How Giving Green Uses AI

Last updated: November 2026

Why We Publish This

This page explains where we use AI, where we don't use AI, and how we protect the people and organizations whose information we hold. We hope it also helps peer organizations working through the same questions.

AI tools and the norms around them change quickly. Therefore, this will be a living, breathing page. We'll make revisions as we learn, especially if we find we got something wrong.

Our Commitments in Brief:

• People make the final call on our recommendations.
• A human is accountable for everything we publish, and we check AI-assisted claims before publication.
• For non-public data (such as donor and grantee information), we only use enterprise-grade AI subscriptions that exclude our content from model training.
• We review this page regularly and update it as our practices change.

The Tools We Use

At Giving Green, we primarily use AI tools under enterprise plans, which exclude our content from model training. This means AI companies can't use our data to improve their models, so our information won't show up in answers to other users. Sensitive information, including donor and grantee data, goes only into these tools.

Before anyone uses a new tool for work, we check:

• The vendor's data terms
• How long the vendor keeps data
• Its security certifications
• Whether we can control who has access

We check again whenever a vendor changes its terms. When we use an AI notetaker in a meeting, we tell participants in advance or at the start of the call.

AI in Our Research

AI helps us work faster, but our researchers determine all conclusions and recommendations. We use AI tools to:

• Find sources
• Summarize literature
• Write and check analysis code
• Draft report sections
• Critique our own analyses before publication
• Copy edit (check grammar, spelling, etc.)

We don't use AI to interpret results, reach conclusions, or set our final recommendations. Donors rely on us for those judgments, so they stay with our researchers.

A person owns every output. Each report, evaluation, and recommendation has a staff member responsible for its accuracy, whether or not AI was involved. We treat AI output as unverified notes:

• When a model suggests a source, we confirm the source exists and says what we claim.
• When it produces a number, we reproduce the number ourselves before publishing it.

Protecting Donor and Grantee Information

Identifiable donor information stays in our donor database unless there's a specific reason to move it. When staff need AI help with donor-related work, we anonymize or aggregate the data first wherever we can. Names and contact details leave the database only for a defined purpose, such as drafting a note to a specific donor.

Confidential grantee and donor information goes only into tools covered by an organizational agreement. This covers personal data, donation amounts, unpublished financials, plans, and draft assessments from the organizations we evaluate. We use that material only in enterprise tools that don't train on our data. Where a confidentiality agreement or research ethics approval sets stricter terms, those terms apply.

Where We Draw the Line

Some uses of AI are off-limits for us regardless of the tool. We don't use AI to:

• Make final charity or grant recommendations
• Make decisions about individual donors, grantees, staff, or partners based on an AI score alone, or profile anyone by sensitive characteristics or demographics
• Infer people's emotions from calls, interviews, or other interactions
• Make final hiring or performance decisions

AI-generated Imagery

On our website, on social media, and in our communications, we use AI-generated images only for abstract or decorative visuals such as icons, textures, and headers. We never use them to depict real people, places, events, or communities.

This statement covers AI-generated imagery across all our channels.

Accountability and What's Next

If something goes wrong, we will act quickly. While Giving Green has not yet had an AI tool cause a security or data breach, we are putting best-practice processes in place to respond to such a breach. If an AI tool exposes donor or grantee information, we will work quickly to limit the damage and find out what happened. When deemed necessary by our own investigation and/or by law, we will notify affected parties.

Questions we're still working through. We don't have settled answers on everything, and we'd rather say so.

• AI's environmental footprint. AI tools run on energy-intensive computing. Our current view is that the productivity-enhancing benefits of AI outweigh the climate downsides, and we do not put climate-related restrictions on AI use. However, we will keep learning about how we as consumers may be able to reduce the carbon footprint of AI use.
• Disclosure of AI use. Norms are still emerging regarding disclosure of AI use in research and communication. We are still working to understand if there is a certain amount of AI use on a product that should warrant disclosure, and how to integrate this into our research and communications guidelines.

Review. We review this page at least annually and whenever our tools or practices change. The date at the top shows when we last updated it.

Contact. Send questions about how we use AI to hello@givinggreen.earth.