GRI Reporting with Claude — Instructor Demonstration
By the end of this session you can
- Build a reporting Project that encodes house voice, prohibited language and evidence rules
- Run a disclosure-level gap analysis across a full data room at a scale no human skim achieves
- Draft disclosure text where every claim is tagged to a source, and stress-test it as a sceptical assurer would
- Generate and quality-check a content index, and map the same content to a second framework
Three-hour agenda
| 00:00–00:15 | The new risk | Fluent, unevidenced prose reads better than the true version. The rule for the module: every sentence carries a source, or it does not ship. |
| 00:15–00:45 | Project setup for reporting | Knowledge base: GRI extracts, last year's report, the materiality output, the Week 2 inventory, the data room index, a style guide. Custom instructions encoding house voice, banned superlatives, the omission policy, the citation requirement. |
| 00:45–01:25 | Live build 1 — gap analysis | Feed the disclosure list and the data room. Produce a gap register with exact source locations, owner, effort and RAG status. Watch it catch three things a skim missed — then invent a plausible source, and see the exact-location rule expose it immediately. |
| 01:25–01:35 | Break | |
| 01:35–02:15 | Live build 2 — drafting | Generate GRI 3-3 and topic disclosures from evidence only, with refusal on unevidenced points. Two mandatory passes: "tag every claim to its source row", then "argue against this text as a sceptical assurance provider". |
| 02:15–02:45 | Live build 3 — index and reuse | Generate and QA the content index. Then map the same disclosures onto BRSR, IFRS S1/S2, ESRS or a buyer questionnaire without rewriting — the interoperability payoff. |
| 02:45–03:00 | Verification protocol | Source-tag audit, reconciliation of every number to the Week 2 inventory, prohibited-claims check, handoff to legal and comms. Compared against the Day 6 times and word counts. |
Claude capabilities taught
- Long-document and multi-document work across a full data room
- Custom instructions as an editorial policy, not just a persona
- Structured extraction into a gap register with mandatory source locations
- The citation-tagging pass as a hallucination detector
- Adversarial self-review: "challenge this as an assurer would"
- Cross-framework mapping from a single content set
Exercise
Class red-team of the instructor's drafted disclosure: find the unsupported claim that was deliberately left in.
Take-home
Day 8 brief issued: continue with your Week 2 company, so the inventory you built becomes the emissions data you now disclose.
Apply →