Product 01 · Green-claims screening for consultants & auditors

Screen the claim before the regulator does.

ReputationIndex is the first tool that screens a green claim against the law before a regulator does — and tells you honestly when it doesn’t know. Every flag carries the exact legal provision it trips. When the evidence isn’t there, the answer is “not determined”, never a guess.

We’re looking for pilot engagements with consultancies and audit firms ahead of the September deadline.

days until Directive (EU) 2024/825 applies — 27 September 2026
The problem is already enforced

Greenwashing now has a price tag

These are verified, documented enforcement cases — real courtrooms, real regulators, real verdicts. Every one of them ships inside the product as a live demo, alongside a catalog of 27 verified cases.

GLS “Climate Protect”
AGCM (Italian Competition Authority), February 2025
€8M fine — the largest Italian greenwashing fine of 2025
Apple Watch “carbon neutral”
Frankfurt Regional Court, August 2025
Banned in Germany — up to €250k per violation
Shein “evoluSHEIN”
AGCM, August 2025
€1M fine
KLM “Fly Responsibly”
Amsterdam District Court, March 2024
15 of 19 statements ruled misleading
FIFA “carbon-neutral World Cup”
Swiss Fairness Commission, 2023
Ordered to stop the claim
The deadline

27 September 2026 — the rules change across the EU

Directive (EU) 2024/825 becomes enforceable EU-wide. The enforcement above happened under the old patchwork; the per-se bans arrive now.

Generic claims banned outright. “Eco-friendly”, “green”, “climate friendly” — prohibited without proven excellence.
Offset-based “carbon neutral” product claims banned outright. No qualification saves them.
The burden of proof shifts to the company making the claim.

The full regulatory picture: CSRD, CS3D, DPP and 2024/825 →

Today’s screening workflow is manual: a consultant reads marketing line by line, cross-references the directive annex, and forms a judgment — hours per brand, consistent only as far as the analyst is. General-purpose AI tools can’t fill the gap: they can’t prove they didn’t invent the claim, aren’t reproducible for an audit file, and give opinions instead of legal citations.

The product

Point it at claims. Get the law back.

Screens

Extracts a brand’s environmental and social claims from text or a URL and runs each one through a versioned catalog of legal rules from Directive (EU) 2024/825.

Cites

Every flag carries its rule id and the exact legal provision — “Offset-based climate neutrality claim, Annex I point 4c” — not a vibe score.

Verifies and scores

Runs flagged claims through evidence verification (MVP — lexical retrieval, mediator–advocate tiers, first-class abstention) and rolls the result into a disclosed, uncalibrated 0–100 risk index.

Abstains

When the source doesn’t establish a fact, the system says “not determined” instead of guessing. No evidence, no penalty.

Explains and delivers

A grounded AI assistant answers questions about the analysis, charts it, proposes compliant rewrites, and exports the audit report as a PDF.

Monitors

Register a brand, harvest its pages, PDFs and news, re-screen on demand and get diff-based alerts on a portfolio dashboard — audits become monitoring.

How it works

Agents where judgment is needed. Deterministic code where the law is.

The model reads language — the one thing it is good at. It never judges the law: the same claims in, the same flags out, every time, stamped with the catalog version.

Source textsite copy, reports, ads, any URL
Extraction agentan LLM reads the claims (model-agnostic)
Quote guardrailno verbatim quote → claim rejected, and shown
Rule engine9 deterministic rules, versioned catalog
Flags + citationsrule id + the exact legal provision
AI model — reads language, never judges the law Deterministic code — reproducible, versioned The law — quoted verbatim, never paraphrased

The wording

“climate-neutral parcel delivery” — offset-funded, method undisclosed
GLS “Climate Protect” · flag OFF-01 — offset-based neutrality claim (per-se)

The law — verbatim from the Official Journal

“Claiming, based on the offsetting of greenhouse gas emissions, that a product has a neutral, reduced or positive impact on the environment in terms of greenhouse gas emissions.”
Directive 2005/29/EC, Annex I, point 4c — inserted by Directive (EU) 2024/825 · applies from 27 Sep 2026

Every flag opens a “Legal mandate” panel like this: the claim’s words beside the law’s words — the exact statute the wording trips, quoted, not a citation label.

The moat

The feature nobody else would build: it refuses to guess

Every attribute of a claim is tri-state: true, false — or not determined. Rules abstain on unknown facts. A confident wrong answer gets your client sued; a tool that says “I don’t know” when it doesn’t know is one an auditor can sign.

In the demo, Apple’s Watch “carbon neutral” claim — banned by a Frankfurt court — is flagged per-se. Right below it, the same brand’s “carbon neutral by 2030” pledge shows zero flags. Not because it’s clean — because those facts were never litigated. The system abstains rather than fabricate a judgment:

not determined
Before the regulator, on the record

Every flag is timestamped the moment it’s raised

The first time a claim trips a rule, the system stamps it — set once, never overwritten, surviving re-audits and re-extraction. A wording risk goes on the record before any regulator moves. Un-adjudicated cases sit on a prediction register of the system’s own wording-level findings, dated to the audit.

2.1 years
median lead of wording-level flags over the regulator’s decision, replayed across 18 dated, decided cases in the catalog
5.3 years
GLS’s offset-based wording was public from October 2019; AGCM’s €8M fine landed in February 2025
27 cases
verified enforcement cases built in — court rulings, ad bans, settlements, ongoing litigation — each labelled honestly
Honesty guardrail. Regulator and court action is not automatically tracked: absence of a known action means only that none is known to this system as of the audit date — never that none exists. The lead-time figures are retrodiction, not validation: lead time is counted only for flags that fire from the claim’s own published wording, never from what a later investigation uncovered. A registered prediction is not a validated one.
Why an auditor can trust it

Trust by construction, not by promise

Verbatim-quote guardrail

A claim is accepted only if its exact quote appears in the source. Hallucination is answered structurally — and rejections are displayed, never hidden.

Deterministic and reproducible

The legal layer is code, not sampling: same input, same catalog version, same flags. Pinnable for any past result in an audit file.

Reviewable line by line

Each rule maps to an article with its verbatim citation; your own lawyers can audit the catalog against the statute.

Model-agnostic

Claude, GPT or an open-weight model on your own infrastructure — compliance logic independent of any AI vendor, with a data-residency path by design.

What we don’t do yet — on purpose

  • Evidence verification is MVP-grade, not final. Retrieval is lexical (token overlap), not embeddings — a documented, acknowledged limit. It abstains rather than fabricate a substantiation verdict either way.
  • No ground-truth greenwashing dataset exists — for anyone. So we don’t quote an unverifiable accuracy number. We optimize for inspectability: every decision traceable to a rule and a quote.
  • The risk index ships uncalibrated — and says so. Explicit, versioned weights, disclosed as expert-judgment defaults, stamped on every snapshot. No false precision.

In compliance tooling, a system that states what it can’t do is worth more than one that claims to do everything.

Who it’s for

A consultant’s tool first, a monitoring platform next

Sustainability consultancies

Pre-deadline claim audits for every consumer-brand client.

Audit & assurance firms

Defensible, citable screening inside existing engagements.

Legal & compliance teams

In-house triage before campaigns ship.

On-demand audits become continuous monitoring: re-harvest, diff the claims, alert on new violations — a portfolio of brands watched every day.

The platform vision

From green claims to full supply-chain reputation intelligence

Corporate reputation now constitutes the majority of enterprise value, yet the systems companies use to protect it remain manual, fragmented and reactive. The EU’s CSRD, CS3D and Digital Product Passport turn supply-chain transparency from best practice into legal obligation. Green-claims screening is the first shipped module of a larger platform: AI-powered due diligence over the whole supply chain.

The platform vision: ingest real-time data from international news feeds, regulatory filings, macroeconomic databases and patent registries; process it with NLP, retrieval-augmented generation and knowledge-graph technology; and deliver a multidimensional reputation score across five dimensions — every component traceable back to its sources.

Financial stability

Credit signals, key ratios and filings, benchmarked against the sector — for public and private counterparties.

Compliance

Certifications (ISO, EMAS, SA8000), licenses and enforcement actions, cross-referenced with human-rights and labor monitoring.

Supply-chain stability

Concentration risks and country exposure, using OECD, IMF, World Bank and BIS data.

ESG

Declared commitments cross-checked against independent sources — greenwashing and social-washing detection by design.

Strategy & innovation

R&D investment, patents, research collaborations and demonstrated resilience through past disruptions.

Every company — not only the largest multinationals — should be able to evaluate a supplier on another continent with the depth of a dedicated risk team. That’s the premise: democratizing supply-chain intelligence, for the SMEs that form the backbone of European industry.

Roadmap

Now — shipped
Claim extraction, rule screening with citations, grounded assistant with charts / SQL / PDF export, URL analysis, 27-case demo catalog, brand harvest with per-source status, evidence verification (MVP) with first-class abstention, a disclosed 0–100 risk index, diff-based monitoring alerts, AI-drafted mitigation plans with human ownership, and Italian-jurisdiction screening (Codice del Consumo).
Next
Embeddings-based evidence retrieval in place of lexical scoring; scheduled monitoring instead of on-demand runs.
Then
Postgres/pgvector migration for multi-user scale; social and ads connectors where platform terms allow.
Scale
Additional jurisdiction rule catalogs — and the five-pillar supply-chain intelligence platform.
Get started

Screen the claim before the regulator does.

We’re looking for pilot engagements with consultancies and audit firms ahead of the 27 September 2026 deadline.

Talk to us

Tell us what you need and we’ll come back to you. No newsletter, no drip sequence — a person reads this and replies.

Reputation Index — screening support, not legal advice. Flags identify wording at risk under a cited provision; they are not accusations, legal assessments or predictions of regulator action.