CSR Reporting: How to Prepare a Clear, Evidence-Based Report

What is CSR reporting?

CSR reporting is how a company documents and communicates its social, environmental, ethical and community responsibilities: the scope it covers, the topics that matter, its policies and targets, the results it can evidence, and the limits of that evidence. Readers include employees, communities, investors, boards and regulators.

Frameworks shape what is disclosed; the evidence underneath decides whether a reader can trust it. The weakest section is often community investment, where results arrive from many grantees, each counting in its own way, and get stitched together when the report is due.

Which CSR reporting standards should you use?

Choose by purpose and applicability: GRI for public reporting on impacts, IFRS S1 and S2 or SASB for investor-oriented disclosure, and ESRS where EU rules bring the company into scope. The summary below is an orientation, not a test of legal applicability.

Scroll horizontally to see all columns →

CSRD is legislation and ESRS are the reporting standards under it, so the rows are not all the same kind of document. Start with each publisher, record the edition you applied, and confirm applicability with qualified advisers rather than a comparison table or a software label.

What must a CSR report contain?

Scope and method, governance, material topics, performance against targets, people and communities, limitations, and assurance status, each backed by evidence you keep. Exact disclosures depend on the standards you apply.

Scroll horizontally to see all columns →

How do you prepare a CSR report, step by step?

Set the boundary and standards first, define every metric and its owner, collect and check the evidence, and only then write and design. Nine steps cover most reports.

1. Set audience, period and boundary. Name the entities, value-chain inclusions and exclusions.
2. Identify applicable standards. Confirm the rules in force for the period.
3. Determine material topics. Record the method, stakeholders consulted and approvals.
4. Define every metric and owner. One dictionary row each: definition, unit, boundary, formula, source, owner, reviewer.
5. Collect evidence as it arrives. Operational records, grantee reports, surveys and feedback, with source and period attached.
6. Check calculations and exceptions. Units, duplicates, missing periods, denominators; keep negative results visible.
7. Draft from approved evidence. Separate measured results, estimates, commitments and interpretation.
8. Review, assure and approve. Owners verify facts, finance checks figures, legal reviews statements, leaders sign off.
9. Publish and keep the register current. Keep a dated copy of what you published, and log corrections rather than regenerating figures.

Ownership sits at the level of each claim, not each chapter. External assurance may test selected disclosures, but it does not move management's responsibility for the report. For metric definitions, see the CSR metrics guide.

How do you report on a grantee or community-investment portfolio?

Treat it as a portfolio report: agree a few metrics and their definitions with every grantee, check each report against that agreement, and add up only the numbers that match. The CSR team is the funder here, whatever the report's cover says.

Picture the fictional regional workforce fund used throughout the course as a company's community-investment program, funding four job-training partners, A to D. The agreement covers enrolled (unique people), completed, placed in a job within 90 days of exit, retained at 12 months and starting wage by track.

Three grantees counted placements within 90 days: 42 at A, 31 at B and 27 at D. Partner C counted within six months and reported 55.

The published line reads: "100 people placed in a job within 90 days of exit, from three of our four training grantees; a fourth reported on a six-month window and is recounting." A total of 155 would look better and mix two measures, and in a public report that is a claim anyone can challenge.

Spend and volunteer hours belong in the report too, labelled as inputs and outputs. The outcome claim is the placement line, and it rests on the shared data dictionary you and your grantees agreed at the start.

How do you compare CSR results with outside data?

Only on the same definition, for the same place and period, and with every known difference written beside the comparison. An outside number answers "compared with what?"; it does not show that your funding caused the result.

For starting wage, map each training track to an occupation code, such as SOC 31-9092 for a medical assistant track, and take the BLS Occupational Employment and Wage Statistics hourly figure for the same area and year. New hires usually start below the median for everyone in a job, so show the 25th percentile beside it, and note that BLS covers all workers, not new hires.

Partner C's single average wage cannot enter the comparison, because it mixes tracks; that is why the fund asked, "Can you split starting wage by track?" Public records help elsewhere too: ProPublica's Form 990 search for a grantee's filings, and Department of Labor enforcement data before you name an employer partner. Compare your results with outside data walks through each source.

How do you make each CSR claim checkable?

Keep a claim register: one row per published statement, holding its wording, definition, source, evidence type, limits and approver. Writers and reviewers then work from the same record, and nothing gets stronger in editing than the evidence allows.

Scroll horizontally to see all columns →

Keep distinctions visible through editing: a grantee's policy is not evidence it works, and a participant's comment is not a measured change. Make every number match its source shows how to write the source note behind each row.

Can AI help write a CSR report?

Yes, for drafting from approved evidence, flagging gaps and assembling tables; not for deciding materiality, choosing favorable evidence or filling missing values. People approve every claim.

Draft the community-investment section of our CSR report.
Inputs: our claim register [paste], each grantee's checked results with statuses [paste], and any outside benchmark with its source, area and year [paste].
Rules:
- Use only numbers in the inputs. Do not calculate new totals or percentages.
- State totals only for grantees marked "matches"; name every held grantee and the reason.
- Label spend and volunteer hours as inputs and outputs, not outcomes.
- Place each benchmark beside its definition differences; never say the program caused a result.
- If something is missing, write "not in our data".
- Cite the register row after every claim.

Sopact Sense supports the evidence side. Each grantee keeps one ID across every report, each grantee's data can sit in its own folder while the CSR team sees aggregated results, field selection keeps names out of what AI receives, and every line of an answer links to a record you can open. Public data loaded as an ordinary survey follows the same rules.

This two-minute video follows portfolio data from a CRM, documents and survey responses into one connected record, then a report for each recipient. Watch for one set of records serving different readers, as your CSR report, board pack and grantee feedback should.

What can a CSR report not claim?

It can report what grantees observed and what adds up across them; it cannot claim the company's funding caused those results without a design that tests it. Say which kind of statement each one is.

Spend, volunteer hours and people trained are outputs; placement and retention are outcomes, and even they are counts, not proof of cause. A benchmark shows how results compare with local figures, not what would have happened without the program. Follow-up data depends on who answered, so show who did not.

A source link does not guarantee a correct conclusion. When AI drafts a section, a named person checks each line against the record, and authorized leaders decide what is published.

Start with one community-investment claim this cycle

Pick one outcome every grantee can report, and take it from agreement to published line before widening the scope.

1. Choose the claim, such as job placement, and write its definition, window and unit as one dictionary row.
2. Confirm the definition with every grantee before the period starts.
3. Check each report against it and mark matches, different, missing or not split.
4. Add only the matches; write each held grantee into the claim with its reason.
5. If you add a benchmark, match its definition, area and year, and list the differences.
6. Fill the claim register row and get a named sign-off.

After the first cycle you have one community-investment claim a reader can check, a register row to copy for the next metric, and a list of definitions to tighten in next year's grant agreements.

Frequently asked questions

Is CSR reporting mandatory?

It depends on the jurisdiction, the company and the reporting period. Some companies fall under rules such as the EU's CSRD, while others report voluntarily against GRI or investor-oriented standards. Confirm current requirements with qualified advisers rather than relying on a generic framework table or a software vendor's label, and record the standard and edition you applied.

What is the difference between CSR, ESG and sustainability reporting?

CSR reporting explains how a company acts on its responsibilities to society. ESG reporting organizes environmental, social and governance information for risk and investor decisions. Sustainability reporting covers an organization's impacts and related risks and opportunities. One publication may use all three labels, so the scope and the standards applied matter more than the title on the cover.

Does every CSR report require double materiality?

No. Materiality approaches differ across standards. ESRS use double materiality, covering impact and financial materiality, while GRI focuses on an organization's most significant impacts. Determine which approach applies, then document the evidence, the stakeholders consulted and the judgment behind each conclusion. A social-value calculation, however careful, is not a financial-materiality assessment.

Who should approve a CSR report?

Follow the company's governance process. Data owners verify their evidence, finance checks calculations and boundaries, legal or compliance reviews material statements, and authorized executives or the board approve publication. Record who approved each claim. External assurance may test selected disclosures, but it does not replace management's responsibility for what the report says.

How should grantee outcomes appear in a CSR report?

Report totals only for grantees that counted the same way, with the definition and denominator beside each, and name any grantee held out and why. Label spend and volunteer hours as inputs and outputs. Where you compare with outside data, use the same definition, area and year, and say what the comparison cannot show.

Can AI write the CSR report?

AI can organize approved evidence, flag missing items and draft sections, provided it works only from your claim register and checked results and writes "not in our data" where something is missing. People remain responsible for materiality, methods, interpretation and publication. Check every AI-drafted figure against its source before it goes to review.