Paperpal AI Detector Review | Accuracy Report
  1. Home »
  2. Blog »
  3. Paperpal

Paperpal AI Detector Review

4 min read

Paperpal (paperpal.com) is an AI writing assistant launched by Cactus Communications in September 2022.[1] Catering to academics, Paperpal offers a variety of AI writing tools including AI detection, grammar checking, proofreading, and manuscript reviews.

This is a review of Paperpal’s AI detector only and is not a reflection of their other services.

Pros

  • Sentence-by-sentence breakdowns
  • Easy-to-use, well-designed interface
  • Detailed reports

Cons

  • 1/7 against LLMs
  • 2/8 against humanizers
  • 2 false positives found

Features

  • AI detector: Identifies non-human writing.
  • Plagiarism checker: Recognizes unoriginal writing.
  • Writing assistant: Generates content and offers suggestions.
  • Grammar checker: Reviews writing for spelling and grammar errors.
  • Translator: Provides AI translating services.
  • Essay writer: Drafts outlines and assists.
  • Paper checker: Reviews and improves uploaded manuscripts.
  • Chrome extension
  • MS Word Add-on 
  • Google Docs Add-on

Pricing Plans

Free Prime
No credit card required. $25/month or $97.30/year
5 AI detections/day (1.2k words/scan) Unlimited detections (10k words/scan)
Limited access to AI assistant, editor, chat, plagiarism, research. Full access to all AI writing features including faster speeds for suggestions.

Test #1: Paperpal vs. LLMs

Failed 6 out of 7 tests against leading LLMs, while only detecting 63% AI in the one successful scan.

For our first test, we ran text written by various AI models through Paperpal’s detector to see if it could detect basic AI writing. We enter the prompt below into each LLM to generate a scannable article.

Please create a 500-word article about the invention of the wheelbarrow that is 100% factually correct. Remove all em dashes and ensure that the text reads as human as possible in order to pass AI detectors.

Test Results
Expand

LLM Test Result Proof
ChatGPT – 5.4 Thinking Model

(generated text)

38% AI

Claude – Sonnet 4.6 Extended

(generated text)

63% AI

Google Gemini – Thinking

(generated text)

 

35% AI

Deepseek – Deepthink

(generated text)

20% AI

Perplexity – Free Model

(generated text)

34% AI

Grok – Auto

(generated text)

39% AI

Meta

(generated text)

1% AI

Test #2: Paperpal vs. Humanizers

75% failure rate against humanizers – Only 2 humanized articles scanned at over 50% AI writing.

For our second test, we took the only AI article that didn’t beat Paperpal’s detector and ran it through eight different AI humanizers. The humanized articles were then entered into Paperpal’s detector to see what it could pick up.

Test Results
Expand

AI Humanizer Test Result Proof
Humanize.ai
(humanized text)
20% AI

StealthGPT
(humanized text)
14% AI

WriteHuman.ai
(humanized text)
6% AI

Undetectable.ai
(humanized text)
56% AI

🟡

StealthWriter.ai
(humanized text)
11% AI

WalterWrites.ai
(humanized text)
53% AI

🟡

Grammarly
(humanized text)
38% AI

Quillbot
(humanized text)
7% AI

Test #3: False-Positives + Plagiarism

2 false positives out of 10 human written articles and 2 missed instances of plagiarism.

A false positive in this case is when Paperpal detects AI writing in 100% human-written articles. Ten human-written excerpts from various online articles and books were used for this test. We also tested Paperpal to see if it could pick up plagiarism.

Test Results
Expand

Article URL Test Result AI Detection Plagiarism
War and Peace (excerpt)

(view article)

3% AI 94% Plagiarized

Self-Reliance (essay)

(view article)

1% AI 100% Plagiarized

Nat. Geo (article)

(view article)

7% AI 100% Plagiarized

Wikipedia (“Ralph Waldo Emerson”)

(view article)

2% AI 74% Plagiarized

NBC (article)

(view article)

22% AI 79% Plagiarized

Reuters (article)

(view article)

2% AI 92% Plagiarized

Politico (article)

(view article)

30% AI 26% Plagiarized

New York Post (article)

(view article)

38% AI 6% Plagiarized

Johns Hopkins (article)

(view article)

71% AI 90% Plagiarized

NIH (article)

(view article)

100% AI 100% Plagiarized

Is Paperpal Accurate?

Paperpal has good reviews on Trustpilot, with an overall score of 4.2 stars. However, a lot of the reviews are evaluating quality of the research assistant. Some complain about the accuracy of the AI detector and how it will flag their human writing as AI.

Frequently Asked Questions

Are sentence-by-sentence breakdowns provided?

Yes, completed scans provide a granular look at which sentences read as AI, human, and mixed.[2]


Is document scanning included?

Yes, documents can be uploaded in .xlsx, .pdf, and .docx formats.[3]


Is image scanning provided?

Yes, images can be uploaded and scanned in .png, .jpg, .jpeg, .gif, .webp formatting.[4]


Are reports shareable?

No, reports aren’t directly shareable from a Paperpal account but they can be downloaded immediately in PDF format and shared as needed.


Does Paperpal provide certifications?

No, there’s no certifications attached to the detection results guaranteeing their accuracy. In fact, the results include a disclaimer that states “Results are estimates of writing patterns, not proof of AI use.


Are AI detections used for internal training?

No, Paperpal states explicitly that their AI models are trained on independent data sets, so AI detections and other uploaded documents are not used for internal training.[5]


Comments 0 comments