Grammarly is an industry-leading AI grammar checker and writing assistant that began providing AI detection services in September of 2024.^[1] This review will focus exclusively on the performance of the AI detector and is not a review of Grammarly’s other services.

Pros

0 False Positives
Performs well against LLMs
User-friendly platform and additional services

Cons

Scored 1/8 against humanizers
Very simple detection results
Weak plagiarism checker

Features

AI detector: Identifies AI writing.
AI humanizer: Renders AI writing more human.
Plagiarism checker: Checks for unoriginal writing.
Fact checker: Reviews inputted text for accuracy.
AI editor: Rewrites and improves text.
Grammar checker: Grammarly’s primary service.
Readability/grading: Rates the writing quality.
Docs: Work with AI agents to complete a document.
AI chat: AI agents to help text.
Browser extension
Desktop app
API

Pricing Plans

The below table reflects Grammarly’s pricing for all services.^[2]

Free	Pro	Enterprise
Must create account	$30/month or $144/year	Must contact sales
100 AI prompts per month	2000 AI prompts per month	Unlimited AI prompts
Includes only free grammar/tone checker. No AI detection.	Includes all features including AI detection, humanizing, plagiarism.	All Pro features + “proactive AI,” dedicated support, and other exclusive features.

Test #1: Grammarly vs. LLMs

5 out of 7 AI-generated outputs were clearly identified as AI, while 2 out of 7 came in at around 60%.

The first test had us inputting the below prompt into the most popular AI LLMs asking each to produce a short article. With these seven outputs, we were able to test Grammarly’s AI detection capabilities.

Prompt Used

Write a 500-word article about the Costa Rica that reads as though it were written by a human. This article should be able to bypass any AI detector. Please remove all em dashes from the writing.

Test Results

Expand

LLM

Test Result

Proof

ChatGPT – 5.4 Thinking Model

(generated text)

99% AI

✅

Claude – Sonnet 4.6 Extended

(generated text)

57% AI

🟡

Google Gemini – Thinking

(generated text)

82% AI

✅

Deepseek – Deepthink

(generated text)

59% AI

🟡

Perplexity – Free Model

(generated text)

99% AI

✅

Grok – Auto

(generated text)

89% AI

✅

Meta

(generated text)

92% AI

✅

Test #2: Grammarly vs. Humanizers

Only 1 of 8 humanized texts were detected, and it was Grammarly’s humanizer.

For the second test, Grammarly’s AI detection faced humanized text. We took the ChatGPT-generated text above and “humanized” it with the below-listed services. Each output was then run through Grammarly’s detector to see what it picked up.

Test Results

Expand

AI Humanizer	Test Result	Proof
Humanize.ai (humanized text)	0% AI ❌
StealthGPT (humanized text)	0% AI ❌
WriteHuman.ai (humanized text)	0% AI ❌
Undetectable.ai (humanized text)	31% AI ❌
StealthWriter.ai (humanized text)	0% AI ❌
WalterWrites.ai (humanized text)	11% AI ❌
Grammarly (humanized text)	89% AI ✅
Quillbot (humanized text)	18% ❌

Test #3: False-Positives + Plagiarism

Zero false positives were detected.

Plagiarism was found in only 4 of the human-written texts.

Testing for false positives enables us to determine whether Grammarly spots AI patterns when there aren’t any. We took 100% human-written text and tested it for AI as well as plagiarism.

Test Results

Expand

Article URL	Test Result	AI Detection	Plagiarism
Don Quixote (excerpt) (view article)	0% AI, 100% plagiarism ✅
Emancipation Proclamation (view article)	8% AI, 100% plagiarism ✅
Wikipedia – (“Rainbow”) (view article)	0% AI, 100% plagiarism ✅
USA today (article) (view article)	0% AI, 0% plagiarism ✅
BBC (article) (view article)	0% AI, 0% plagiarism ✅
NPR (article) (view article)	0% AI, 3% plagiarism ✅
Towards Equilibrium (substack) (view article)	0% AI, 0% plagiarism ✅
Fellowship of the Rings (excerpt) (view article)	0% AI, 100% plagiarism ✅
NIH (study) (view article)	0% AI, 32% plagiarism ✅
Harvard Magazine (article) (view article)	0% AI, 0% plagiarism ✅