How AI Works
Artificial intelligence is reshaping how organizations communicate, research, and campaign—and animal advocacy is no exception. From donor appeals to creating social media content, understanding how AI actually works will help you use it more effectively.
What Is AI?
AI is an umbrella term for machines that mimic human-like intelligence. When you use tools like ChatGPT or Claude, you're interacting with a so-called Large Language Model (LLM) trained to predict what text should come next, based on massive amounts of web pages, books, articles, and other text sources.
Through training, the AI repeatedly tries to predict what word comes next in a sentence, getting feedback on whether it was right or wrong, thus identifying patterns in language and generating relevant responses.
This has an important implication: AI reflects its training data. Since speciesist language is the norm on the internet, most AI models will default to speciesist framings unless you guide it otherwise. When you chat with AI in your Open Paws tools, you are chatting with a model that was fine tuned to be non-speciesist.
How AI Processes Your Prompt
When you type a prompt, AI analyzes the context of every word and how they relate to each other. Then it generates a response informed by its training, any custom instructions, your prompt, and everything it has generated so far. This is why AI is non-deterministic: The same prompt can produce different results. Here is a video that explains this project.
Image and video generation works differently from text, but the core principle remains: pattern recognition at massive scale.
Image generation tools were trained on billions of images paired with text descriptions. They learned the visual patterns associated with words. Video generation tools extend this process across time. They've learned not just what things look like, but how they move—how grass sways, how animals walk, how light shifts. Your prompt guides both the visual content and the motion patterns.
Think of every element in an image as a dice roll. Ask for "a pig" and the AI rolls the dice on everything else: Species, running or lying down, background, lighting, style, angle? All random. The more you specify in your prompt, the more dice you lock in place. "A pig" produces something generic. "A pig lying in front of a sanctuary barn, golden hour lighting, photorealistic" will give you a more predictable result.
The limitations are similar to text: these tools reflect their training data. Stock imagery aesthetics dominate. Genuine depictions of animal suffering or industrial farming may be restricted or poorly rendered because such images were filtered from training data or are underrepresented. You may need to iterate significantly or combine AI-generated elements with authentic footage.
Limitations You Need to Know
AI is powerful, but has two main limitations to be aware of:
Hallucinations: AI can confidently state false information—inventing statistics, citing non-existent studies, or misattributing quotes. Always verify facts, especially for public-facing content.
Bias: Training data contains societal biases, including speciesist assumptions. AI may default to language that centers human interests or minimizes animal experiences unless explicitly guided.
Getting Better Results
Your outputs are only as good as your inputs. A few principles:
Be specific. "Write a fundraising email" produces generic content. "Write a 200-word fundraising email for monthly donors about our undercover investigation into [specific issue], emphasizing [specific impact]" produces usable drafts.
Provide context. Share your organization's voice guidelines, past examples of successful content, or specific facts you want included. Tip: Ask AI to ask you questions. This will help you give additional context for your prompt.
Request non-speciesist language. AI will often default to phrases like "animal welfare" or "humane treatment." If your organization uses rights-based framing, say so explicitly. In AI chat tools like chatGPT or Claude, you can add custom instructions that will apply to all of your chats.
Iterate. Treat first outputs as rough drafts. Ask for revisions, alternative approaches, or specific improvements.
Verify everything. Never publish AI-generated statistics, citations, or factual claims without independent confirmation.
AI won't replace the passion, judgment, and moral clarity that drive effective animal advocacy. But it can handle repetitive or time-consuming tasks, freeing you to focus on strategy, relationship-building, and the work only humans can do. Advocates who learn to use these tools thoughtfully will move faster and reach further—which means more impact for the animals!