Key takeaways
AI works best on the preparation layer of ESG reporting, such as aggregating and standardizing data, while consequential judgment calls stay with a person.
AI-drafted narrative needs a documented review trail and, under the EU AI Act, must be labeled unless a qualified person reviews and owns it.
Corporations should put AI's efficiency gains toward easing the burden on nonprofits.
As an impact leader, you own the sign-off on your environmental, social and governance (ESG) report, but getting to the end result can be time-consuming. Giving, volunteering and grants data all sit in separate systems, so someone must reconcile them by hand and turn the data into a narrative, all before the deadline.
This is where AI can shine. It can help with collecting and standardizing data, mapping it to a framework and drafting the narrative.
What is AI for ESG being used for today?
The unglamorous prep work is where AI is helping teams today. In practice, AI is doing four things:
- Finding the data
- Standardizing it
- Flagging what's missing
- Drafting narrative from program records
None of these are judgment-free tasks. Deciding how to resolve a data conflict, which gap-filling approach makes sense or how to phrase a narrative accurately all call for a person's read on the program. What AI changes is where that judgment gets applied. Instead of starting from raw, scattered records, the person reviewing the work starts from an AI-assisted first draft.
KPMG puts fragmented data collection and management at the top of the 2026 sustainability reporting challenge list, with leading practice shifting from manual to automated and reactive to predictive.
AI-assisted narratives
As discussed, AI can help provide a first draft of the narrative based on patterns in the data. But keep in mind, AI will not have firsthand knowledge of the program. It's up to you to refine and revise based on the outcomes of your program.
If your report goes through assurance, there is an added layer of rigor inside an assured statement. Your provider will ask where your reporting numbers came from, how the document was prepared and what controls sat around it. Those are factors the IAASB's implementation guidance tells practitioners to weigh. ISSA 5000, the purpose-built sustainability assurance standard, takes effect for periods beginning on or after Dec. 15, 2026, replacing the general-purpose ISAE 3000 (Revised). Adoption timing varies by jurisdiction, so confirm which applies to your next report.
Either way, three practices hold:
- Data lineage: Every figure AI produces or transforms needs a traceable path to a source record.
- Version history: Reviewers need to see what the model produced and what a person changed.
- Documented review: Sign-off has to exist as a record. Name who reviewed what, and when.
Does AI-drafted narrative have to be labeled?
Since Aug. 2, 2026, Article 50 of the EU AI Act has required a label on AI-generated text published to inform the public on matters of public interest. There's an exception: if someone with relevant knowledge and professional judgment on the subject matter reviewed it, and a person holds ultimate legal responsibility for publishing it, the label isn't needed.
AI for funding decisions
Funding is the clearest example of AI's role: supporting the decision-maker without replacing them. Only 23% of large corporate firms said AI should be used to decide grant approvals and denials, compared with 70% who expect it to be essential in identifying fraud and anomalies, according to the State of Corporate Purpose 2026 report. This will be welcome news for nonprofits, 51% of which are concerned about corporate donors using AI to score grant proposals, and 64% of which fear AI will miss the critical human nuance required to understand their true impact.
And among large corporate firms, 77% worried that using AI to vet and recommend nonprofits would reinforce existing funding biases, and 75% worried it would exclude smaller, community-led organizations. Those are the exact outcomes a person reviewing the work is positioned to flag.
The same caution should be applied for any ESG-adjacent decision with real consequences for people or communities:
- Access: Who gets into a program or receives a benefit.
- Disclosure: What a company says publicly about its own performance. That's a materiality question.
- Community effect: Which communities get prioritized, and which get left out. That's a question about who bears the cost.
In each case, AI can prepare the preliminary analysis the human decision-maker needs to make the call.
How do I start using AI in ESG reporting?
Start with the tasks where a wrong answer is visible and easy to correct, which usually means data assembly. Instead of pulling giving, volunteering and grantmaking numbers into three separate reports by hand, the Benevity Reporting Studio brings Donate, Volunteer and Grants Management data together, so less of the week goes to assembling numbers.
Then set the rule before you need it. Decide which calls stay with a person, document it and keep the trail behind anything AI drafts.
Offload busywork and give your CSR program the data foundation it needs to move faster.
Learn more about Reporting Studio, or explore an interactive demo.
Frequently asked questions (FAQs)
Which ESG frameworks does employee giving and volunteering data report into?
Contributions are reported as economic value distributed under Global Reporting Initiative (GRI) 201, and Business for Societal Impact (B4SI) is the established framework for measuring inputs, outputs and outcomes of community investment.
Is AI reliable for calculating greenhouse gas or emissions data?
AI can assist with emissions accounting but can't calculate emissions. Emissions accounting is a specialized discipline with its own verification standards, and International Financial Reporting Standards (IFRS) S2 requires measurement in accordance with the Greenhouse Gas (GHG) Protocol Corporate Standard.
How do I measure giving and volunteering outcomes, beyond activity?
Benevity partners with True Impact and the Impact Genome Registry for standardized outcomes measurement and benchmarking against comparable programs.
About the author
