ESG reporting has become one of the most resource-intensive obligations in the finance function. Between the CSRD, ISSB standards, CDP questionnaires, and an ever-expanding list of investor requests, controllers and CFOs are being asked to produce disclosure-grade sustainability data on a quarterly cadence — often using the same spreadsheets and manual processes that were built for a world where ESG was a footnote in the annual report. The result is predictable: version control chaos, inconsistent emission factors, audit trails that live in someone’s inbox, and reporting cycles that consume weeks of analyst time.
The deeper problem is not the data itself. Most organisations already capture energy consumption, headcount, supplier spend, and travel volumes somewhere in their systems. The friction sits in the translation layer: mapping operational data to framework-specific metrics, reconciling boundaries between financial and operational control, documenting methodology decisions, and producing narrative disclosures that satisfy both regulators and internal audit. That translation layer is where Claude Code changes the economics of ESG reporting.
Claude Code is not a dashboard or a reporting platform. It is an agentic assistant that reads your files, understands your folder structure, and executes multi-step tasks across documents, datasets, and templates. For ESG teams, this means you can point it at a folder of utility invoices, a GHG protocol mapping file, and last year’s disclosure, and ask it to produce a reconciliation with a documented audit trail. The work that used to take a controller three days of copy-paste and cross-referencing can be compressed into a supervised session where the analyst reviews outputs rather than assembles them.
Why the prompt structure matters more than the model
Most ESG automation attempts fail at the briefing stage, not the execution stage. A vague instruction like “summarise our Scope 3 data” produces a vague output that no auditor will accept. The prompts below use a structured anatomy: task, success criteria, file context, reference pattern, and a mandatory clarification step before execution. This mirrors how a good controller would brief an analyst — and it is the single biggest lever for output quality.
First, read these files completely before responding:
[raw_energy_data.csv] — monthly electricity, gas, and fuel consumption by site, with meter references
[emission_factors_2026.md] — approved emission factors, sources, and applicable regions
[prior_year_disclosure.md] — last year’s published Scope 1 and 2 figures with methodology notes
[entity_boundary.md] — list of entities in scope under operational control approach
Here is a reference for what I want to achieve:
[Upload last year’s reconciliation workbook as markdown, showing how each line item maps from source data to disclosed figure]
Here’s what makes this reference work:
Every disclosed number has a source column, a factor column, a calculation column, and a reviewer note. Boundary exclusions are documented inline. Restatements from prior periods are flagged separately. Tone is factual, no narrative padding, and every assumption is labelled as an assumption.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: Reconciliation table in markdown, one row per site per scope, plus a methodology note of 300-400 words
Recipient’s reaction: The external auditor should be able to trace any disclosed figure to a source row without asking a follow-up question
Does NOT sound like: A sustainability marketing summary, a high-level dashboard commentary, or anything with rounded figures presented as exact
Success means: Zero unresolved mapping gaps, every exclusion documented, and the methodology note passes internal review without edits
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
Notice what the prompt does not do. It does not ask Claude to “be accurate” or “follow best practices.” It specifies the artefact, the traceability requirement, and the failure mode to avoid. For a CFO reviewing output, that traceability requirement is the whole game — an ESG number without a source trail is a liability, not a disclosure.
From reconciliation to narrative: the second half of the cycle
Once the numbers are locked, the next bottleneck is the narrative. CSRD and ISSB disclosures require explanatory text that connects metrics to strategy, explains year-on-year movements, and describes governance around the data. This is where finance teams often hand off to sustainability or communications — and where consistency quietly breaks down. A second structured prompt keeps the narrative anchored to the same source files, so the story and the numbers never drift apart.
First, read these files completely before responding:
[reconciliation_final.md] — locked Scope 1, 2, and 3 figures with variance to prior year
[strategy_notes.md] — approved decarbonisation initiatives, timelines, and capex commitments
[regulatory_checklist.md] — required narrative elements per ESRS disclosure requirement
[tone_guide.md] — house style, prohibited claims, and approved terminology
Here is a reference for what I want to achieve:
[Upload a peer company’s management commentary section or your own prior-year version as markdown]
Here’s what makes this reference work:
Each metric movement is explained with a cause, a magnitude, and a forward-looking statement. Claims are hedged where evidence is incomplete. No superlatives. Every initiative mentioned ties to a capex line or a dated milestone. Paragraphs are short and each opens with the metric, not the narrative.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: Management commentary, 800-1000 words, structured by ESRS topic
Recipient’s reaction: The reader should understand what moved, why, and what happens next — and should not be able to accuse us of greenwashing
Does NOT sound like: A press release, an investor pitch, or a document that promises outcomes we have not funded
Success means: Every claim maps to a source file, every forward statement has a date or a dependency, and legal review raises no unsupported assertions
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
The practical tip for finance leaders evaluating this approach: start with the reconciliation prompt, not the narrative prompt. Numbers first, story second. If Claude Code can produce a traceable reconciliation from your existing data exports, you have proven the workflow end to end — and you have a defensible artefact to show internal audit before you scale it across entities.
What to try next: run the first prompt against a single site or entity, time the cycle, and compare the output against your current process. Then extend the same structure to Scope 3 categories, water and waste metrics, or the social disclosures under ESRS S1 and S2. The prompt anatomy stays the same — only the context files change.
Published on 26 September 2026 on growwithgpt.com