Pangram Labs — Free Mini Course

You've Read AI-Written Content Today. Did You Know It?

Here's a stat that should make you stop: researchers estimate that up to 40% of content published online is now AI-generated — and the number is climbing.

• ▸Most people frame AI detection as an academic integrity issue — students using ChatGPT on essays. That's real. But it's the tip of the iceberg.
• ▸Here's where undetected AI content creates genuine risk:
• ▸🎓 **Education** — Students submitting AI-written work don't develop the skills they're supposed to. Grades lose meaning. Institutions lose credibility.
• ▸📰 **Publishing & media** — AI-generated articles that slip through erode reader trust and can spread misinformation at scale.
• ▸💼 **Hiring** — Cover letters and work samples written by AI tell you nothing about the actual candidate. You might hire someone who can't do the job.
• ▸⚖️ **Legal & compliance** — Some industries (finance, healthcare) have disclosure requirements. Using undisclosed AI-generated content can create liability.
• ▸The old way: rely on gut instinct and Google searches. The new way: use a purpose-built tool with research-backed detection.

• ▸Pangram Labs is an AI detection platform built specifically to identify AI-generated text — with a focus on accuracy and transparency.
• ▸Unlike basic detectors that spit out a percentage and call it a day, Pangram Labs is:
• ▸🔬 **Research-backed** — built on peer-reviewed methodology, not a black box
• ▸📊 **Transparent about confidence** — it tells you how certain it is, not just what it found
• ▸🆓 **Free tier available** — scan individual documents without a credit card
• ▸🏢 **Built for teams** — educators, content teams, and enterprise clients all have workflows
• ▸The core product is simple: paste or upload text, run a scan, get a detailed report with highlighted sections and confidence scores.
• ▸Think of it as a second opinion — a fast, reliable one — for any document where authenticity matters.
• ▸Pangram Labs won't replace your judgment. But it will sharpen it.

• ▸Ever wonder how a tool can tell if text was written by a machine? Here's the plain-English version.
• ▸AI language models like ChatGPT generate text by predicting the most statistically likely next word, over and over. This creates patterns that are subtly different from how humans write.
• ▸Pangram Labs detects these patterns by analyzing:
• ▸📉 **Perplexity** — how 'surprising' is the text? Humans are unpredictable. AI tends to choose safe, expected words. Low perplexity = more likely AI.
• ▸📈 **Burstiness** — humans write in uneven bursts — sometimes complex sentences, sometimes simple ones. AI writing tends to be eerily consistent in complexity.
• ▸🔁 **Statistical fingerprints** — certain phrase structures and word combinations appear more often in AI output than in human writing.
• ▸Important caveat: these are signals, not proof. No detector is 100% accurate. Pangram Labs is honest about this — it gives you a confidence score, not a verdict.
• ▸Use it as evidence, not a judge. Understanding the mechanics behind AI detection — not just 'run the scan' — means you'll use it smarter and more responsibly than most people who download these tools.

• ▸Using Pangram Labs for the first time takes about a minute. Here's exactly what to do:
• ▸1️⃣ **Go to pangramlabs.com** — create a free account (no credit card needed)
• ▸2️⃣ **Paste your text** — copy the document you want to check and paste it into the scan window. Minimum ~100 words for a reliable result.
• ▸3️⃣ **Run the scan** — click Analyze. Pangram processes the text in a few seconds.
• ▸4️⃣ **Review the report** — you'll see an overall AI probability score plus a highlighted version of the document showing which sections triggered detection.
• ▸5️⃣ **Check the confidence level** — Pangram tells you how confident it is in the result. A high-confidence flagging is worth taking seriously. A low-confidence result means you need more context.
• ▸Pro tip: For best results, scan the full document — not just excerpts. Pangram's signals improve with longer text, and a single AI-written paragraph in an otherwise human document can still get caught.
• ▸Let's talk about reading those results next.

• ▸Pangram gives you a score. Now what?
• ▸Here's how to interpret what you're seeing without overreacting or under-reacting.
• ▸**AI Probability Score:**
• ▸- **0–30%** — Likely human. Low AI signal. Proceed normally.
• ▸- **30–60%** — Mixed signal. Could be edited AI content, or naturally consistent human writing. Worth a closer look.
• ▸- **60–85%** — Strong AI signal. Flag for review. Don't act on this alone — ask for the original draft or process.
• ▸- **85%+** — Very high confidence. The statistical signature is strong. Still a signal, not a confession — but worth serious follow-up.
• ▸**Confidence Score:** This is separate from the AI probability. It tells you how reliable that probability estimate is. A 75% AI score at 90% confidence is much more actionable than a 75% score at 40% confidence.
• ▸**The golden rule:** Never take action based solely on a Pangram score. Use it to guide a conversation, not end one.
• ▸That's what separates fair use from misuse.

• ▸Here's something most AI detection tools won't tell you: AI can be rewritten to evade detection.
• ▸It's called **AI humanization** — tools like Undetectable.ai or even just re-prompting ChatGPT with 'rewrite this to sound more human' can lower AI probability scores significantly.
• ▸This is real. You should know it. Here's how it affects you:
• ▸⚡ **A low score doesn't mean human** — it might mean the AI output was processed afterward. This is why Pangram's confidence levels matter: humanized AI often creates unusual patterns that lower confidence even as it lowers the AI score.
• ▸⚡ **A high score is still just a signal** — even without evasion, Pangram can sometimes flag human writing (false positives). Non-native English writers are particularly affected. Don't treat a high score as definitive proof.
• ▸⚡ **Context is everything** — the score should be one input in a larger picture that includes the person's history, the circumstances, and any other evidence.
• ▸Pangram Labs is transparent about these limitations. That's not a weakness — it's what makes it a trustworthy tool.

• ▸Pangram Labs serves two very different audiences — and knowing which camp you're in shapes how you use it.
• ▸🎓 **Educators:**
• ▸- Scan submitted essays before grading — flag suspicious submissions for closer review
• ▸- Use results to open a conversation, not issue an accusation
• ▸- Pair with your institution's academic integrity policy — Pangram provides evidence, policy provides process
• ▸- Free tier is often sufficient for individual teachers; department plans for scale
• ▸💼 **Content teams and businesses:**
• ▸- Screen freelancer submissions before publishing — protect your brand voice
• ▸- Audit content pipelines where AI use isn't disclosed
• ▸- Legal and compliance teams scan external documents (filings, reports, submissions)
• ▸- API access available for teams processing large volumes automatically
• ▸In both cases: **document your process.** If you use Pangram to flag something and then take action, you want a record of the score, the confidence, and what you did next.
• ▸Due process protects everyone.

• ▸Ready to put this into practice? Here's what to do in the next 7 days:
• ▸✅ **1. Create your free account** — pangramlabs.com, no card required. Poke around before your first real scan.
• ▸✅ **2. Run a test scan on something you know** — paste an article you wrote yourself. See what score comes back. This calibrates your intuition for what 'human' looks like in the system.
• ▸✅ **3. Scan a ChatGPT output** — ask ChatGPT to write a short paragraph, then scan it. Watch the score. Now you have a reference point.
• ▸✅ **4. Try the humanization test** — take that same ChatGPT output, ask ChatGPT to 'rewrite this to sound more human,' and scan again. See how much the score changes. Eye-opening.
• ▸✅ **5. Share your process with one colleague** — whether you're in education, content, or HR, AI detection is becoming a shared workflow skill. Get someone else on board.
• ▸The goal isn't to become suspicious of everything. The goal is to have a fast, reliable gut-check when authenticity matters.
• ▸You now have one.