AI content detector

Shows how human your email, post or article reads, with every AI tell highlighted and a plainer line for each. It reads the text, never who wrote it.

Cold emailHi Jane, I hope this message finds you well. I came across Northwind's...

110 words, 13 AI tells, sample text

Content type

Emails, posts, articles or pages. The rules change with the content type.

Sample text, checkedGuessed: Cold email rulesRules run in your browser, the AI check runs only if you ask

17human

Reads as AI

80 and up reads human, under 50 reads as AI

Cold email110 wordsVaried rhythm

5 AI verbs, 2 stock phrases, 2 AI adjectives, 2 hype words, 1 AI opener and 1 AI pattern. Fix the red marks first: dead phrases and stock openers cost the most, because readers spot them first. The score reads patterns in the text, never who wrote it.

12 phrases1 pattern12 sentences

Where it reads as AI

Numbers in the text match the notes. Each note has a plainer way to say it.

Your cold email

13 marks

Hi Jane,

I hope this message finds you well. I came across Northwind's recent expansion and was truly impressed by your innovative approach to logistics.

In today's fast-paced world, staying ahead isn't just about working harder - it's about working smarter. That's where Outbound Labs comes in. We leverage cutting-edge AI to streamline your outreach, boost engagement, and unlock new growth.

Our seamless platform empowers teams to connect with the right prospects at the right time. It's a game-changer for companies looking to elevate their pipeline.

Would you be open to a quick call to explore how we can help Northwind achieve its goals?

Looking forward to connecting!

Best regards,
Tom

Costs the mostCosts pointsThe rules run in your browser.

1“I hope this message finds you well” Dead phrase-10

A stock line from thousands of templates. Readers skip it on sight. Delete it and open with the reason you're writing.

2“I came across” AI opener-8

The opener models reach for first. It reads as generated personalization. Say what you saw: "Saw your post on corridor pricing"

3“innovative” AI adjective-5

Puffed-up adjective that models overuse. Say the concrete thing instead. Drop it, or say the concrete property.

4“In today's fast-paced world” Dead phrase-10

A stock line from thousands of templates. Readers skip it on sight. Start with a specific fact about the reader.

5"Not just X, but Y" Pattern-8

The sound of a model reaching for emphasis it hasn't earned. Say the Y part on its own.

6“leverage” AI verb-5

One of the words that spiked in writing after ChatGPT. A plain verb reads human. "use"

7“cutting-edge” Hype word-6

Marketing hype. It reads as a brochure, and some filters score it. Say what it does, with one number.

8“streamline” AI verb-5

One of the words that spiked in writing after ChatGPT. A plain verb reads human. Say what gets faster or cheaper, and by how much.

9“unlock” AI verb-5

One of the words that spiked in writing after ChatGPT. A plain verb reads human. "get" or "open up"

10“seamless” AI adjective-5

Puffed-up adjective that models overuse. Say the concrete thing instead. Drop it, or say the concrete property.

3 more marks in the text.

Sentence rhythm

Words per sentence, in order. People mix short and long. Models even them out.

9.2 words a sentence on averageVariation 0.65: varied, reads human

Second opinion

Ken AI

Ken AI reads the whole text, scores how human it sounds, and rewrites the lines with the most tells.

Check it as another kind of text

Each content type has its own rule set. Pick one to re-check the same text.

Cold email grader

Then

Spam word checker

Email me this report

The score, every AI tell with a plainer alternative and, if you ran it, Ken AI's rewrites. Your full text isn't included.

Every tell with its fixKen AI rewrites includedYour text is never stored

Ken writes cold email that reads like a person.

Every draft runs against these rules, and the lines that read as AI get rewritten before anything sends.

Start free - 1,000 credits, no card

What the AI content detector looks for

Paste text and pick what it is: a cold email, a LinkedIn post, a blog article or one of the other content types. The detector highlights the patterns that make writing read as machine-made and offers a plainer version of each.

Stock phrases: "I came across", "I noticed that", "caught my eye", "quick question", "circling back", "just wanted to".
Hype and puffery: "game-changing", "cutting-edge", "world-class", "revolutionary", "in today's fast-paced world".
Constructions: em dashes, "not just X, but Y", lists of three used more than once, and stock closers like "in conclusion" in articles and marketing copy.
Rhythm: sentences that all run the same length. People write a short line, then a long one. Models tend to even them out.

Each content type has its own rule set, so a LinkedIn post isn't judged like a press release.

How to read the result

The meter shows how human the text reads. Each highlight is one tell, with the reason and an alternative. Fix the red marks first. Dead phrases and stock openers cost the most, because readers spot them in the first line. Sentence rhythm is the last thing to tune.

The result says "reads as AI", never "written by AI". It measures patterns in the text and can't know who or what wrote it. A person who writes in corporate filler will score low, and an AI draft that someone edited hard will score high. That's the right outcome, because your reader judges the text, not where it came from.

How to make AI-assisted writing sound human

Cut the opener. Start with the thing you came to say.
Replace every word of praise with the fact behind it. "Impressive growth" becomes "you doubled the team since March".
Break the rhythm. Follow a long sentence with a three-word one.
Delete the last paragraph if it only restates the first.
Swap em dashes for a period or a comma.
Read it out loud. Anything you wouldn't say to a colleague goes.

The optional AI pass adds a second-opinion score and suggests plainer versions of up to four of the lines with the most tells.

Why sounding like AI costs replies

Buyers have read enough machine-written outreach to know the tells. An email that opens with "I came across your profile and was impressed" gets filed as automated before the second line. On LinkedIn the same patterns make a post feel generic, and it gets scrolled past.

The rules here are the writing rules Ken applies to every campaign email it drafts. Running your draft through them takes a few seconds and needs no account.

Questions

Can an AI detector prove who wrote a text?

No, and no tool can. Detectors measure patterns, and people use the same patterns. OpenAI shut down its own AI text classifier in 2023 because of its low accuracy. This tool tells you how much your text reads as AI and shows why, so you can fix it. Don't use it, or any detector, to accuse someone of using AI.

What are the most common signs of AI writing?

Stock openers ("I came across", "I hope this finds you well"), hype words ("game-changing", "cutting-edge"), the "not just X, but Y" construction, lists of three, em dashes, and closers like "in conclusion". Even sentence length is another tell. People vary their rhythm more than models do.

Does it work for LinkedIn posts and blog articles?

Yes. Pick the content type and the rules change with it, because the tells differ. A cold email gives itself away in the opener and the ask. An article does it in the intro and the summary at the end. If you don't pick a type, the detector guesses from the text and you can change it.

Is my text stored?

The rules run in your browser, so your text doesn't leave the device for the main result. The optional AI pass sends the text to a model to score it and suggest rewrites. The tool doesn't keep it afterwards. If you email yourself the report, it lists the flagged phrases, not the full text.

How do I make ChatGPT writing sound less like AI?

Give the model one real fact to build on, then edit what it writes. Delete the opener and the closing summary, replace praise with specifics, swap em dashes for periods, and vary sentence length. Then paste the result here to catch what you missed. Adding "sound human" to the prompt rarely does it on its own.