" as a sign that content was paraphrased with an AI tool. These lines are the easiest to find, and a search for the words "I hope", "certainly" and "as an AI" across your site takes a few minutes.

The dash, and why one sign is never enough

From 2022 the long dash, the punctuation mark called the em dash, was treated as a sign of AI writing. Wikipedia's guide now lists it under "historical indicators", with dates from 2022 to September 2026. The page explains that newer models have been adjusted to use it less, and some to avoid it entirely.

This history teaches something useful. Each single sign rises and falls with the models in use, and a careful human writer can produce any of them by choice. Professional writers use dashes, and Wikipedia's list of "ineffective indicators" adds that perfect grammar, formal language and "bland" writing prove nothing on their own.

What does identify slop reliably is the combination: many signs at once, on page after page, together with the absence of facts only the business could know. That is why the fix is always the same, whatever the sign: give the reader a real fact, in plain words, from a named source.

Can a tool decide if text was written by AI?

Many owners would like a tool to answer the question for them. Wikipedia's guide tells its editors not to rely only on AI detection tools "such as GPTZero and Pangram", because "these tools have non-trivial error rates" and can be confused by small changes to the text or by models they were not trained on. The same page reports that a 2025 study found people's ability to tell AI text from human text "no better than random chance".

For a business the conclusion is practical. Google judges pages on quality and value, and none of its documents on AI content asks site owners to pass a detector. Use the signs in this chapter as an editing checklist, then judge the page by what it gives the reader.

A table of signs and fixes

The table below collects the signs from this chapter into one place, with an example of each as it appears on business websites and the plain rewrite. Print it, keep it next to the screen, and use it on every draft.

Signs of human writing

Wikipedia's page also lists signs of human writing, and they make a good target for editing. Human writers, the editors found over 25 years of articles, use simple "is" and "has" sentences more often. They choose plain words: "wrote" over "authored", "moved" over "relocated", "used" over "utilized", "tried" over "attempted". They make definite statements, such as "was the first" or "is the only", when the facts allow it.

For a business, the strongest sign of human writing is a detail no model could know: the name of the street where the job was, the question a customer asked last Tuesday, the reason the team stopped using a supplier. Chapter 5 explains how to collect those details before you start writing.

Part of a larger guide

This page is one part of Avoiding AI slop. The other parts:

• What AI slop is
• What Google says
• The editing process
• First-hand experience
• Checking facts
• Bylines and disclosure
• Publishing at scale

Questions people ask

What words make writing sound like AI?

Wikipedia's guide lists words chatbots use far more often than people, including delve, crucial, pivotal, testament, showcase, vibrant, tapestry, intricate, meticulous, garner, foster, underscore, enhance, bolstered, boasts and landscape in the abstract sense. One or two may be coincidental, the guide says, but many together are "one of the strongest tells for AI use." Replace each with the ordinary word or a specific fact.

What is negative parallelism in AI writing?

It is a sentence that denies something nobody claimed so that the next part sounds larger, as in "It's not just a kitchen, it's a lifestyle." Wikipedia lists the forms "not only... but", "not X, but Y" and "Y rather than X" as common in chatbot text. The fix is to state the positive fact once, with a detail the reader can check.

Is the em dash a sign of AI writing?

It was for a time. Wikipedia's guide now lists overuse of the em dash under "historical indicators" for the period 2022 to September 2026, because newer models have been adjusted to use it less. A dash alone proves nothing, since professional writers use it, but many dashes together with other signs still suggest a pasted chatbot draft.

Do AI detectors work?

Not reliably enough to decide on their own. Wikipedia's guide tells editors not to rely solely on detection tools such as GPTZero and Pangram, because they "have non-trivial error rates" and can be confused by small text changes. It also reports a 2025 study in which people did no better than chance. Judge a page by its facts and usefulness instead.

Why do AI drafts use lists of three so often?

Wikipedia's guide observes that chatbots "often use this structure to make superficial analyses appear more comprehensive." Three items sound complete, so the model fills the third place even when it has only two real points, or none. Keep a list of three only when each item is a distinct fact, and give each one its proof.

What should I search for on my site to find pasted chatbot text?

Search your pages for phrases written for the chatbot user: "I hope this helps", "Certainly!", "Let me know if", "As an AI language model", "In this article, we will explore" and "In conclusion". Google's rater guidelines name "As an AI language model" as a sign of paraphrased AI content. Finding and removing these lines takes a few minutes.

Does perfect grammar mean a text was written by AI?

No. Wikipedia's guide lists perfect grammar among its "ineffective indicators", noting that many writers come from professional backgrounds. Formal or "bland" writing proves nothing on its own either. Look instead for groups of signs together: overused words, inflated claims, vague sources and a lack of facts that only the author could know.

What are signs of human writing?

Wikipedia's editors found that human-written articles use simple "is" and "has" sentences more often, plain words such as "wrote", "moved" and "used" in place of "authored", "relocated" and "utilized", and definite statements such as "was the first" when the facts allow. On a business site, the clearest sign is a detail only the business could know.

Should I just remove the AI words and publish?

No. Wikipedia's guide warns that treating the signs as the only problem "could just make detection harder" while the real problems stay: wrong facts, invented sources and empty content. Remove the words, and also check every fact at its source and add your own experience. A page that only changed its vocabulary is still slop.

Why does AI writing exaggerate how important things are?

Wikipedia's guide notes that chatbot output "skews positive and verbose by default", and phrases such as "a pivotal moment" or "a testament to" make any fact sound weightier. Wikipedia lists this "undue emphasis on significance, legacy, and broader trends" as a common sign. On a business site, replace the phrase with the fact itself: what happened, when, and why it matters to the customer.

Sources

1. Wikipedia:Signs of AI writing, Wikipedia, WikiProject AI Cleanup. Last edited 4 November 2026, accessed 4 November 2026.
2. Search Quality Evaluator Guidelines, Google. Version of 11 September 2025, accessed 4 November 2026.
3. Creating helpful, reliable, people-first content, Google Search Central. Last updated 2026-10-05, accessed 4 November 2026.

Keep reading

• How to avoid AI slop: the full guide
• The editing process from AI draft to published page
• What first-hand experience adds that AI cannot
• What AI slop is, and how to spot it on your site
• Run the free Get Found Check

Want pages that sound like your business and bring enquiries?

Licheo's agents write and update your pages from your own facts, and a person checks every page before it goes live. Start with the free Get Found Check to see how customers find you today, or contact us for pricing.