Your AI Questions, Answered
Recently, we hosted an AMA on AI for animal advocacy and you raised some really thoughtful questions. We passed those questions along to the AI experts on our team and now we're bringing their answers to you – so you can clear up your doubts, get unstuck, and get the support you need in your AI journey.
1. Should animal advocates be using AI at all, given the ethical and environmental costs?
The short answer: Yes, because opting out has a cost too, and we think it's higher than the cost of using AI thoughtfully. Factory farming kills tens of billions of land animals and trillions of fish every year. If a tool can help us do more research, reach more people, plan better campaigns, and free up staff time for the work that needs human minds and hearts, we think it's the more ethical choice to use it carefully than to refuse it on principle.
The longer answer: The concerns are real. Pollution from data centers, AI's water use, communities bearing the local costs, and the politics of some of the largest AI companies are all worth taking seriously. We take them seriously too.
And we still think animal advocates should use these tools. Not because the concerns don't matter, but because opting out has costs too, and for a movement this small fighting a problem this large, we think those costs are higher.
You can acknowledge the concerns and avoid over-reliance, while also accepting that animals are exploited at massive scale, that animal ag is already using AI to be more efficient at that exploitation, and that anything that makes you faster and stretches your team's limited resources means opting out has worse consequences than opting in.
This is similar to how many of us think about other compromised tools we already use. Smartphones are built with materials sourced from conflict zones. Social media platforms are owned by people whose politics most of us oppose. We use these tools because the alternative hands the field to people who have no qualms at all. AI is similar, though the stakes and the timeline are different.
A few things we'd add:
Per-prompt costs are smaller than the headlines suggest, but the system-level impact is still significant. Independent estimates put a typical text prompt's energy use in the range of a few seconds of TV viewing, and water use well below an hour of videoconferencing or streaming. That doesn't make the data center buildout fine. It just means that your individual queries are not where the most consequential decisions are being made. Those are made by infrastructure builders, regulators, and the companies themselves.
Use it for what it's good at, not for everything. A lot of unnecessary AI energy use comes from tasks that didn't need AI in the first place. There are also certain risks with using AI for tasks we're not familiar with – i.e., tasks where we can't evaluate whether the output is high quality. It's better to stick to tasks where you can tell whether the output actually makes sense.
Picking your provider matters. Where you can, choose providers that publish transparency reports, offer non-training options, and don't have your money flowing toward causes you'd otherwise oppose.
The question for us is not whether AI is perfectly clean, because it's not. The question is: given the world we're actually in, how can we use the tools available to us to do the most good while reducing the harms we can reduce? For animal advocates, we think the strategic use of AI is justified. Factory farming is one of the largest sources of suffering on earth. If AI helps us fight that system more effectively, refusing to use it may not be the more ethical choice.
2. What are the most useful ways nonprofits can actually apply AI to food systems work and animal advocacy?
We think about this in categories rather than individual tools, because tools change every few months and categories don't. Here's where the biggest leverage tends to sit:
Doing existing work better and faster. Writing (i.e. emails, social media content internal documentation), translation, grant applications, cleaning and analysing data, are all areas where a thoughtful AI workflow can take a multi-day task down to a multi-hour one.
Monitoring the world for opportunities and threats. This includes new factory farm permits, company expansions, legislation, funder priorities, media narratives. AI is useful for monitoring and catching things that would otherwise require someone’s constant attention.
Meeting help. AI can transcribe meetings and analyse them. AI can summarise meetings to extract next steps, catch you up on relevant material for meetings you missed, and analyse across multiple meetings for common patterns. If you connect AI to your other tools (Hubspot, Asana, Calendar, Gmail etc) AI can pull together large amounts of relevant information in minutes to prepare you for meetings.
Supporting research and strategy. Summarizing evidence, comparing options, identifying gaps in your thinking, pressure-testing assumptions, and understanding complex systems faster. Treat it like a research assistant who's read everything but hasn't lived anything, because that's roughly what it is.
Accelerating alternative protein and food systems research. AI is already being used in protein design, and GFI has highlighted AI-based work on optimizing cultivated meat media for cost, cell growth, and climate impact.
Helping more people take action. AI can help with campaign planning, diet change, and finding local opportunities to get involved.
Improving operations so staff can spend more time on high-value work. This includes CRM updates, intake forms, volunteer coordination, data cleanup, email triage, knowledge management.
The idea is that AI doesn't replace strategy, judgment, or relationships. Instead, it gives our movement more leverage on the parts of advocacy that are bottlenecked by time and capacity.
3. How do I encourage colleagues to try AI without becoming "that person" who won't shut up about it?
No individual staff member should feel like they have to be "the AI person" forever. Our top recommendation: push for org-level adoption. AI can likely help most jobs at all levels of seniority and experience. . Ideally your leadership is bringing AI into team meetings, planning processes, and eventually role expectations.
If your organization isn't there yet, here are a few things you can do:
Start at the top. If you get your managers, directors and leaders excited about AI, they’ll encourage the rest of the organisation to adopt it.
Show, don’t tell. "It used to take me hours every month to write these reports, but using AI they now take me minutes, do you want to see how?" is more compelling than "you should really be using this." If you’re brave enough, offer to demonstrate your use-case at a team meeting!
Solve a problem they already have. Ask a colleague what part of their work is repetitive, annoying, or stuck. Then offer to help with that specific thing using AI. Nobody wants to feel like work they enjoy is being taken away from them. But most people are open to doing less of the work they dislike or getting help with something they’re stuck on.
Address their concerns. Some people are skeptical because they’ve read about AI's labor impacts, they care deeply about their craft, or they're worried about losing skills. Their concerns are not silly, and dismissing them will only make adoption harder.
4. Can I move my projects from one AI platform to another?
Mostly yes, with some caveats. ChatGPT and Claude both have a “projects” feature that is almost identical. In Gemini, they are called “gems” and are also very similar. A "project" inside Claude, ChatGPT, Gemini, or AI Camp is usually some combination of:
Custom instructions (the “system prompt” that runs at the start of every chat)
Uploaded files and documents
Connected apps
A folder of related chats
The first three travel pretty well. You create a new project on the new platform and copy them over. The last one (your chat history) usually does not move cleanly between tools.
If those past chats matter, two options:
Export them if your current tool supports it. Some do, some don't. Worth contacting their support to ask.
Have the AI summarize them for you before you migrate. Ask the old tool to write up the important decisions, context, and outputs from your conversations. Then upload that summary as a resource in the new project. Faster than starting over and lighter lift than a full data export.
5. A new model drops every week and I have no clue what's going on. Is there a guide?
You almost certainly don't need to keep up with every weekly model update. The top AI models are constantly leap-frogging each other; one month Gemini is the best, ChatGPT the next, then Claude, then back to ChatGPT. For most people, it's enough to check in only when your current tool is failing at something you need. For example, there are many models released each month that are “open source/weights” that can run on your computer. However, in practice almost none of these are worth investigating except if your work specifically requires them (for example, if you have data that can never leave your computer). The better question isn't "what's the best model overall?" It's "what's the best model for the work I actually do?"
You can also have AI do the monitoring for you. Once a month, run something like this:
I work at an animal advocacy nonprofit. My main AI use cases are research, writing, strategy, data analysis, meeting notes, and workflow automation.
The AI tool I use the most is [INSERT TOOL: e.g. Gemini, Claude, ChatGPT, Perplexity, Copilot]. Please tell me whether there were any important updates to that tool in the last 30 days, but also search for other major AI model and tool updates that might be relevant to my work.
Focus only on changes that would matter for a nonprofit team. Summarize:
1. What changed
2. Why it would matter for our work
3. Whether we should change tools or workflows
4. What to test next
Ignore hype and minor benchmark changes unless they affect real work.
Anthropic's model-selection guidance is right that you should test models on your actual prompts and documents before switching.
We're also planning to release a dedicated Knowledge Base with practical resources for using AI. Stay tuned for updates.
And if you're looking to strengthen your AI skills – registrations are open for free, expert-led courses from Amplify, VH's AI training program for animal advocates!
🔸 Sign up now 🔸
6. If I schedule a recurring task in Gemini (or any other tool), will it remember the context I gave it?
When you create a recurring task, you explicitly give it context. It will use that context in all runs. However, in most tools it does not remember previous runs of that task. So if you have a Gemini scheduled task that looks for the latest alt protein news every morning, it may surface the same news story a few days in a row because it won’t remember that it surfaced it yesterday. In some tools there are ways around this, which generally follow the pattern of having each recurring task end with having the model write up what it did to a memory document. Then you tell it to start the next run by reading that document.
7. I'm not comfortable using connectors because of sensitive data. How should I think about this?
This is a reasonable concern, and we don't think the answer is "just trust the vendor." A few things worth knowing:
This isn't unique to AI. When you use Google Drive, Gmail, Slack, Asana, Salesforce, or Airtable, you're already trusting vendors with data, security practices, employees, subcontractors, and future policy decisions. AI doesn't create a new category of risk.
Account type matters a lot. Free personal accounts on most major AI tools may use your conversations and data for training newer models. Work accounts, and nonprofit plans usually don't, and on paid personal plans you can turn it off. However, this can change, so you should always read the actual policy. Google's Gemini privacy hub is a good example of what to look for.
Watch out for the "lethal trifecta." There is a particular risk that comes from combining three permissions in one AI agent:
Access to your private data
Exposure to potentially malicious content (like web pages or emails from outside)
The ability to take actions or send communications.
Any two of those is fine. All three together is where things go wrong. For example: if your ChatGPT agent can:
Read your emails
Hackers can email you (so ChatGPT can read a hackers email)
ChatGPT can send emails
You are at risk because a hacker can now convince ChatGPT to leak your emails to them.
Use connectors deliberately. Start with low-sensitivity connectors first, like a project management tool or a calendar, before connecting your full inbox or document system. Use read-only access where you can. Review your connector permissions every few months. Always keep a human in the loop before anything goes external.
You can scope what the AI sees. Remember that you can directly tell your AI which things to look at which things not to touch. For example, you may have the Asana connector enabled, but you can tell you agent “only look at the Vegan Challenge Project, do not ever look at the Finance Project”. It will not follow your instructions with 100% reliability, but this can limit unnecessary risk.
If you want a low-risk starting point, try a project management tool, a shared docs folder of public-facing materials, or a calendar.
To wrap it up
Acknowledge the costs. AI's environmental footprint, labor impacts, and ownership concentration are legitimate concerns. There's no point pretending otherwise.
Use it deliberately. Pick tools that align with your values and use them for tasks where they actually save time or open up new capabilities.
Keep humans in the process. Review outputs, edit heavily, and don't ship anything important without a human reading it first.
Don't replace human connection where it matters. Just as human presence matters in street outreach, make sure your content doesn't lose its human quality, especially content meant to engage other people. We're a social movement, after all.
Explore AI at all levels of your organization. AI policies, training for all staff, and leadership buy-in all contribute to ensuring everyone benefits from AI.
Stay critical – read the skeptics. Check in with yourself regularly on whether your AI habits need updating. The tools will keep changing, and so should the ways we use them.
Want to sharpen your AI skills?
Amplify has two advanced courses coming up between July 20 and September 7, plus a redesigned version of the core course starting in October.
Registrations are open – save your spot:
