Reading through a company's sustainability report and spotting the BS takes expertise and hours of work. Now imagine doing that for thousands of companies across millions of web pages. That's where AI comes in.
The Problem With Manual Detection
There are roughly 230 different environmental labels in the EU alone. Companies publish sustainability claims across websites, reports, packaging, social media, and advertising. No human team can monitor all of it.
The EU's own research found over half of environmental claims were misleading — but discovering that required a massive manual study. We need something faster.
How NLP Identifies Greenwashing
Natural Language Processing (NLP) — the branch of AI that understands human language — can analyze environmental claims at scale. Here's the process:
1. Claim Extraction
First, the AI scans text and identifies sentences that contain environmental claims. It looks for patterns like:
- Environmental keywords: "sustainable," "eco-friendly," "green," "carbon neutral"
- Comparative claims: "greener than," "reduced impact," "better for the planet"
- Quantified claims: "50% less emissions," "100% recyclable"
- Certification references: "certified organic," "FSC approved"
2. Claim Classification
Each extracted claim gets classified by type:
- Vague claims: No specific, measurable assertion ("eco-friendly packaging")
- Specific claims: Measurable and verifiable ("made with 60% post-consumer recycled content")
- Comparative claims: Benchmarked against something ("30% less water than our 2020 product")
- Absolute claims: Blanket statements ("zero waste," "carbon neutral")
3. Credibility Scoring
This is where it gets interesting. The AI evaluates each claim against several factors:
- Specificity: Does the claim include numbers, timelines, and scope?
- Evidence: Is supporting data referenced or linked?
- Certification: Does it reference recognized third-party standards?
- Scope clarity: Does it specify what the claim applies to?
- Known red flags: Does it use terms the EU has flagged as commonly misleading?
4. ClimateBERT and Domain-Specific Models
General AI models are decent at understanding language, but they miss domain-specific nuances. That's why models like ClimateBERT — trained specifically on climate and environmental text — perform significantly better.
ClimateBERT was trained on climate-related texts from corporate reports, scientific papers, and news articles. It understands the difference between genuine technical claims and marketing fluff in ways that general models can't.
Our Greenwashing Scanner uses this technology to analyze any website for potentially misleading environmental claims.
What AI Can and Can't Do
What It Does Well
- Identifies vague, unsubstantiated language at scale
- Flags claims that use known problematic patterns
- Compares claims against regulatory requirements
- Processes thousands of pages in minutes
- Provides consistent, bias-free analysis
What It Can't Do (Yet)
- Verify whether specific data points are true
- Access private supply chain information
- Replace human judgment for complex, nuanced claims
- Guarantee legal compliance — it's a screening tool, not a lawyer
Why This Matters Now
With the EU Green Claims Directive requiring substantiation of all environmental claims, companies need to audit their communications proactively. Manual audits are expensive and slow. AI-powered tools can do a first pass in seconds, flagging the highest-risk claims for human review.
It's not about replacing human expertise — it's about making that expertise scalable.
Try It Yourself
Our free scanner analyzes any URL for potential greenwashing. Enter a website, and within seconds you'll get a breakdown of environmental claims found, their credibility scores, and specific concerns flagged by our AI.