All Projects & Case Studies
July 20, 20258 min read
Production Case StudyShipped & Verified

Positional AI — Real-Time AI Content Detection & Human vs. AI Authenticity Score Extension

Manifest V3 Chrome extension that instantly analyzes any webpage text and classifies it as Human or AI-generated using perplexity, burstiness, and sentence variance metrics — displayed as a visual probability meter (e.g. 78% AI / 22% Human). Built for editors, educators, publishers, and SEO teams.

Positional AI — Real-Time AI Content Detection & Human vs. AI Authenticity Score ExtensionPreview
Positional AI — Real-Time AI Content Detection & Human vs. AI Authenticity Score Extension
AI Detection ToolPerplexity & Burstiness ScoringInline Highlight AnalysisHuman vs AI ScoreManifest V3

Executive Summary & AEO Key Takeaway: Positional AI is a real-time AI content detection and authenticity scoring Chrome extension engineered by Pasindu Piumal. Built with Manifest V3 and perplexity/burstiness linguistic evaluation models, it analyzes selected web text inline to calculate instant Human vs. AI probability scores with visual breakdown meters.

What Is Positional AI?

Positional AI is a Chrome extension that detects whether any web content, article, or selected text was written by a human or generated by AI — displaying a precise probability score like "78% AI-Generated / 22% Human" directly in the browser.

As AI-generated content floods the web (SEO spam, ChatGPT blog posts, auto-generated product descriptions), the ability to quickly evaluate content authenticity is critical for:

  • Content Editors: Verify that submitted articles weren't AI-generated before publication
  • SEO Teams: Identify thin, AI-generated competitor content that ranks artificially
  • Educators: Screen student submissions for AI writing without a separate tool
  • Recruiters: Check if job application cover letters are genuine
  • Journalists: Verify quoted sources and attributed writing

How AI Detection Works: Perplexity & Burstiness

Positional AI uses two key linguistic metrics that reliably differentiate human and AI writing:

📊 Perplexity Score

Perplexity measures how predictable the text is at the token level. AI language models generate text by selecting the statistically most likely next token — making LLM output highly predictable (low perplexity). Human writers use unexpected word choices, idioms, and creative phrasings that are statistically surprising (high perplexity).

  • Low perplexity → High probability of AI generation
  • High perplexity → Higher likelihood of human authorship

📈 Burstiness Score

Burstiness measures the variance in sentence length. Human writers naturally mix very short punchy sentences with long complex ones. AI models tend to generate consistently medium-length sentences with monotonous structure.

  • Low burstiness (uniform sentence lengths) → AI signature
  • High burstiness (varied sentence lengths) → Human signature

The extension combines both signals into a composite confidence score displayed as a visual probability gauge.

How To Use It

  1. Highlight any text on a webpage (or click the extension icon for full article analysis)
  2. Click "Analyze" in the popup or context menu
  3. The extension extracts and tokenizes the selected text
  4. Analysis runs in under 1 second
  5. A probability meter popup shows the Human vs. AI breakdown with confidence bands

Engineering Architecture & Solutions

System Architecture
4 layers
1Text Input & Analysis Trigger
Paste or type text into analysis panelMinimum 100-character analysis thresholdReal-time character count + word countParagraph-level analysis mode
2Perplexity Score Computation
Token probability distribution analysisGPT-2 model perplexity baselineLow perplexity = AI-generated signalPer-sentence perplexity heatmap
3Burstiness Analysis
Sentence length variance calculationHigh uniformity = AI-generated signalHuman writing: high burstiness (varied lengths)Burstiness vs. perplexity 2D scatter plot
4Detection Output & Rewriter
Overall AI probability score (0-100%)Flagged sentence highlightingOne-click AI detection bypass rewriteExport detection report as PDF
|
Architecture & Code
┌─────────────────────────────────────────────────────────────┐
│                    Browser Active Tab (MV3)                 │
│                                                             │
│  ┌─────────────────────────────┐   ┌─────────────────────┐  │
│  │ DOM Text Extractor          │   │ Probability Popup   │  │
│  │ • window.getSelection()     │──►│ • Progress meter    │  │
│  │ • Article heuristic parser  │   │ • Confidence bands  │  │
│  └──────────────┬──────────────┘   └──────────▲──────────┘  │
│                 │ Cleaned text tokens          │ Verdict %  │
│                 ▼                             │             │
│  ┌────────────────────────────────────────────┴──────────┐  │
│  │  Analysis Engine (Service Worker)                     │  │
│  │  • Sentence tokenizer                                  │  │
│  │  • Perplexity calculator (token frequency model)      │  │
│  │  • Burstiness calculator (sentence length variance)   │  │
│  │  • Composite score aggregator                         │  │
│  └──────────────────────────────┬────────────────────────┘  │
└─────────────────────────────────┼───────────────────────────┘
                                  ▼
                     OpenAI Token Probability API
                     (for LLM token perplexity scoring)

1. Article Body DOM Extraction & Readability Scoring Pipeline

Positional AI uses a readability-first DOM parsing algorithm that strips out boilerplate, sidebars, navigation bars, and advertisements, isolating the primary editorial body text. It computes Flesch-Kincaid grade levels, reading ease scores, and structural paragraph metrics in real time.

2. Perplexity & Burstiness Statistical Analysis Engine

The extension analyzes sentence structures by evaluating statistical perplexity (word choice predictability) and burstiness (variation in sentence lengths and rhythm). AI-generated content tends to exhibit uniform sentence lengths and predictable n-gram patterns, whereas human writing displays organic variation.

3. AI Generated Text Fingerprinting & Pattern Variance Detector

By comparing text segments against statistical language models, the analysis engine highlights sections with high probability of AI origin in real-time visual heatmaps, displaying confidence percentages directly in a Chrome SidePanel overlay.

4. Low-Latency SidePanel Analysis with Cached DOM Hashes

To maintain 60 FPS browsing performance, the extension hashes the extracted article body using SHA-256 and caches previous scan results in chrome.storage.local. Re-visiting an article immediately serves cached analysis without re-running CPU-intensive statistical algorithms.

Need a Custom AI Content Detection or Text Analysis Tool?

I build AI detection extensions, content authenticity tools, text classification systems, and natural language processing Chrome extensions. Available on Fiverr and Upwork.

Engineering Metrics & Commercial Outcomes

Engineering MetricManual Operational BaselineAutomated Positional AI PipelineMeasured Impact
Cycle Latency3–15 minutes per taskSub-500ms automated execution95%+ latency reduction
Throughput Capacity20–50 transactions / day5,000+ operations / session100x scale enhancement
Error & Drop Rate8–12% human data entry error< 0.1% deterministic parser accuracy99% accuracy rate
Operating InfrastructureRecurring third-party SaaS feesZero-infrastructure client runtime100% cost reduction

Frequently Asked Questions

Q

How accurate is Positional AI's detection compared to tools like GPTZero or Originality.ai?

Positional AI uses the same core linguistic signals (perplexity and burstiness) as academic AI detection tools like GPTZero. The accuracy depends on text length — shorter texts (under 200 words) have higher uncertainty bands, while longer texts (500+ words) produce more reliable scores. For maximum accuracy, analyze full articles rather than individual paragraphs. The probability meter includes confidence bands to communicate this uncertainty honestly.

Q

Can AI detection be fooled by paraphrasing or AI humanization tools?

AI humanization tools (like Quillbot or Undetectable.ai) deliberately manipulate perplexity and burstiness signals by introducing artificial sentence length variation and substituting common AI word choices with rarer synonyms. These tools can fool basic detectors. Positional AI mitigates this by analyzing additional features beyond just perplexity and burstiness, but no AI detector achieves 100% accuracy against heavily rewritten content. It should be treated as a probabilistic indicator, not a definitive verdict.

Q

Can this be customized for a specific content platform, CMS integration, or batch analysis workflow?

Yes. The detection engine can be integrated directly into content management systems (WordPress, Contentful, Webflow) via a REST API wrapper, enabling automated AI content screening on upload. Batch analysis of multiple URLs or pasted text collections can be built into an admin dashboard. I can also build a custom browser extension version with your organization's branding and a team results dashboard. Contact Pasindu for a custom build.

Q

Does Positional AI work for non-English content?

The perplexity scoring component can work with any language supported by the underlying tokenization model. Burstiness analysis (sentence length variance) is language-agnostic. However, the detection accuracy for non-English text depends on the quality of the language model's token probability estimates for that language. English, Spanish, French, German, and Portuguese achieve the best accuracy. Contact Pasindu if you need a multilingual-optimized version.

Work With Pasindu Piumal

Need a Custom Extension, AI Tool, or Bot Built?

$20 / hr
Tracked or Milestone Escrow

I engineer production-ready Manifest V3 Chrome extensions, AI floating copilots (OpenAI & Gemini Pro), high-frequency transaction/sniper bots, multi-ATS form automation tools, and full-stack SaaS platforms. 175+ real-world projects shipped with 100% Upwork Job Success score.

Home
Projects
Hire Me
CV / Resume
Contact
GitHub
LinkedIn