> For the complete documentation index, see [llms.txt](https://jia-mei-he-shun.gitbook.io/lunote-ai-humanizer-handbook/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://jia-mei-he-shun.gitbook.io/lunote-ai-humanizer-handbook/ai-detector-handbook.md).
# AI Detector Handbook
Choose an AI detector, understand what its score means, compare free and paid plans, and build a review workflow that keeps a human decision in the loop.
## Introduction
### Why should we detect AI writing?
AI-assisted writing is now common in education, publishing, marketing, hiring, and internal business work. The useful question is rarely just “Was AI involved?” A reviewer usually needs to know whether the writing follows the relevant policy, whether the claims are supported, whether the author can explain the work, and which passages need closer review.
An AI detector can help locate patterns that resemble model-generated writing. It can also help an author self-check a draft, help an editor focus a review, or help a teacher decide which questions to ask. It cannot read intent, prove misconduct, verify every source, or reconstruct the writing process.
{% hint style="info" %}
**Core rule:** treat every AI detector result as a probability signal, not a verdict.
{% endhint %}
### What does this handbook do?
* Explains what an AI detector measures and why different tools disagree.
* Compares free access, paid plans, result clarity, integrations, and practical limits.
* Reviews 10 mainstream AI detector products from a user’s perspective.
* Shows how to interpret representative reports without treating a score as proof.
* Provides workflows for students, educators, publishers, teams, and multilingual reviewers.
### What this handbook does NOT do
* It does not promise that any detector is 100% accurate.
* It does not treat AI detection as the same thing as plagiarism detection.
* It does not recommend punishment, rejection, or publication decisions based on one score.
* It does not rank products only by vendor accuracy claims.
***
## Part I: Understanding AI Detector
### Chapter 1 | What Is an AI Detector?
### 1.1 From yes/no judgment to probability evidence
An AI detector analyzes language patterns and estimates how closely a passage resembles text produced by a large language model. Depending on the product, a report may include:
* an overall AI probability or percentage;
* human-written, mixed, or AI-assisted categories;
* sentence- or passage-level highlights;
* explanations such as repetition, predictability, sentence rhythm, or low variation;
* supporting workflow evidence such as writing history, report export, or plagiarism checks.
The score describes similarity to patterns in the detector’s model. It does not identify the writer’s intent and does not prove who typed the words.
### 1.2 Why AI detectors disagree
Two detectors can give different results because they use different training data, model versions, thresholds, language coverage, text-length requirements, and definitions of “AI-assisted.” A small edit can also change the signal distribution.
Disagreement is not automatically evidence that one tool is broken. It is a reason to inspect the highlighted passages, confirm that the sample is long enough, check the supported language, and compare the score with real process evidence.
### 1.3 What AI detectors can help with
AI detectors are most useful for:
1. screening drafts before submission or publication;
2. locating passages that need closer human review;
3. comparing an original draft with a revised version;
4. combining AI signals with sources, citations, plagiarism checks, and writing history;
5. building a consistent review workflow for a team.
They are least useful when a reviewer submits a short passage, ignores the detector’s language limits, or treats one percentage as a final decision.
### Chapter 2 | Common Problems and Solutions of AI Detector
### 2.1 Six common problems of AI Detector
#### Problem 1: False positives
Formal, repetitive, technical, or highly edited human writing can resemble AI-generated text. A false positive becomes harmful when the score is treated as proof instead of a prompt for review.
{% hint style="info" %}
**Safer response:** inspect the highlighted text, ask for drafts or notes when appropriate, and consider the consequences before making a decision.
{% endhint %}
#### Problem 2: Model drift
AI writing models, paraphrasers, and humanizers evolve. A detector that performed well on one model or benchmark may behave differently on newer outputs.
{% hint style="info" %}
**Safer response:** use current product versions and maintain a small internal test set that reflects your actual content.
{% endhint %}
#### Problem 3: Mixed authorship
A document may contain original writing, AI-generated sections, copied quotations, grammar corrections, or human revisions. A single document-level score can hide that mixture.
{% hint style="info" %}
**Safer response:** prefer sentence or passage highlights and review the writing process, not only the total.
{% endhint %}
#### Problem 4: Short samples
Short text gives a detector fewer patterns to analyze. Headlines, captions, bullet lists, and short answers are especially unstable.
{% hint style="info" %}
**Safer response:** use the minimum length recommended by the product and, when possible, evaluate a coherent section rather than an isolated sentence.
{% endhint %}
#### Problem 5: Language gaps
Multilingual support is not the same as equal performance in every language. Results can vary with translation, regional vocabulary, non-native writing, and code-switching.
{% hint style="info" %}
**Safer response:** test each language separately with known human and AI samples before using a detector operationally.
{% endhint %}
#### Problem 6: Score confusion
An AI probability is not a plagiarism score, authorship certificate, policy decision, or explanation of intent.
{% hint style="info" %}
**Safer response:** state what the score represents, record which tool produced it, and require a human reviewer to make the final decision.
{% endhint %}
### 2.2 User Rule of Thumb for 2026
{% stepper %}
{% step %}
#### Start with explainable results
Start with a detector that explains where to look, not only a total score.
{% endstep %}
{% step %}
#### Use coherent passages
Use longer, coherent passages whenever possible.
{% endstep %}
{% step %}
#### Add supporting evidence
Combine detection with drafts, notes, citations, source checks, and policy context.
{% endstep %}
{% step %}
#### Review data handling
Do not submit confidential text until you have reviewed the tool’s data-handling terms.
{% endstep %}
{% step %}
#### Test each important language
Test every important language separately.
{% endstep %}
{% step %}
#### Evaluate before buying
Before purchasing an API or institution plan, evaluate 50–100 representative internal samples.
{% endstep %}
{% step %}
#### Document the decision
Document the human decision and the evidence used to reach it.
{% endstep %}
{% endstepper %}
***
## Part II: Best AI Detector Practices Guide
### Chapter 3 | What AI Detector Products Are Free?
“Free” can mean a public text box, a daily quota, a monthly word allowance, a short trial, or a limited report. Before choosing a product, check the sample length, upload support, report detail, account requirement, and what happens after the free allowance ends.
### 3.1 User-oriented detector categories
#### Self-check and quick review
Prioritize a fast input flow, readable score categories, sentence-level highlights, and a clear next action. Lunote and QuillBot are designed around this kind of immediate review.
#### Education and writing-process review
Prioritize writing history, class workflows, downloadable reports, and careful false-positive guidance. GPTZero, Pangram, and education-focused Copyleaks plans are relevant candidates.
#### Publishing, SEO, and agency review
Prioritize plagiarism checking, source review, readability, scan history, site scanning, and editorial collaboration. Originality.ai and BrandWell are positioned for these workflows.
#### Enterprise and platform integration
Prioritize API access, team roles, security, data controls, analytics, and integration with learning or content systems. Copyleaks, GPTZero, Pangram, and Winston AI offer broader integration paths.
#### Multilingual review
Prioritize explicit language lists and your own language-specific test set. Do not assume a headline language count guarantees equal results across all languages.
#### Budget-first review
Compare the free quota with your real document size. A large per-scan box may be more useful than a generous monthly total if most work consists of long documents.
### 3.2 Top 10 Best AI Detector Products
#### 1. Lunote AI
Lunote AI Detector is designed for fast, explainable review. It shows AI-generated, mixed, and human-written shares, then points the reviewer to specific sentences. The next step can remain manual or move selected passages to Lunote AI Humanizer.
**How to use**
{% stepper %}
{% step %}
Paste a supported passage.
{% endstep %}
{% step %}
Run the detector.
{% endstep %}
{% step %}
Review the overall shares and highlighted sentences.
{% endstep %}
{% step %}
Check facts, context, and writing history.
{% endstep %}
{% step %}
Revise only the passages that genuinely need work, then run another check if useful.
{% endstep %}
{% endstepper %}
**Pricing and limits**
* Guest access: two free checks without sign-up, up to 600 characters per check.
* Free account: five AI detector checks per day, up to 600 characters per check.
* Pro Annual: $99.99 per year, shown as $8.33 per month; 1,000 detector checks and up to 5,000 characters per check.
* Pro Monthly: $1.95 for the first three days, then $29.99 per month unless cancelled before the trial ends; the detector allowance matches Pro Annual.
**Key selling points**
* AI-generated, mixed, and human-written result shares.
* Sentence-level highlights.
* Direct handoff to AI Humanizer for selected passages.
* Low-friction guest access.
* The report is designed to suggest a review path instead of presenting an unexplained percentage.
**User verdict:** best first stop for users who want a focused detector, clear sentence-level signals, and a direct review-to-revision workflow.
[Open Lunote AI Detector](https://lunote.ai/ai-detector)
#### 2. GPTZero
GPTZero is strongly positioned for education and writing-process review. Its value extends beyond a score through sentence highlights, reports, classroom workflows, and writing-history features.
**How to use**
Paste or upload a document, run the scan, review highlighted passages, and pair the result with drafts or writing history for sensitive decisions.
**Pricing and limits**
* Free plan: up to 10,000 words scanned per month.
* Premium: $12.99 per month when billed annually, with up to 300,000 words per month.
* Professional: $24.99 per month when billed annually, with up to 500,000 words per month and larger workflow capacity.
* Team, institution, and API options are available separately.
**Key selling points**
* Education-focused reporting and integrations.
* Sentence-level evidence.
* Writing-process and document-history features.
* Public guidance that detector results should not be treated as a final verdict.
**User verdict:** best for students, teachers, and institutions that need process evidence around a detection result.
[Visit GPTZero](https://gptzero.me/) · [See GPTZero pricing](https://gptzero.me/pricing)
#### 3. Originality.ai
Originality.ai is aimed at publishers, agencies, and SEO teams. It combines AI detection with plagiarism checking, readability, scan history, team reporting, and content-operation tools.
**How to use**
Paste text, upload a file, use the dashboard or extension, select the appropriate detector mode, and review the AI result together with plagiarism and editorial evidence.
**Pricing and limits**
* A limited free checker is available on the public product page.
* Pay as you go: $30 for 3,000 credits; one AI-detection credit covers 100 words.
* Pro: $12.95 per month when billed annually, with 2,000 subscription credits per month.
* Larger team and API requirements follow separate plans.
**Key selling points**
* Strong publishing and agency workflow.
* AI detection plus plagiarism, readability, and scan history.
* Credit-based usage that can fit irregular review volume.
**User verdict:** best for publishers and agencies that need content-integrity checks inside a repeatable editorial operation.
[Visit Originality.ai](https://originality.ai/) · [See Originality.ai pricing](https://originality.ai/pricing)
#### 4. Copyleaks
Copyleaks is a broad content-integrity platform for individuals, education, and enterprise teams. It combines AI detection, plagiarism checking, multilingual support, API access, and learning-system integrations.
**How to use**
Run a text or file scan, inspect the report, and evaluate the platform’s API, learning-system, security, and reporting options before adopting it across a team.
**Pricing and limits**
* A free public AI detector is available for basic checks.
* Personal: $16.99 monthly or $13.99 per month when billed annually.
* Pro: $99.99 monthly or $74.99 per month when billed annually.
* One unified credit covers up to 250 words or one image. Education and enterprise pricing is customized.
**Key selling points**
* AI and plagiarism checks in one platform.
* Support for 30+ AI-detection languages.
* API, learning-system, and team administration.
* Enterprise security and reporting paths.
**User verdict:** best for organizations that need detection inside a larger procurement, education, or compliance workflow.
[Visit Copyleaks AI Detector](https://copyleaks.com/ai-detector) · [See Copyleaks pricing](https://copyleaks.com/pricing)
#### 5. Grammarly
Grammarly’s AI Detector sits inside a broader writing ecosystem. It is useful for authors who want to review AI signals alongside grammar, citations, plagiarism, and authorship records.
**How to use**
Use the free web checker for a quick estimate or use AI detection inside a supported paid Grammarly workflow. Review the percentage and questionable passages, then add citations or writing-process evidence where required.
**Pricing and limits**
* A free web AI detector is available.
* AI detection inside Grammarly products is a paid feature on supported plans.
* Grammarly Pro is $12 per month when billed annually; the standard monthly price is $30.
**Key selling points**
* Detection in a familiar writing environment.
* Connections to plagiarism, citations, editing, and authorship.
* Clear public guidance that the score is an estimate rather than objective proof.
**User verdict:** best for everyday writers who want responsible AI review inside their existing editing workflow.
[Visit Grammarly AI Detector](https://www.grammarly.com/ai-detector) · [See Grammarly plans](https://www.grammarly.com/plans)
#### 6. QuillBot
QuillBot combines detection with paraphrasing, grammar, summarizing, citation, and humanizing tools. Its report distinguishes AI-generated, human-written and AI-refined categories and provides explanation cards.
**How to use**
Paste at least 80 words, run the detector, review the score and explainer cards, and decide whether the issue is authorship transparency, citation, or writing quality.
**Pricing and limits**
* Free: up to 1,200 words per scan and six scans per day.
* Premium: $8.33 per month when billed annually, with unlimited detection scans and fuller explanation support.
**Key selling points**
* Clear distinction between AI-generated, AI-refined, and human writing.
* Sentence-level review and explanation cards.
* Detection and revision tools in the same environment.
**User verdict:** best for writers who want a free detector and optional revision tools without changing platforms.
[Visit QuillBot AI Detector](https://quillbot.com/ai-content-detector) · [See QuillBot Premium](https://quillbot.com/premium)
#### 7. Winston AI
Winston AI combines text detection, plagiarism checking, writing feedback, document scanning, OCR, AI-image detection, and shareable reports.
**How to use**
Paste text, upload a file, import from a URL, or scan a document image. Review the human score and prediction map, then add plagiarism or report features when needed.
**Pricing and limits**
* Free access begins with a 14-day trial and 2,000 credits.
* Essential: $18 per month.
* Advanced: $29 per month.
* Elite: $49 per month.
* Annual billing can reduce the effective monthly price. Plan credits and features differ.
**Key selling points**
* Text, image, and document scanning.
* OCR for pictures and handwriting.
* Plagiarism, reports, and team features.
* Multilingual detection across a published language list.
**User verdict:** best for reviewers who need text detection plus OCR, image checks, and shareable reporting.
[Visit Winston AI Detector](https://gowinston.ai/ai-content-detector/) · [See Winston AI pricing](https://gowinston.ai/pricing/)
#### 8. ZeroGPT
ZeroGPT is a free-first AI detector with a large text box and a broad utility set. Paid plans add larger per-scan limits, batch files, reports, and other writing tools.
**How to use**
Paste text or upload a supported file, run the detector, inspect highlighted sentences, and avoid submitting confidential material until you have reviewed the privacy terms.
**Pricing and limits**
* Free: up to 15,000 characters per AI detection.
* Pro: $7.99 per month when billed annually, with up to 100,000 characters per detection.
* Plus and Max annual plans are listed at higher allowances and prices.
* API pricing is available separately.
**Key selling points**
* Large free input allowance.
* Paid batch-file and report options.
* Broad set of connected writing utilities.
**User verdict:** best for budget users who need fast, large quick checks and can apply careful score interpretation.
[Visit ZeroGPT](https://www.zerogpt.com/) · [See ZeroGPT pricing](https://www.zerogpt.com/pricing)
#### 9. BrandWell
BrandWell’s free detector is positioned for marketing, SEO, and brand-content review. It focuses on whether a draft reads as robotic or templated and connects to a wider content platform.
**How to use**
Paste text, enter a URL, or upload a supported file. Use the result to guide an editorial review of facts, tone, repetition, and brand fit rather than to chase a pass score.
**Pricing and limits**
* The public AI Content Detector is free.
* The detector entry shows a 2,500-character limit.
* Broader BrandWell products and services have separate commercial plans.
**Key selling points**
* Clear marketing and SEO use case.
* Text, URL, and file input paths.
* Connections to content creation, brand voice, and optimization tools.
**User verdict:** best for marketing teams that want a quick signal before a fuller editorial and brand review.
[Visit BrandWell tools](https://brandwell.ai/tools/)
#### 10. Pangram
Pangram is a focused detector for schools, universities, publishers, and teams. It emphasizes segment analysis, major-model coverage, multilingual detection, and integrations.
**How to use**
Paste or upload text, scan for AI assistance, inspect document segments, and keep the result with process evidence when the decision has real consequences.
**Pricing and limits**
* Starter: free with five daily credits.
* Premium: $20 monthly or $180 annually.
* Pro: $60 monthly or $540 annually.
* Developer access uses pay-as-you-go credits; enterprise and education pricing is separate.
**Key selling points**
* Segment-level AI assistance review.
* Support for major language models and 20+ languages.
* Chrome extension, education integrations, and API paths.
**User verdict:** best for users who want a focused detector with stronger institutional and integration options than a basic free text box.
[Visit Pangram](https://www.pangram.com/) · [See Pangram pricing](https://www.pangram.com/pricing)
***
## Part III: Clever AI Detector Practical Examples
### Chapter 4 | Best AI Detector Tools by Use Case Segment
The five examples below use the same review logic: inspect a sufficiently long passage, examine highlighted sections, add writing-process context, and let a person make the final decision. The images are representative interface examples, not independent accuracy benchmarks and not a ranking by score.
### 4.1 Lunote
**Best fit:** quick self-check, editorial triage, and a direct detector-to-humanizer workflow.
**Review path**
{% stepper %}
{% step %}
Read AI, mixed, and human shares as an overview.
{% endstep %}
{% step %}
Open the highlighted sentences.
{% endstep %}
{% step %}
Compare those sentences with the author’s sources and drafts.
{% endstep %}
{% step %}
Send only genuine writing-quality problems to AI Humanizer.
{% endstep %}
{% step %}
Recheck the revised passage if the comparison helps the reviewer.
{% endstep %}
{% endstepper %}
### 4.2 GPTZero
**Best fit:** education, student self-check, and cases where writing-process evidence matters.
**Review path**
{% stepper %}
{% step %}
Review the document-level result and sentence contribution.
{% endstep %}
{% step %}
Check writing history, drafts, or document replay when available.
{% endstep %}
{% step %}
Ask the author to explain the flagged section.
{% endstep %}
{% step %}
Record the human conclusion separately from the detector score.
{% endstep %}
{% endstepper %}
### 4.3 Pangram
**Best fit:** schools and teams that want document segments, integrations, and a focused detection product.
**Review path**
{% stepper %}
{% step %}
Identify which segments carry the signal.
{% endstep %}
{% step %}
Compare the segment with neighboring human-written passages.
{% endstep %}
{% step %}
Check whether AI assistance was permitted and disclosed.
{% endstep %}
{% step %}
Use the organization’s policy—not the detector alone—to determine the response.
{% endstep %}
{% endstepper %}
### 4.4 Grammarly
**Best fit:** writers who already use Grammarly for editing, citations, plagiarism, and authorship records.
**Review path**
{% stepper %}
{% step %}
Review the estimated AI percentage.
{% endstep %}
{% step %}
Add citations where AI-assisted research or sourced language requires attribution.
{% endstep %}
{% step %}
Use Authorship or writing records when available.
{% endstep %}
{% step %}
Make content edits for clarity and accountability, not only to lower a score.
{% endstep %}
{% endstepper %}
### 4.5 QuillBot
**Best fit:** a free quick check with explainers and optional revision tools.
**Review path**
{% stepper %}
{% step %}
Confirm that the passage meets the minimum useful length.
{% endstep %}
{% step %}
Distinguish AI-generated from AI-refined sections.
{% endstep %}
{% step %}
Read the explanation cards before rewriting anything.
{% endstep %}
{% step %}
Keep facts, citations, and intended meaning intact during revision.
{% endstep %}
{% endstepper %}
### A responsible review workflow
The detector is one layer of a larger review:
{% stepper %}
{% step %}
#### Collect context
Drafts, notes, sources, citations, and the applicable policy.
{% endstep %}
{% step %}
#### Run detection
Use an appropriate sample and record the tool.
{% endstep %}
{% step %}
#### Review passages
Inspect highlights and add another signal only when it is useful.
{% endstep %}
{% step %}
#### Decide by a human
Document the reasoning and avoid automatic score-based penalties.
{% endstep %}
{% endstepper %}
***
## Q\&A About AI Detector
A. Are AI detectors accurate?
They can be useful at identifying patterns, but no detector is perfect. Accuracy varies by model, language, sample length, editing level, and the type of writing. Treat the result as a review signal.
B. Can an AI detector prove plagiarism?
No. AI detection estimates whether text resembles model-generated writing. Plagiarism checking looks for overlap with sources. A passage can be original but AI-generated, or copied without any AI involvement.
C. Why did my own writing get flagged?
Formal structure, repeated sentence shapes, predictable transitions, technical language, or a short sample can resemble AI patterns. Keep drafts and sources, inspect the flagged passage, and do not assume the score proves anything about intent.
D. Should I use more than one detector?
Use a second detector when the decision is important and another signal will genuinely improve the review. Do not average several scores into false certainty. Differences should lead to closer examination, not a harsher automatic judgment.
E. What is the safest student workflow?
Keep notes, outlines, sources, and drafts. Run a self-check on a coherent passage, review highlights, correct real writing problems, cite allowed AI assistance, and follow the institution’s policy.
F. What is the safest teacher workflow?
Treat the score as a reason to review. Check the assignment instructions, compare with earlier work, ask the student to explain the writing process, and document the final human judgment. Do not penalize a student only from a detector result.
G. What is the safest publisher workflow?
Combine AI detection with plagiarism checks, source verification, fact-checking, readability, brand review, and contractual disclosure rules. The publication decision belongs to an editor.
H. Is a screenshot enough evidence?
No. A screenshot can record one tool’s result at one moment, but it usually lacks model version, complete input, language settings, writing history, and policy context.
I. Should I paste confidential text into a free detector?
Only after reviewing the product’s privacy, retention, deletion, and model-training terms. For confidential work, use an approved enterprise route or remove sensitive details before testing.
J. Which detector should I start with?
Start with the decision you need to make. Lunote is a practical first stop for a fast, sentence-level review and a clear revision path. Education teams may prioritize GPTZero or Pangram; publishers may prioritize Originality.ai; enterprises may prioritize Copyleaks.
K. Should I choose free or paid?
Use free access when the document size and frequency fit the limits and you do not need team controls, saved reports, integrations, or higher privacy requirements. Choose paid access when larger capacity or organizational workflow features solve a real need.
[Try Lunote AI Detector](https://lunote.ai/ai-detector)
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:
```
GET https://jia-mei-he-shun.gitbook.io/lunote-ai-humanizer-handbook/ai-detector-handbook.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.