A likelihood, not a verdict
The percentage is how likely Lantern thinks it is that an AI model wrote the text. Under 35% reads as likely human-written, and 65% and above as likely AI-generated.
Paste a post, an essay or a review, and Lantern estimates how likely it is that an AI model wrote it, and how sure it is.
Lantern needs at least 25 words, and its answers get firmer past 150.
You can check 3 texts and 1 image a day here for free, without an account.
The score card appears here. It shows how likely it is that the text was written with AI, how sure Lantern is, and what stood out.
You get a likelihood, how sure Lantern is of it, and what stood out. It’s the same score card the extension shows beside what you read.
The percentage is how likely Lantern thinks it is that an AI model wrote the text. Under 35% reads as likely human-written, and 65% and above as likely AI-generated.
Confidence says how much evidence the estimate rests on. Short passages can’t support a firm answer, so Lantern holds its confidence back below 150 words.
The card lists what moved the score, such as patterns common in AI writing tools, and says so when the passage is too short to tell much.
No AI detector is always right. These are the cases where Lantern’s text scores deserve the most caution.
More answers about scores, the extension and plans are on the support page.
Yes. Anyone can check 3 passages and 1 image a day here, without an account. The Lantern extension, which checks the text and images you come across as you browse, takes a plan from $9 a month.
It’s Lantern’s estimate of how likely it is that an AI model wrote the text, from 1% to 99%. It never says 0% or 100%, because no score is proof. Under 35% reads as likely human-written, 65% and above as likely AI-generated, and anything in between as mixed signals.
No AI detector is always right, and Lantern doesn’t pretend otherwise. It gives a likelihood rather than a yes or no, says how sure it is, and is most reliable on longer passages of everyday English prose. Treat a high score as a reason to look closer, never as proof.
At least 25 words. With less, there isn’t enough to judge, so Lantern says so instead of guessing. Confidence stays low under 60 words and moderate under 150, so a few paragraphs give the firmest answer. You can paste up to 8,000 characters.
A language model trained on large collections of human and AI writing reads the passage as a whole and estimates how likely it is to be machine-written. Lantern calibrates that estimate, so a percentage means the same thing from one check to the next, and works out how much the passage’s length lets it trust the answer.
Not reliably. Human writing polished by AI, and AI writing that someone reworked, often scores in the middle. A mixed result can be a sign of exactly that kind of shared authorship.
Only as one input among several. Writing by non-native English speakers and very formulaic writing can look machine-made, and a score is never proof. Never use it as the only basis for a decision about a person.
No. The text is checked and then discarded. Lantern keeps the result and counts the check toward the day’s free checks, and never stores what you pasted unless you choose to share a passage of it.
The Lantern extension checks the posts, reviews and photos you come across and leaves a quiet score beside each one. Plans start at $9 a month, and you choose how many checks you need.