AI writing
Signs of AI writing, and how to fix each one
The common signs of AI writing are inflated words such as "delve", "tapestry" and "testament to", stock openers and recaps, sentences that all run the same length, "not just X but Y" contrasts, groups of three, and emoji bullets with bold labels. Drafts from older AI tools also put a long dash in most sentences, though many newer tools avoid it. None of them proves a text was written by AI, since people write this way too, so use them as an editing list: cut the padding, say the plain thing and put your own specifics back.
AI writing tools produce text that is correct and easy to skim, and after a while it all sounds alike. Readers have learned the signs of AI writing: the same handful of inflated words, an opener that could start any article, a tidy recap at the end, and sentences that march along at one length.
This guide goes through every group of signs that the free Check phrases button in our AI humanizer marks, with a plain swap for each and before and after examples we wrote ourselves. Treat every sign as a reason to edit. None of them tells you who wrote a text.
Why AI drafts show the same signs
An AI writing tool builds text one word at a time and favors words that are likely to come next. Likely words are safe words, so drafts drift toward a formal, general style that fits any topic. The tools are also tuned to sound helpful and complete, which is why so many drafts announce what they will cover, add a balanced “on the other hand” and then sum up what you just read.
None of these habits is wrong on its own. The trouble is density. One “pivotal” in a report is fine; five stock words, three recaps and a dash in every sentence make a page read as filler, whoever wrote it.
The shift in vocabulary is measurable. A study of over 15 million biomedical abstracts from 2010 to 2024 found a sudden rise in style words such as “delves”, “showcasing” and “underscores” once AI chat tools were widely available, and estimated that at least 13.5% of the 2024 abstracts had been processed with AI (Kobak et al., Science Advances). Wikipedia’s editors keep a longer field guide for spotting AI text in encyclopedia articles, linked in the sources below.
Common AI words and what to use instead
These are words AI uses a lot more than most people do, and Check phrases lists them under Overused words. The right column gives the plain word to try first. Often the better fix is to cut the word and say the specific thing it was standing in for.
| AI word or phrase | Try instead |
|---|---|
| delve, delve into | explore, look at, cover, or cut it |
| tapestry, a rich tapestry of | mix, range, or name the parts |
| a testament to | shows, or cut it |
| navigate the complexities, navigate the challenges | handle, deal with, work through |
| realm, in the realm of | area, field, or cut it |
| the digital landscape, the business landscape | the market, the field, the situation |
| pivotal | important, main, or show the effect |
| crucial, crucially | important, or say what happens without it |
| seamless, seamlessly | smooth, simple, or name the step that got easier |
| robust | strong, reliable, or say what it withstands |
| leverage | use |
| foster | encourage, build, help |
| underscore, emphasize | shows, or make the point directly |
| showcase, highlight | shows, or cut the commentary |
| intricate, intricacies | detailed, complex, the details |
| meticulous, meticulously | careful, or say what was checked |
| commendable | good, or say what was good |
| garner, bolster, boast | gets, supports, has |
| interplay | how X and Y affect each other |
| enhance, enhancement | improve, or say what changes |
| vibrant | lively, busy, or describe it |
| comprehensive, holistic | full, complete, or cut it |
| multifaceted, nuanced | complex, or name the sides |
| paramount, invaluable, indispensable | important, useful |
| transformative, revolutionize, game-changer | say what changed and by how much |
| groundbreaking, cutting-edge, state-of-the-art, unparalleled | new, modern, or name the feature |
| elevate, unlock, unleash, empower, supercharge | improve, open, help, or say how |
| embark, journey | start, process, path |
| ever-evolving, fast-paced, rapidly changing | cut it, or say what changed |
| myriad, plethora, multitude | many, or give the number |
| cornerstone, bedrock, linchpin, hallmark | basis, core, or cut it |
| synergy, paradigm shift | say what the teams do together |
| utilize, facilitate, streamline | use, help, simplify |
| truly, deeply, incredibly, remarkably, undeniably | cut it, or add the evidence |
| align with, resonate with | fits, matches, or say how |
| valuable insights, valuable resources | name the insight or the resource |
| a key role, a key factor, key takeaways | main, big, or say why |
| a wide range of, a diverse array of | many, or list them |
Swap a word only where the plain one says the same thing. “Robust” is the right word for a statistical test, and “landscape” is fine in a sentence about hills.
AI phrases to avoid, group by group
Single words are only part of it. Most of what makes a draft sound machine-made is whole phrases that could sit in any text on any topic. The tables follow the groups that Check phrases shows, with some of the smaller groups merged.
Stock phrases and openers
| Phrase | Fix |
|---|---|
| It's important to note that, It's worth mentioning that | Delete the lead-in and say the thing |
| In today's fast-paced digital world | Start with your first real point |
| In the world of marketing | In marketing, |
| When it comes to pricing | On pricing, |
| Let's dive in, Let's break it down, Let's take a closer look | Delete it |
| In this article, we will explore | Start with the first point |
| At the end of the day, In a nutshell, The bottom line is | Cut it |
| has never been more important, more important than ever | Say what changed |
| Imagine a world where, Picture this | State the benefit |
| The possibilities are endless, Only time will tell, The future looks bright | End on your last real point |
| the power of, the magic of, the art of | Cut it |
| a double-edged sword, strike the right balance | State both effects plainly |
Puffed-up importance and advert tone
| Phrase | Fix |
|---|---|
| plays a crucial role in, plays a pivotal role in | Say what it does |
| serves as a, stands as a | is |
| a powerful reminder of, a stark reminder that | State the fact |
| left an indelible mark, a lasting legacy | Say what changed |
| paved the way for, set the stage for, marked a turning point | Say what happened next |
| rich cultural heritage, woven into the fabric of | Name the tradition, or cut it |
| reflects broader trends | Name the trend |
| underscores the importance of | Make the point directly |
| nestled in the heart of, hidden gem, breathtaking | Give the location plainly |
| Whether you're a beginner or a seasoned pro | Name who it is for |
| take it to the next level, look no further, say goodbye to | Say what improves |
| unlock the full potential, harness the power of | Say what the reader can now do |
| a commitment to excellence | Say what you actually do |
| thoughtfully crafted, expertly curated | made from, built for |
| user-friendly, hassle-free, effortless | Describe the feature |
Stacked connectors, vague sources and recaps
| Phrase | Fix |
|---|---|
| Additionally, Furthermore, Moreover (to open a sentence) | also, and, or nothing |
| Notably, Importantly, Interestingly | Cut it |
| That being said, With this in mind, This ensures | but, so, or nothing |
| Experts say, Studies show (with no source) | Name the source, or own the claim |
| may potentially, it could be argued that | Keep one hedge |
| While specific details are limited, Based on available information | Cut it, or state what is known |
| Despite these challenges, faces several challenges | Name the challenge |
| In conclusion, In summary, Overall, Ultimately | End on your last real point |
| Conclusion, Final Thoughts, Key Takeaways (as headings, mostly from older tools) | Drop the label |
| By embracing these strategies, we can build a brighter future | Cut it |
Email lines, social posts and leftover chat lines
| Phrase | Fix |
|---|---|
| I hope this email finds you well | Start with the reason you are writing |
| Please don't hesitate to reach out | Just reply if you need anything. |
| I wanted to reach out, touch base, circle back | I'm writing about... |
| Thank you for your time, understanding and consideration | Keep one thanks |
| Here's the thing, Here's why | State the point |
| Let that sink in, Read that again | Delete it |
| I'm thrilled to announce, I'm humbled to share | Give the news first |
| Agree? Thoughts? Drop a comment below | Ask a real question, or cut it |
| A long row of hashtags at the end | Keep two or three that people search |
| I hope this helps! Certainly! Great question! | Delete it |
| As an AI language model, As of my last update (mostly older tools) | Delete it |
| [Your Name], [Insert date], [Company] | Fill it in |
| Citation codes copied from a chat window | Delete them |
| What a great question, an incredible team, amazing results | Cut it, or say what they did |
Signs in the shape of the text
Some of the strongest signs lie in how sentences and pages are built. Check phrases marks the common ones under Sentence habits and Formatting habits, and the Sentence rhythm panel shows when every sentence runs the same length. Rhythm and repetition across a whole page still need a read-through.
Sentences that all run the same length
AI drafts often keep every sentence at about the same middling length, each with the same build: subject, verb, a clause, a full stop. Any one sentence is fine. A page of them reads flat.
Before: Our team reviewed the quarterly results in detail. The numbers showed steady growth across all of our regions. Customer feedback was positive in most of the areas we measured. We plan to build on this progress in the next quarter.
After: Our team reviewed the quarterly results in detail. Every region grew steadily. Customer feedback was positive in most of the areas we measured, and we plan to build on this progress in the next quarter.
The facts are the same. One sentence is cut down to its point, and the last two are joined. Mix a short sentence in where you want a point to land, and let a longer one carry detail. The Sentence rhythm panel in the humanizer shows your sentence count, the average length and whether the variety is low, medium or high.
Groups of three
Three adjectives, three verbs, three bullet points, again and again: “fast, secure and reliable”, “collaborate, innovate and grow”. Real points rarely come in threes every time.
Before: Our platform is fast, secure and reliable, helping teams collaborate, innovate and grow.
After: Pages load in under a second, and every file is encrypted.
Keep the one or two qualities that matter and back each with a fact. If you do not have the fact, the word was padding.
“Not X but Y” contrasts
“It’s not just a tool, it’s a revolution.” “This isn’t about speed. It’s about trust.” The pattern sets up a claim nobody made so it can knock it down. Its cousin is “No fluff, no jargon, just results.”
Before: This isn’t just a budgeting app. It’s a whole new way to think about money.
After: This budgeting app sorts your spending into categories as you go.
Say what the thing is. Keep a contrast only when someone really does believe the first half.
Tacked-on endings and dramatic reveals
Watch for sentences that end in a comment starting with an -ing word, such as “highlighting its commitment to” or “ensuring a smooth experience”, and for set-ups such as “The result?” or “The best part?” before a plain fact. “From small startups to global brands” is another stock sweep that names no one.
Before: The city opened three new parks in 2025, highlighting its commitment to green space. The result? Visits to public parks rose by a third.
After: The city opened three new parks in 2025, and visits to public parks rose by a third.
Overuse of the long dash
AI drafts lean on the long dash (the em dash) to bolt asides and punchlines onto sentences, often several times a paragraph: an aside wrapped in a pair of long dashes in the middle, then another before a reveal at the end. Editors and novelists have used the long dash for centuries, so a single one means nothing. One in most sentences is what readers notice.
This sign is fading. Many newer AI writing tools avoid the long dash entirely, so a text without one tells you nothing. Drafts from older tools, which are still in wide use, often have one in most sentences.
Use commas for an aside, a colon before an explanation, parentheses for a true side note, or a full stop. The Fix long dashes button in the humanizer swaps em and en dashes for commas, “to” in number ranges or hyphens, on your device, and Undo puts them back.
Emoji bullets, bold labels and headings
Formatting gives drafts away as often as words do:
- an emoji at the start of every bullet
- every bullet opening with a bold label and a colon
- headings in Title Case, or a heading with a colon and a subtitle, such as “Email Marketing: A Complete Guide”
- headings that start with “Unlocking”, “Mastering” or “Navigating”
- arrows in place of words, and a horizontal rule between every section
- bold scattered through every paragraph
Use plain bullets or full sentences, write headings in sentence case that say what the section covers, and bold one thing at most.
Signs of AI writing in essays, resumes, emails and posts
The same habits show up everywhere, but each kind of text has its own favorites.
Academic writing and essays
Common AI words in academic writing are “delve into”, “intricate”, “pivotal”, “underscores”, “showcasing”, “a nuanced understanding” and “plays a crucial role”, plus the lead-in “It is important to note that”. Structural signs matter more: every paragraph built the same way (topic sentence, three points, a one-line summary), “studies show” with no citation, a balanced view that never takes a position, and a conclusion that repeats the introduction.
State your argument in the first paragraph. Back each claim with a named source you have read. Cut the summary line at the end of each paragraph, and make the conclusion say something the introduction did not, such as what your argument means or what is still open. If your course or journal has rules on AI tools, follow them, and keep your notes and drafts: they show how you worked far better than any word list.
Resumes, CVs and cover letters
AI words to avoid in a resume or CV are the inflated verbs and adjectives: leverage, utilize, facilitate, streamline, foster, elevate, robust, seamless, meticulous and cutting-edge. Cover letters drafted with AI often open with “I am thrilled to apply” and close with “I am confident that my skills align perfectly with your needs”.
Before: Leveraged cutting-edge tools to streamline processes and enhance team efficiency.
After: Cut invoice approval time from 6 days to 2 by moving approvals into one shared tracker.
The after needs your real numbers, which is exactly why it reads as yours. In a cover letter, name one thing you did that matches the job and one reason you want this job in particular.
Emails
AI-drafted emails stack politeness: “I hope this email finds you well”, “I wanted to reach out regarding”, “Please don’t hesitate to reach out” and “Thank you for your time and consideration”, often in one short message. Here is the same email with the details it was hiding:
Before: I hope this email finds you well. I wanted to reach out regarding the upcoming project deadline. Please don’t hesitate to reach out if you have any questions. Thank you for your time and consideration.
After: Can we move the report deadline from Friday to Tuesday? The sales figures arrive on Monday. Just reply if Tuesday doesn’t work for you.
Lead with the request and the reason, and thank people once.
Social posts
AI-written posts open with a hook such as “Here’s the thing” or “I’m thrilled to announce”, run one sentence per line, list points with emoji bullets, add “Let that sink in” and end with “Agree?” and a row of hashtags. Lead with the news, write in short paragraphs, ask a question you want answered, and keep two or three hashtags people search.
How to fix AI-sounding writing
Work through a draft in this order. Cutting comes before swapping, because there is no point polishing a sentence you will delete.
- Run the checks. Paste the draft into the AI humanizer and press Check phrases. It marks most of the words and phrases in the tables above, sorted into groups such as Overused words and Sentence habits, with a plain swap for each. Weaker signs, such as “Notably” or “the power of”, are marked only when they come up 3 or more times. The check runs on your device with no daily limit, and your text is not sent. The list covers English for now.
- Cut. Delete lead-ins, signposts such as “In this article”, stacked connectors and recap paragraphs. Most drafts get shorter with nothing lost.
- Swap. Replace inflated words with plain ones, but only where the plain word says the same thing.
- Put the specifics back. Numbers, names, dates and examples are what make writing sound like a person who knows the subject. No tool can add facts only you have.
- Vary the rhythm. Split a long sentence, join two short ones, and check the Sentence rhythm panel.
- Fix the formatting. Turn emoji bullets and bold labels into plain bullets, and use Fix long dashes if every sentence has one.
- Read it aloud. Anything you would not say to a colleague is worth another look.
Here is one paragraph taken through those steps:
Before: In today’s fast-paced digital landscape, effective communication plays a crucial role in fostering seamless collaboration. Additionally, it’s important to note that clear emails can significantly enhance productivity, streamline workflows and empower teams to unlock their full potential. In conclusion, prioritizing clarity is not just a best practice but a necessity.
After: Clear emails save everyone time. When a message says what you need and by when, people can answer it once instead of writing back with questions.
The opener, the lead-in, the three-part list, the “not just X but Y” and the recap went. What was left was one claim, and the after says what clarity does instead of calling it crucial.
When a draft needs more than spot fixes, press Rewrite in the same tool. You get 3 free rewrites a day of up to 1,000 words each (fewer in some languages), with no sign-up, in a Natural, Simple, Professional or Casual voice. The rewrite is built to keep your numbers, names, quotes, links and formatting, and the page lists any number, link or quote missing from the result. Your text is sent only when you press Rewrite, and we keep neither it nor the result (the details are on our privacy page). Read the rewrite, check the facts and make it sound like you.
If the text you want to check is in a picture, such as a screenshot of a post or a photo of a printed page, copy the words out first with screenshot to text or image to text.
A prompt to avoid these signs
If you draft with an AI writing tool, you can ask for fewer of these signs up front. Add a style note like this one to your request, and change it to suit your text:
Write in plain English for a general reader. Prefer short, common words, and avoid delve, tapestry, testament, pivotal, crucial, seamless, robust, leverage, foster, showcase and landscape. Start with the first real point, with no lead-in such as “In today’s world” or “It’s important to note”. Do not end with a recap or “In conclusion”. Vary sentence length and keep some sentences short. Use a contrast such as “not just X but Y” only when the contrast is real, and list three things only when there are three. Do not use the long dash; use commas, colons or full stops instead. Use plain bullets with no emoji and no bold labels. Keep every fact, number and name I give you, and add none of your own.
Treat the result as a first draft. Tools follow style notes loosely, and a word you ruled out can come back as a close synonym, so still run Check phrases and read the draft yourself. A style note is a writing aid, not a way past AI detectors, and we make no promise about what any detector will say.
Checklist for AI-sounding writing
Run through this before you send or publish:
- The first sentence says something specific about this topic.
- No sentence opens with a lead-in such as “It’s important to note”.
- Inflated words are swapped wherever a plain word says the same.
- Every claim that something matters says why, or what happens without it.
- Sentences vary in length, and some are short.
- Lists have as many items as the point needs, not always three.
- Contrasts such as “not just X but Y” are there only when the contrast is real.
- Long dashes are rare, and none bolts a punchline onto a sentence.
- Bullets are plain, headings are in sentence case, and bold is used once at most.
- The end makes a final point instead of repeating the middle.
- Sources are named, not “experts say”.
- No leftover chat lines, placeholders or citation codes remain.
What these signs cannot tell you
These signs are not proof. Apart from the leftover chat lines, the words and phrases in the tables above were all in use long before AI chat tools. Journalists use the long dash, speechwriters teach the rule of three, and the formal style taught in schools and offices shares many of these habits. So does writing by people working in their second or third language.
AI text can also lose every sign. A short instruction or a quick edit removes most of them, so a clean text says nothing either way.
AI detectors do not settle the question either. A test of 14 detection tools found that none was accurate or reliable, and that simple changes to the text made them worse (Weber-Wulff et al.). Another study found that detectors consistently misjudged essays by non-native English writers as AI-generated while judging native speakers’ essays correctly (Liang et al.).
So judge clarity, not origin. Ask whether a text says something specific, whether it is accurate, and whether any sentence could go without the reader missing it. Those questions improve any draft, whoever wrote it. If you need to know how a piece was written, talk to the writer and look at their notes and earlier drafts. The checks in our humanizer are writing checks in the same spirit: they count matches and give no score.
Pictures have their own clues and, often, better evidence. See how to tell if an image is AI-generated, or read the labels a file carries with the AI image checker. Some AI tools can also hide an invisible statistical watermark in the text they write, which no reader can see in the words; what SynthID is explains how that works.
Check a draft for these signs
Paste a draft into the free uncrop.si AI humanizer and press Check phrases: it marks stock words and phrases by group, with a plain swap for each, on your device and with no daily limit.
Questions
What are the most common AI words?
The words AI uses far more often than most writers include delve, tapestry, testament, realm, pivotal, crucial, seamless, robust, leverage, foster, underscore, showcase, intricate, meticulous, vibrant, multifaceted, and landscape in phrases such as "the digital landscape". The AI words list in this guide gives a plain alternative for each. One of them in a page means little; several in one paragraph is what readers notice.
What AI words and phrases should you avoid in writing?
Avoid the ones that add nothing: lead-ins such as "It's important to note that", openers such as "In today's fast-paced world", "Moreover" at the start of every paragraph, recaps such as "In conclusion", and inflated verbs such as "leverage" and "elevate" where "use" and "improve" say the same. You do not need to ban words. Cut or swap them when a plain word does the job.
What are common AI words in academic writing and essays?
In research abstracts, style words such as "delves", "showcasing" and "underscores" rose sharply after AI chat tools spread, according to a study of over 15 million abstracts. In essays, watch for "a nuanced understanding", "plays a crucial role", "It is important to note", "studies show" with no citation, and a conclusion that repeats the introduction. Follow your course's rules on AI tools, and keep your notes and drafts.
What AI words should I avoid in a resume or CV?
Leverage, utilize, facilitate, streamline, foster, elevate, robust, seamless, meticulous and cutting-edge, especially when several share one bullet. Replace them with a plain verb, what you did and a number: "Cut invoice approval time from 6 days to 2" says more than "Streamlined processes to enhance efficiency".
Can you prompt an AI writing tool to avoid these signs?
Partly. A short style note helps: ask for plain words, varied sentence lengths, no lead-ins or recaps and no long dashes, and tell it to keep your facts and add none. This guide has a style note you can copy. Tools follow such notes loosely, so check the draft and edit it yourself afterwards. A style note is a writing aid, not a way past AI detectors, and we make no promise about what any detector will say.
Is the long dash a sign of AI writing?
Only as a pattern, and mostly in drafts from older tools. Older AI writing tools often use the long dash (the em dash) several times a paragraph, but many newer ones avoid it entirely, so a text without one tells you nothing. Writers and editors used it long before AI, so one dash proves nothing either. If your draft has one in most sentences, swap some for commas, colons or full stops. The Fix long dashes button in our AI humanizer does this on your device, and Undo puts them back.
Can AI detectors tell if text was written by AI?
Not reliably. A test of 14 detection tools found none of them accurate or reliable, and a separate study found that detectors consistently misjudged essays by non-native English writers as AI-generated. Treat the signs in this guide as editing notes, not evidence, and judge a text by whether it is clear and accurate.
Sources
The research we drew on, checked on 28 October 2026.
- Kobak et al., Science Advances (2025): excess vocabulary in over 15 million biomedical abstracts
- Weber-Wulff et al., International Journal for Educational Integrity (2023): testing of detection tools for AI-generated text
- Liang et al. (2023): AI text detectors misjudge writing by non-native English speakers
- Wikipedia: Signs of AI writing, a field guide kept by Wikipedia editors
Free tools for this
Related guides
AI images
How to tell if an image is AI-generated
Five steps that work: get the original file, read its labels, check for SynthID, look for visual clues and trace the source. What each test can prove.
AI images
Content Credentials (C2PA) explained: how to verify where an image came from
C2PA Content Credentials are a signed record of the app, AI or camera that made an image. What they show, valid vs trusted, and how to verify them free.