Curated sources, not training memory
The knowledge base holds the sustainability reporting legal acts, EFRAG implementation guidance and published sustainability reports from listed German companies. What is not in it is not offered as evidence.
CSRD · ESRS · Double materiality
Answers on CSRD sustainability reporting — with the source, the version and the date.
CSRD-GPT does not draw its answers from a language model’s memory. It draws them from a maintained collection of legal texts, guidance and published sustainability reports, in which the successive versions are held apart. You can see what every statement rests on.
Short answer
CSRD-GPT is an AI expert assistant for reporting under the Corporate Sustainability Reporting Directive (CSRD) and the European Sustainability Reporting Standards (ESRS). It answers technical questions, drafts report sections and runs gap analyses, working from a curated source collection in which ESRS (2023) and ESRS (2026) are kept separate. Every answer names its source. Built by impact-in-company for sustainability consultants, in-house teams, audit practices, law firms, agencies and reporting companies. Plans from €49 a month.
Why the version decides
Since 2026, several versions of the sustainability reporting standards apply side by side.
This is why a general language model becomes unreliable here: it has seen both versions in training and holds no dividing line between them. For a financial year 2026, that dividing line is the answer.
1,000 / 450
employees and €m net turnover — both thresholds must be exceeded for
the CSRD reporting obligation to apply.
Directive (EU) 2026/470 (“Omnibus I”), Official Journal 26 Feb 2026
61%
fewer mandatory data points in the revised sustainability reporting
standards compared with the 2023 version.
Delegated act C(2026) 5010 final of 3 July 2026
19 Mar 2027
deadline for Member States to transpose the amended directive into
national law.
Article 4 of Directive (EU) 2026/470
2 versions
ESRS (2023) and ESRS (2026) both apply to financial years 2026,
with an option to choose.
Transitional provision in C(2026) 5010 final
A dated breakdown with primary sources is on Who has to report, and when?
The difference
Not how the answer is phrased, but where it comes from.
A general model produces a plausible answer from training knowledge. On familiar topics it is often right — and you cannot tell from the answer when it is not. In an assured report, that is the problem.
Versions maintained separately
ESRS (2023) and ESRS (2026) exist as separate holdings, not as one merged corpus. That is what makes it possible to ask what changed between them, rather than only what “the standards” say.
A separate space per engagement
Your own documents sit in knowledge spaces kept apart by client or project. For a consultant with several clients, that separation is the precondition for bringing in any material at all.
Three use cases
Where the difference actually shows.
A general model answers a single technical question almost as well. These three tasks are where the two come apart.
Case 01
Which version applies to my financial year?
“What changes for E1-1 between ESRS 2023 and ESRS 2026, and which version applies to a financial year 2026?” — the question that shows whether versions are held apart.
See the version comparison → Case 02Gap analysis against your own documents
Holding a transition plan, a prior-year report or a data collection against the disclosure requirements and naming what is missing. That requires an isolated knowledge space.
See the gap analysis → Case 03Verification down to the citation
Not just reading a reference but opening the passage and checking it. The point at which a formulation becomes an auditable working step.
See source verification →Who it is for
Five working situations, five different questions to the same body of sources.
Sustainability consultants
Independent or in small practices: more engagements without buying in junior capacity, at the same technical depth.
Read more →In-house sustainability teams
Producing the report internally without triggering an advisory engagement for every question of interpretation.
Read more →Audit and assurance
Holding disclosures against the legal text, following materiality assessments, taking citations into the working papers.
Read more →Law firms
Settling questions of interpretation against the wording, keeping versions apart, tracking national transposition.
Read more →Mid-sized companies
First the question of whether an obligation exists at all — and if so, from which financial year and to what extent.
Read more →Agencies
Writing the report on a client’s behalf — with the disclosure requirement, the version and the citation behind every chapter.
Read more →Adjacent regulations
PPWR, EUDR, ESPR, the EU Construction Products Regulation, EcoVadis: same construction, different knowledge base. What is in preparation.
Read more →Knowledge base
What is in the collection.
Legal acts and standards
- Delegated Regulation (EU) 2023/2772 — the sustainability reporting standards, 2023 version
- Delegated act C(2026) 5010 final — the revised standards, held separately
- Directive (EU) 2022/2464 (CSRD) and its amendment by Directive (EU) 2026/470
- Directive (EU) 2025/794 postponing the application dates
- The voluntary standard for companies outside the reporting obligation
Interpretation and practice
- EFRAG implementation guidance, including on double materiality
- Published sustainability reports from listed German companies as drafting evidence
- Commentary on contested disclosure requirements
- Your own documents in a space kept separate per client or project
The collection is maintained, not accumulated: superseded legal versions are removed rather than supplemented, so that an answer cannot rest on a version that no longer applies.
Frequently asked
What to know before you try it.
What is CSRD-GPT?
CSRD-GPT is an AI expert assistant for reporting under the Corporate Sustainability Reporting Directive (CSRD) and the European Sustainability Reporting Standards (ESRS). It answers technical questions, drafts report sections and runs gap analyses, working from a curated body of legal texts, guidance and published sustainability reports. Every answer names its source.
How is CSRD-GPT different from ChatGPT or Microsoft Copilot?
In where the answer comes from. A general language model produces a plausible answer from training knowledge; it has seen ESRS 2023 and ESRS 2026 as the same material and holds no dividing line between them. CSRD-GPT retrieves passages from a maintained source collection in which the versions are kept apart, and returns the citation with the answer. The full comparison.
Are my client documents used for training?
No. Uploaded documents sit in their own knowledge space, separate from the rest of the system, and are not used to train models. On the Premium plan you can create a separate space per client or project so that material does not travel between engagements.
Does CSRD-GPT replace professional review?
No. It produces evidenced intermediate results — citations, drafts, gap lists — that still need professional review. The citation exists for exactly that purpose: you can check every statement against the legal text instead of trusting a formulation.
Access
Request access.
CSRD-GPT is being built and is currently being tested with consultants, in-house teams and auditors. If you would like to take part, get in touch — you will hear from a person, not from a queue.
- No purchase, no subscription
- You help decide what gets built next
- Replies come from the people building the product