EcoClaim.eu VS Python

EcoClaim.eu's answer

Three groups.

E-commerce operators selling into the EU who use words like "eco-friendly", "sustainable" or "climate neutral" in their copy or on their packaging.

Sustainability and compliance officers at larger companies who need a documented review of their own website and evidence they can keep on file.

Agencies and consultancies running compliance audits for clients, who need white-labelled output, structured exports, and a way to show a client what changed since the last audit.

EcoClaim.eu's answer

Most tools search your text for banned words. EcoClaim reads the page twice, in two different ways, and makes the two readings agree before it reports anything.

One pass reads the copy. A second reads the images — packaging, badges, product shots — because a claim on a label carries the same legal weight as a claim in a sentence. Those two passes routinely disagree: the image layer recognises a legitimate certification mark while the text layer flags the same badge as self-invented. A reconciliation step resolves those conflicts deterministically before the report is written, so you never receive two contradictory verdicts about the same object.

It also knows when not to flag something. A claim backed by a genuine, recognised certification scheme is cleared rather than reported — unless the label is being misused, which is still a violation. That exception handling is the difference between a report a compliance officer trusts and a list of false positives they stop reading.

Every finding carries the provision it breaches — down to the specific Annex point of the Directive — the exact wording that triggered it, why that severity was assigned, and a rewritten version of the sentence that would comply.

And it is deterministic. The same page scanned twice produces the same findings. For a document you may have to defend to a regulator, that matters more than it sounds.

EcoClaim.eu's answer

Because the hard part of this is not finding the word "eco". It is judging whether a particular claim, on a particular kind of product, in a particular country, is actually unlawful — and then telling you what to write instead.

Detection is graded, not binary. Findings come back red, yellow or green with the reasoning attached. A quality label from a genuine industry association is not called fraudulent; it is marked as needing evidence on file, which is precisely what a regulator would ask for. Blunter tools call everything a violation, and the report becomes noise.

The rules carry exceptions, examples and statutory citations, and they are applied through the lens of your sector and your market. The same phrase is assessed differently for cosmetics than for industrial B2B, and national layers for Germany, France, Spain, Italy, the Netherlands and Poland apply local enforcement practice and penalties on top of the EU baseline.

Every flagged claim comes with a suggested rewrite that keeps the marketing intent while removing the legal exposure — so the output is a work list, not a verdict.

Findings are then ranked by how much each fix resolves. On a recent audit, four changes cleared 79 of 92 findings.

And it is a one-time purchase, not a subscription. Compliance is an audit, a round of fixes, and a re-check months later — priced accordingly, with credits that never expire.

EcoClaim.eu's answer

The Empowering Consumers for the Green Transition Directive applies across the EU from 27 September 2026. It turns generic environmental claims into an unfair commercial practice, with penalties reaching 4% of annual turnover.

Most companies affected have no idea. The claims are spread across product pages, about pages and packaging images written over years by people who meant well.

The tools that existed were either keyword lists that flag "green" wherever it appears, or law-firm audits costing thousands. Neither helps someone with 300 product pages and a September deadline.

EcoClaim was built to sit between the two: read the whole site, cite the actual rule, and say what to change.

EcoClaim.eu's answer

Next.js and TypeScript on Vercel, with PostgreSQL for storage.

Analysis of texts and images runs on LLMs specifically trained against a structured rule set derived from the EmpCo Directive and UCPD Annex I, with national overlays per member state. A separate crawler service handles page fetching and rendering.

The rules are versioned and deterministic where they can be — the same page scanned twice produces the same findings.