Run a Leadership 360 or Engagement Survey on Org360 →

Our AI Policy

Nonprofit Builder’s AI guidelines for consultants, as a shared standard of professionalism and integrity. These guidelines aim to help us use AI responsibly, transparently, and always in ways that strengthen trust with our clients and partners.

1. Safety and privacy: do no harm

The following are essential guidelines to use AI safely:

  • Always and only use a paid account for any client work. This allows you to opt out of having your data used for training, meaning fragments of it may be contained in the output when the model is used. If you use a free account, you have no choice, your data will be used for training.
  • Be sure you opt out of using your data for training purposes. ChatGPT has an option for this, but you may be shocked to learn that by default it is turned off, meaning they will be sharing your data for training. So make sure you find and correct that right now (turn off “Improve the model for everyone”).
  • Never put personal data (such as person names, emails, organization names) into an AI prompt. And never put private client data (such as strategy documents, interviews, MEL data) into an AI without the client’s explicit consent.

Instead of Gemini, Claude or ChatGPT, where you can never be sure of what is happening with your data in the background, Tinfoil.sh offers a promising privacy model. Tinfoil secures both the server environment and the communication channels. Still, your requests may be transmitted to and processed in the United States. Tinfoil allows you to choose between several open AI models like DeepSeek or GPT-OSS for only $10 monthly. These models may, however, be less advanced than current proprietary ones.

However, even this approach is not bulletproof. Tinfoil.sh has servers in the USA and has to abide by US jurisdiction. When handling sensitive information that should not be exposed to potential US court orders, even this option should be avoided. A European, or better yet locally hosted, alternative within a carefully selected jurisdiction should be preferred.

There are several reasons why a client may not want you to use AI. Security and privacy is one of them. Or the client may simply prefer that you do all the work yourself: they may feel that using AI is taking a shortcut. Or they may not be comfortable with AI for ethical reasons, such as the consequences for the environment. Regardless of their reason, they are your client, and out of respect, you need to follow their wishes.

This leads us to the following three guidelines:

  • Always discuss and agree on your intended use of AI at the outset of the engagement, before you even submit your SOW. This will protect you from painful misunderstandings further along. We may even add an AI field to the SOW to make sure this topic is covered.
  • If your client is not happy with your planned use of AI, or if you have not expressly discussed it, then just don’t. It's as simple as that. Out of respect for the client, and to protect your relationship of trust with the client.
  • Document your use of AI: provide an “AI Usage Sheet” as part of your final report or handover, detailing tool names (e.g., Mistral, DeepSeek), data entered, and client adaptations.

What you may do without client consent

You can use AI freely for activities that do not involve client-specific information and do not generate deliverables. These uses are equivalent to a web search. Examples include:

  • Researching funders, trends, or best practices.
  • Exploring your client’s thematic domain.
  • Learning new OD methods or tools.
  • Asking general questions that do not reveal client identifiers.

When you must obtain client consent

Consent is required whenever AI touches client information or contributes to a client-facing output, including when you:

  • Enter client data, identifiers, descriptions, internal issues, or draft text into an AI.
  • Use AI to produce or rewrite any part of a deliverable.
  • Ask AI to analyze, recommend, or diagnose client-specific situations.

Grey zone: using AI for quality assurance, asking AI to check grammar, clarity, tone, or structure, or “discussing” a client issue with AI in vague terms. These uses are allowed only if they do not reveal sensitive or identifying information. When in doubt, generalize or anonymize, or request consent.

3. Quality assurance

You are the consultant (you're in the driver's seat)

AI is an assistant. It cannot replace your thinking, judgment, or deep understanding of the client. You must frame the task, direct the work, and ensure the final result reflects your professional input. For example, don’t simply ask the AI for a "job description", but give it context about the role as it applies to your client’s situation and challenges.

Check carefully for mistakes and bias

When you review that draft, be ruthless.

  • Check for Mistakes: Read every single word. AI "hallucinates" and makes up facts, citations, and data. You are responsible for factual accuracy.
  • Check for Bias: AI models are trained on existing data, which means they can have a strong "conservative" bias (politically neutral, but in the sense of defaulting to the status quo and resisting new ideas). Watch for this, as well as cultural or social biases.

Leaving even a small mistake can lead the client to think the whole document was produced by an AI and this will discredit you. So be extra careful with this. Also beware to polish out typical AI writing style or its use of punctuation (like “em dashes”).

AI docs are always a draft. Make them your own.

AI output is only a first draft. Never, ever copy-paste it directly into a client deliverable. You must edit, rewrite, challenge, and build upon it. If it feels generic, it's because it is. Your job is to add your unique insight, creativity, and strategic value.

4. Don't bill for time saved with AI: reinvest it

It is not ethical to bill for hours you didn't work. If AI speeds up a task, reinvest that "saved" time. Use those extra hours to go deeper on analysis, train their team, co-create strategies, or offer follow-up coaching. This is how you turn efficiency into client value. In full transparency with your client of course.

5. Educate yourselves to educate your clients

Be transparent. Explain how you plan to use AI (e.g., "I'll use it to summarize our research and generate a first draft of the market scan"). Reassure them it will not replace your strategic brain, judgment, or creative input. Where appropriate, help them build capacity by showing them examples from their own context. But don't push if you see they are reluctant.

If you plan to use AI, you must invest in continual learning. Learn the tools, how to use them, and their strengths and limitations.

  • Be curious and try to "think AI first" when you start a new task.
  • AI is also about process automation, data analysis, and more, beyond LLMs.
  • Evaluate alternatives such as Mistral (France), Qwen (Alibaba, China), DeepSeek (multilingual, open research), or BLOOM (multi-language, open science project), and tailor the tool choice based on the nonprofit’s region and community.
  • Follow what's going on in terms of privacy of AI tools to make sure you use them safely.
  • Search the "Hugging Face model hub" for AI models designed for diverse languages and regions.

6. Mind data protection and AI regulation

The EU has passed the AI Act that is going to be applicable step by step. It has a broad extraterritorial reach. It suffices to use the output of an AI system within the EU. The main requirements are:

  • AI Literacy (Art. 4 AI Act): There are no required certificates, but you need some proficiency. Working through this document could be a start but additional training will be required in the future.
  • Transparency (Art. 50 AI Act, required starting August 2026): If you use generated AI images or text, make sure that you tag them (machine-readable) as AI generated. If you have other people than yourself directly interacting with AI, additional obligations will apply.
  • Be careful not to use AI where it is forbidden. Art. 5 AI Act lists forbidden uses that mostly fall into the domain of analysing or influencing humans.
  • There are high risk uses that require additional precautions starting August 2026 (see Art. 26 AI Act).

Data protection regulation applies on top of the AI Act.

7. Finally, appreciate your value as a human

Remember, you are a human working with humans. AI cannot offer genuine support, empathy, humor or a real connection. A perfect plan is just one piece of the puzzle. Helping your client build confidence, navigate their human challenges, and feel inspired: that is the value they will always be grateful for.

Questions about this policy? Write to support@nonprofitbuilder.org.