Signs of AI Writing: 12 Mistakes That Make Content Look AI Generated

The most common signs of AI writing are inflated importance ("plays a pivotal role", "a testament to"), stock words like "delve", "tapestry" and "vibrant", sentences built as "it's not X, it's Y", ideas always grouped in threes, vague claims such as "experts say", and generic examples with no real detail. Add heavy formatting, a summary that starts "In conclusion", and leftover chatbot phrases like "I hope this helps", and readers can often tell a draft came straight from AI.

None of these signs proves AI was used, and using AI to help you write is not wrong. The problem is text that sounds like nobody in particular wrote it. This guide covers 12 signs of AI writing, what to do about each one, how people and detectors actually spot AI text, what Google says about AI content, and a simple editing routine to make AI assisted writing sound human.

Signs of AI writing at a glance

12 signs your content looks AI generated, and how to fix them

These patterns come mainly from Wikipedia's guide to the signs of AI writing, which volunteer editors use to spot AI generated articles. The guide itself stresses that they are clues, not proof: plenty of humans write this way too.

1. Puffing up the importance of everything

AI drafts love to tell you that something "plays a pivotal role", "marks a significant milestone" or "stands as a testament to" something bigger. Ordinary facts get dressed up as historic moments.

Fix: state what happened and why it matters to the reader, in plain words. "The new road cut the trip from two hours to forty minutes" beats "the road stands as a testament to the region's commitment to progress".

2. Stock AI vocabulary

Some words appear far more often in AI text than in human writing: delve, tapestry, testament, intricate, interplay, landscape, meticulous, pivotal, underscore, bolster, crucial and vibrant. Researchers saw this in real data. A 2025 study in Science Advances analysed more than 15 million medical research abstracts and found that words like "delves", "showcasing" and "underscores" suddenly became far more common after ChatGPT arrived, which suggested at least 13.5% of 2024 abstracts were written with AI help.

The favourite words change as models change, and Wikipedia notes that "delve" dropped off sharply in 2025. So the lesson is not to ban a list, but to notice words you would never say out loud.

Fix: swap each one for the word you would use in conversation: "explore" or "look at" instead of "delve into", "important" instead of "pivotal", "shows" instead of "underscores".

3. A tourist brochure tone

AI often slips into promotional language: places are "nestled in the heart of" somewhere, products "boast" features, communities are "vibrant" and offerings are a "diverse array". It reads like an advert even when you asked for information.

Fix: remove the adjectives and keep the facts. If something really is impressive, show it with a number or a detail.

4. "It's not X, it's Y"

Lines such as "It's not just a phone, it's a lifestyle" or "This isn't about money, it's about freedom" are a classic AI habit. So are "Not only... but also..." and "No fluff, no filler, just results." Used once, they can work. Used in every section, they become a tell.

Fix: make the point directly. "This phone's battery lasts two days" says more than a contrast formula.

5. Everything comes in threes

AI loves the rule of three: three adjectives, three benefits, three examples, again and again. Real writing has two reasons sometimes, and five other times.

Fix: list the points you actually have. If two are strong and the third is filler, drop the third.

6. Vague attributions

"Experts argue", "industry reports suggest" and "observers have noted" sound authoritative but say nothing about who said what. AI uses them to fill gaps, and readers increasingly distrust them.

Fix: name the source and link to it, or remove the claim. "A 2025 Stanford study found..." is worth ten "experts say".

7. Avoiding the word "is"

AI drafts often replace simple verbs with grander ones: something "serves as" a hub, "stands as" a landmark, or "features" a garden. Each one alone is fine, but together they make writing stiff.

Fix: use "is", "are" and "has" freely. Plain verbs read as confident.

8. Formula openings and endings

AI openings like "In today's digital age" or "In an era of rapid change" waste the first line. AI endings often start with "In conclusion", "Overall" or "In summary", or close with a section on challenges and the future that adds nothing new.

Fix: open with your main point or a real situation. End with what the reader should do next, not a summary of what they just read.

9. Formatting that shouts

Signs include Title Case On Every Heading, bold text in nearly every paragraph, bullet points that each start with a bold mini heading, emoji used as bullets, and tables where a sentence would do.

Fix: use sentence case headings, bold only the one thing a skimmer must not miss, and use lists only for real lists or steps.

10. Chatbot leftovers

These are the most obvious signs, because they were never meant for the reader: "Certainly! Here is your article", "I hope this helps", "Would you like me to...", "[Insert company name]", stray Markdown symbols like double asterisks, and odd citation codes copied from a chatbot's answer.

Fix: read the whole piece before publishing and delete anything addressed to you rather than to your reader.

11. Generic examples and no first hand detail

AI examples tend to be safe and interchangeable: "a small business owner might use this to save time." Nothing in them could only have come from you.

Fix: add what only you know: the exact tool you used, what went wrong, a real price in naira, a conversation with a customer, a place you visited. Google's own guidance asks whether content shows first hand experience, such as having actually used a product or visited a place.

12. Every sentence sounds the same

AI often produces paragraphs where every sentence is a similar length and follows the same pattern. The result is smooth but flat.

Fix: read your draft out loud. Where it drones, break a long sentence in two, or combine two short ones. Writing centres recommend reading aloud precisely because your ear catches what your eyes skip.

The em dash debate

The long dash, called an em dash, has become one of the most talked about signs of AI writing. AI models tend to use it far more than most everyday writers, often in places where a comma, a colon or brackets would be normal, and often for dramatic, salesy emphasis.

But the em dash is a weak sign on its own:

• Humans used it first. Writers from Emily Dickinson onwards have relied on it, and AI models learned the habit from human writing.

• It can now be turned off. In November 2025, OpenAI's chief executive Sam Altman announced that ChatGPT would finally follow custom instructions telling it not to use em dashes. You set this in Settings, under Personalization and Custom instructions.

• Newer models use fewer. Wikipedia's guide now treats heavy em dash use as more typical of older models, and says it is only useful alongside other signs.

If you want your writing to feel human, the em dash itself is not the problem. The problem is using it as a crutch for punchy statements in every paragraph.

How to tell if something is written by AI

People are better at this than you might think, if they use AI themselves. In a 2025 study presented at the ACL conference, researchers asked people to judge 300 articles. A panel of five people who often use ChatGPT for writing got just 1 of the 300 wrong, beating most commercial detectors, even when the AI text had been paraphrased to hide it. People who rarely used AI did barely better than chance.

The experienced readers did not rely on one clue. They looked for clusters: AI vocabulary, an overly formal tone, a lack of original ideas, and writing that was clear but empty. So if you are checking a piece of writing, look for several signs together, not one word or one dash.

Can AI detectors tell?

Not reliably enough to accuse anyone.

• OpenAI withdrew its own detector. In July 2023, OpenAI shut down its AI text classifier because of its low accuracy. At launch it caught only 26% of AI written text and wrongly flagged human text 9% of the time.

• Detectors can be biased against non native English writers. A 2023 Stanford study found that seven popular detectors flagged more than half of essays written by non native English speakers as AI generated, while correctly judging essays by US students. That matters for students in Nigeria and anywhere English is a second language.

• Even the best tools admit limits. Turnitin says its false positive rate is under 1% for documents with at least 20% AI writing, but higher for shorter texts and smaller amounts of AI. Vanderbilt University turned off Turnitin's AI detector in 2023, calculating that even a 1% error rate would wrongly flag about 750 of its 75,000 papers a year.

If you write your own work, protect yourself: keep your drafts and version history, and be ready to explain how you wrote it. For students, our guide on using AI to study without cheating explains where the lines usually are.

Does Google penalise AI content?

No, not for being AI generated. Google's Search Central guidance says it focuses on the quality of content rather than how it is produced, and that appropriate use of AI is not against its guidelines.

What Google does penalise is scaled content abuse: producing many pages mainly to manipulate search rankings, whether by AI or by people. Its spam policies give the example of using generative AI to create many pages without adding value for users. Google's helpful content guidance also asks whether a page shows first hand experience and original information, and says AI content should be reviewed and fact checked before publishing.

So the AI writing mistakes in this guide are not just a style issue. Generic, unchecked, interchangeable content is exactly what struggles in search.

How to make AI writing sound human

Use AI for a first draft or an outline if it helps, then edit it like this:

1. Start with your own notes. Give the AI your real experience, examples and opinions before it writes, so the draft has something specific to work with.

2. Delete the chatbot leftovers and any placeholder text.

3. Search for the stock words in the table above and replace each with the plain word.

4. Cut the formulas: contrast lines, automatic groups of three, "In today's" openings and "In conclusion" endings.

5. Replace vague claims with named sources, or remove them.

6. Add one detail per section that only you could write: a number, a name, a place, a mistake you made.

7. Fix the formatting: sentence case headings, less bold, fewer bullet points.

8. Read it aloud and change anything you would never say.

9. Fact check every claim, especially numbers, dates and quotes. AI can state wrong facts confidently.

To make ChatGPT sound more human from the start, add custom instructions such as "Write in plain British English, use short sentences, avoid em dashes and words like delve or vibrant, and never start with a summary." Our guide to hidden ChatGPT features shows where to set them, and our comparison of the best AI for writing covers which assistant writes most naturally.

A quick before and after

Before: "In today's digital age, email marketing plays a pivotal role in business growth. It's not just a channel, it's a powerful tool that helps businesses connect, engage and convert."

After: "Most of our repeat orders last year came from one monthly email to past customers. It costs us almost nothing to send, so it is the first thing we would set up again."

The second version is shorter, specific and believable, because it comes from real experience.

Why it matters: readers trust it less

Looking AI generated costs you trust. In a 2024 study of nearly 5,000 people in the US and UK, published in PNAS Nexus, simply labelling headlines as AI generated made people rate them as less accurate and less worth sharing, even when the headlines were true and even when a human had written them. The researchers found people assumed AI content had no human checking it.

That is why the fixes above work. Specific details, named sources and a clear human voice show the reader that a real person stands behind the words. For practical examples, see how to write emails that do not sound like a bot.

Frequently asked questions

What are the signs of AI writing?

Common signs of AI writing include inflated importance ("pivotal", "a testament to"), stock words like "delve" and "tapestry", contrast formulas such as "it's not X, it's Y", ideas always in threes, vague sources like "experts say", heavy formatting, "In conclusion" endings and leftover chatbot phrases. They are clues, not proof, so look for several together.

How can you tell if something is written by AI?

Look for a cluster of signs rather than one. AI text often sounds smooth but empty: formal, generic, full of stock phrases and missing specific details or opinions. Research shows people who use AI tools often are much better at spotting AI text than people who do not.

Is using em dashes a sign of AI?

Not on its own. AI models often overuse em dashes, but humans used them long before AI, and ChatGPT can now be told to avoid them. Treat a heavy use of em dashes as one possible clue among many, never as proof.

Are AI detectors accurate?

Not accurate enough to rely on alone. OpenAI withdrew its own detector in 2023 for low accuracy, and a Stanford study found detectors flagged more than half of essays by non native English speakers as AI written. Keep drafts and version history to show how you wrote your work.

Does Google penalise AI generated content?

Google does not penalise content just because AI helped write it. It focuses on whether content is helpful, original and trustworthy. It does penalise mass produced pages created mainly to manipulate rankings, so review, fact check and add real experience to anything AI drafts.

How do I make ChatGPT sound more human?

Give it your own notes and examples first, set custom instructions asking for plain language, short sentences and no em dashes or stock words, then edit the draft yourself: remove clichés, add specific details, name your sources and read it aloud before publishing.

What words does AI overuse?

Words often linked to AI writing include delve, tapestry, testament, pivotal, intricate, underscore, landscape, meticulous, bolster, crucial and vibrant, plus phrases like "plays a pivotal role" and "in today's digital age". The list changes as AI models change.

Is it wrong to use AI to help me write?

No. Using AI for ideas, outlines or a first draft is fine in most work, as long as you follow your school's or employer's rules, check the facts and make the final piece your own. Some schools and publications ban or limit AI use, so check before you submit.