Ask in plain English. Get answers your auditor can verify.

The Certigence assistant is grounded on your evidence and prompted to cite the ESRS clause, evidence file or survey row behind every claim it makes.

Four jobs it was built for

  • Draft disclosures: Pull together an ESRS narrative, a TCFD statement or a SFDR article — from your own evidence base.
  • Answer RFIs: Paste an investor, bank or customer ESG questionnaire; get footnoted answers for review.
  • Flag gaps: Tell it the standard you report against; it lists missing data points, weak evidence and likely auditor questions.
  • Localise: Translate disclosures and narratives across English, Norwegian, Swedish, Danish, and German without losing fidelity.

AI assistant

  • Cited — Answers link back to source — ESRS clause · evidence file · survey row
  • Zero — Prompts used for training — Your tenant data stays yours
  • 5 — Frameworks referenced — ESRS · CSRD · SFDR · TCFD · GRI
  • SCC — GDPR transfer safeguard — Claude + GPT-4 class models, encrypted at rest and in transit

Every claim has a citation

The assistant only answers from your tenant's evidence. If it doesn't know, it says so. Every sentence links to the source — not a synthetic summary.

  • Inline citations on every generated paragraph.
  • Retrieval transparent — see exactly which documents were used.
  • Refusal by default when evidence is thin, rather than confabulation.
  • Preserves verbatim quotations when the auditor needs them.

Your data is not training fuel

Customer prompts, responses and evidence are never used to train shared models. We use enterprise-tier endpoints with tenant-scoped retrieval; retention terms are configured with the model provider.

  • Zero-retention contract terms with the model provider where available.
  • Tenant-scoped retrieval: no cross-customer bleed, enforced in DB.
  • Encrypted at rest (AES-256) and in transit (TLS 1.3).
  • GDPR by design; security practices documented on request.

How the assistant answers

  • Retrieve: Tenant-scoped search over your evidence, disclosures and structured data.
  • Reason: Claude or GPT-4 class model with tool use for calculations and lookups.
  • Cite: Every sentence gets a footnote linking back to the source.
  • Log: Full transcript saved for review, reuse and audit.

Put the assistant to work on your next disclosure

FAQ

Which model is it?

We route to the strongest appropriate model per task — Claude Opus/Sonnet and GPT-4 class for narrative generation, faster models for classification. The choice is transparent in the audit log.

Can I restrict it to specific evidence sources?

Yes. Per workspace, you can scope retrieval to your own uploads, only verified evidence, or a specific ESRS topic.

How do you prevent hallucinations?

Retrieval-augmented generation with a refusal policy when evidence is thin, plus a lightweight post-hoc citation verifier. Methodology and versioning are public.

Is it available in all my languages?

English, Norwegian, Swedish, Danish, and German are first-class. Additional languages on request.