Statamic Greenwashing Checker

Flags greenwashing risks in Statamic entries while editors write. Covers the EU rules on environmental claims (Directive 2024/825, “EmpCo”), which apply from 27 September 2026.

Phrases like “eco-friendly”, “climate-neutral thanks to offsetting” or “100 % green” become per-se violations across the EU under EmpCo. This addon sends entry text to an LLM of your choice. It marks risky claims in the field where they appear, explains the risk and suggests a rewrite that editors can apply with one click.

The addon is a public beta. It flags potential risks. It is not legal advice and does not establish compliance.

Features

• Checks while editing: a sidebar panel checks unsaved text. Findings appear below each affected field, including fields nested in Bard, Replicator, Grid and Group. Risky phrases in Bard are highlighted.
• Suggested rewrites: each finding has one to three replacements. Apply replaces only the quoted phrase and keeps the editor’s undo history.
• Nine risk categories, mapped to the EmpCo/UCPD articles. German sites can also show the matching UWG references.
• Context-aware: your real certificates and licences (“substantiations”) and house rules go into the prompt, so documented claims are not flagged. B2B collections skip the consumer-only per-se rules.
• Review workflow: editors can ignore a finding, or approve it with a documented reason (separate permission). Approvals are stored as content files you can commit to Git.
• Publish warning: publishing an entry with open high-risk findings asks for confirmation. It never blocks.
• Bulk scans: scan saved content from the CLI or from a CP utility that ranks entries by risk.
• Any Prism provider: OpenAI, Anthropic, Gemini, Mistral, OpenRouter, Ollama and more, via Prism. Results are cached per text, so unchanged copy is never sent twice.
• German and English UI. Explanations follow the editor’s CP language; suggested rewrites follow the entry’s language.

Review all findings	Scan saved content

Requirements

• PHP 8.3+
• Statamic 6.20+, or Statamic 5.74+ on Laravel 11/12 with a reduced feature set (see Statamic 5)
• An API key for an LLM provider supported by Prism. Each check is one paid API call.

Installation

composer require el-schneider/statamic-greenwashing-checker

Set a model and its provider key in .env. The model string is <prism-provider>/<model>:

GREENWASHING_CHECKER_MODEL=openai/gpt-6-luna
OPENAI_API_KEY=sk-...
 
# or, for example, through OpenRouter:
# GREENWASHING_CHECKER_MODEL=openrouter/openai/gpt-6-luna
# OPENROUTER_API_KEY=sk-or-...

Provider keys come from Prism’s own config/prism.php. Until a key is configured, the panel shows a setup notice and nothing is sent.

Open any entry: the Greenwashing check panel appears in the sidebar for all collections.

Configuration

Editors and admins configure most things in the CP under Addons → Greenwashing Checker → Settings:

To change the model, timeout or the defaults for these settings, publish the config:

php artisan vendor:publish --tag=greenwashing-checker-config

The comments in config/statamic/greenwashing-checker.php explain each option. CP settings override the config file.

Exclude fields

Fields with internal or non-marketing text can opt out in the blueprint. On a Replicator, Grid, Group or Bard field this excludes everything inside it:

-
  handle: internal_notes
  field:
    type: textarea
    greenwashing_checker: false

Panel placement

placement can be sidebar (default), tab (separate tab) or none. With none, add the Greenwashing Checker fieldtype to a blueprint where you want the panel.

Customize the prompt

The system and user prompts are Blade views. Publish them to adjust wording or add rules:

php artisan vendor:publish --tag=greenwashing-checker-prompts

Cached results do not notice prompt edits. Delete storage/statamic/greenwashing-checker/units/ to have all text checked again.

Permissions

Running a check requires permission to edit the entry.

Scanning existing content

php please greenwashing:scan                      # all enabled collections and sites
php please greenwashing:scan --collection=products --site=default
php please greenwashing:scan --dry-run            # count unchecked text, no API calls
php please greenwashing:scan --prune              # delete stored results no entry uses

The CLI writes explanations in the app locale. The Utilities → Greenwashing check page starts the same scan from the CP and lists entries by open high-risk findings. Long scans need a queue worker. The default sync queue runs them inline.

Data, storage and cost

• What is sent: the text of changed fields, a capped amount of surrounding entry context (max_context_chars), your substantiations and house rules. This can include unpublished copy. Opt out sensitive fields.
• When: on Check, one second after an editor leaves a changed field (check_on_blur), and on scans. Set check_on_blur to false for manual-only checks.
• Cost: one structured LLM call per check, for uncached text only. Results are keyed by text and analysis settings, so unchanged copy reuses its findings.
• Storage: results are stored in storage/statamic/greenwashing-checker/ without expiry. Approvals are stored in content/greenwashing-checker/approvals/. Commit and back up approvals like other content.
• No telemetry. Provider retention policies apply.

Statamic 5

Statamic 5 has no addon settings API and no Inertia CP. Analysis, caching, the scan command and the review sheet work the same way. These parts are Statamic 6 only:

• CP settings. Configure through the config file instead; substantiations, house rules, audience and per-site references stay empty.
• Inline field boxes, Bard highlights, publish warnings and approvals. The Statamic 5 panel lives in the sidebar, and Apply works on top-level text, textarea and markdown fields.
• The interactive scan utility. Statamic 5 gets a read-only list; use greenwashing:scan to scan.

Limitations

• Only entry text fields are checked. Assets are not: PDFs such as data sheets or brochures, images, logos and labels can still contain risky claims.
• The model can miss claims or flag harmless ones. Truncated context can hide a qualification made elsewhere on the page.
• Entry text is passed to the model as data, but prompt injection remains possible.
• Legal references cover the EU and Germany. Other national implementations can be contributed, see CONTRIBUTING.md.

License

MIT. Support through GitHub issues, no SLA.