Look Through The Faunalytics Lens
We are now several years into the boom of generative AI, and in many ways animal advocates are still trying to figure out how to use these tools without inheriting their problems. We’ve written about this twice already on the blog: once about the reliability problems with using AI as a research tool, and once about the broader state of AI in the movement and the questions advocates should be asking themselves before going “all in.” Both pieces circled the same underlying question without answering it directly: what would an AI tool built specifically for animal advocacy research actually look like — one that solves for the problems we’ve been naming rather than inheriting them?
This blog introduces our earnest and thoughtful attempt at an answer: the Faunalytics Lens, now publicly available on our website. On the surface, it’s a tool that navigates the Faunalytics Library — over 6,000 human-vetted and summarized research entries, as well as all of our original studies and our broad range of other resources. With access to that corpus, it answers questions with direct citations to the source material. It’s a tool purpose-built for one job: making our vast collection of materials and data faster to look through and easier to use, without the reliability problems that come with general-purpose AI tools.
Why is it called the Lens? We picked this metaphor because it sums up what we hope the tool does for advocates:
- A lens clarifies — it makes the research landscape easier to see.
- A lens focuses without narrowing — it brings the relevant research into sharpness without excluding the broader picture.
- A lens requires someone to look through it — advocates are active participants in using it.
- A lens reorients — it shifts perspective, which speaks to the tool’s behavior of reframing any query toward animal advocacy.
What The Lens Does And Doesn’t Do
In plain terms, the Lens is a chat interface. You type an advocacy question and the Lens looks through the Library, then answers with citations linking back to the original entries. You can ask follow-up questions, copy responses, and dig deeper. One exciting feature is that you don’t have to have a fully formed question to take advantage of what it offers. For example, if you just want a summary of a bunch of information about bees, you can type “bees.” If you know you want to research something related to factory farms, environmental impact, and public opinion, you can throw in a few keywords and the Lens will do the rest, asking clarifying and follow-up questions as needed.
The Lens can do this because it takes any prompt — even a prompt that may be hostile to animals, such as “how do I get rid of the rats in my house” — and reframes it in animal advocacy terms before it hits our database. This means it returns a response that always places effective animal advocacy at the forefront.
A powerful value-add of the Lens is that you don’t just have to ask it for research — you can ask it to help you generate ideas for outreach materials, fundraising, campaign messaging, and more. It will ask follow-up questions to help refine your idea, and even give you specific options to try out. Best of all, it will link to the sources in our Library that support the advice it’s giving you.
A further value-add that will be especially useful to our global audience is that the Lens can converse in a variety of languages: ask a question in Spanish, and get a Spanish response! In our testing with languages that the team is fluent in (French, Spanish, Polish), responses were high quality and translated results well. However, this feature should be considered experimental — please let us know about your experiences with it!
Just as important as what the Lens does is what it won’t do.
First and foremost, the Lens is not connected to the open web. If a study isn’t in our Library — where it’s been vetted and edited by a human — the Lens will not reference it. While this may seem like a limitation, it means that the Lens will only include research and findings from the 6,000+ studies that our team has already determined are useful to advocacy, and summarized with our highest standards of accuracy.
If the data to answer your question doesn’t exist in our Library, the Lens will tell you so, and frame it as a research gap rather than making something up. This point matters more than it might sound, and that diverges from the general-purpose tools most advocates are familiar with.
Finally, since the Lens has a finely tuned persona that centers animal advocacy at its core, it will not give you answers that would support exploiting animals. In other words, don’t bother asking it for a chicken soup recipe, and you won’t have to worry about it providing recommendations that support animal exploitation.
Why Source-Grounding Matters
In our first blog on AI and research, we covered three persistent problems with general-purpose LLM summarization: confident hallucinations, confirmation bias, and unsupported extrapolation. Our Research Library Manager Meghann Cant walked through specific cases where she’d caught LLMs fabricating quotes, reversing study findings, and adding “speculation” beyond what the source actually said. The takeaway was that nothing replaces careful human verification — and that the burden of verification, in a general-purpose AI workflow, falls entirely on the individual user. We also noted in that blog that only 8% of people always check the sources behind AI overviews. In other words, most AI users don’t verify the outputs they’re getting, and that’s a significant problem.
The Lens is built differently in three concrete ways, each addressing one of those problems directly:
- It pulls only from the Faunalytics Library. This is called “source-grounding,” and we singled it out in a blog as one of the more promising approaches to using AI for research. Source-grounded models don’t draw on the open internet or on whatever the underlying LLM happens to have absorbed and emphasized during training. They only work with the corpus you give them. In our case, that corpus is over 6,000 entries that Faunalytics volunteers and staff have already read, summarized, and vetted — built from over 25 years of curation. The Lens can’t go off and pull from a random Substack post, a low-quality news article, or an AI-generated summary somewhere else on the web. It works with vetted material, or it tells you about research knowledge gaps.
- Citations are mandatory and clickable. Every claim the Lens makes links back to the Library entry it came from, and because of our database architecture, it does so with an extremely high degree of accuracy. Our hope here is to make verification the default, not the exception. Although we’ve already vetted the source material, you can easily click straight through to the original summary and confirm.
- It’s animal-aligned by design. The Lens is built to be an animal advocacy researcher. It won’t provide recipes for animal-based dishes or suggest hurting animals for pest control or conservation, but will redirect such questions toward animal-friendly alternatives. Most general-purpose chatbots are happy to do all of those things on request, or require special prompting to avoid these pitfalls. The Lens’ system prompt — the underlying instructions that shape how it responds — makes our movement’s ethical commitments part of how the tool behaves rather than something the user has to enforce themselves.
None of this makes the Lens infallible — nothing is, and we approach anyone who makes such claims with a high degree of suspicion. But it puts the user in a much better position to catch errors if they happen because the path back to the source is one click away.
The Loop, Revisited
There’s a deeper argument running underneath the Lens, and it’s worth emphasizing because it might be easy to miss in an announcement post.
In our blog on AI and research, we made the case that animal advocacy research — the often unglamorous work of producing studies, reading other people’s studies, writing summaries, vetting findings, and building libraries and databases — is vital to the continued usefulness of AI in our movement. When you ask a general-purpose LLM about animal advocacy, it might cite Faunalytics, Rethink Priorities, and other research organizations. AI needs human-generated research to remain useful.
In many ways, the Faunalytics Lens operationalizes that argument. The tool works so well because of the upstream labor that’s been happening in our Resource Department for years: Myself, Che Green, Casey Bond, Meghann Cant, and others managing and curating the Library over the last two decades, dozens of volunteer writers around the world summarizing studies, our research team prioritizing what gets included in our major studies. Every entry in the Library was produced, written, and edited by a human before the Lens could surface it. The Lens depends on that invisible labor, much like AI depends on the labor of producing all of the data it scrapes and ingests.
This matters because it shapes what kind of tool the Lens is in the broader ecosystem of AI products. A tool that pulls from the open web makes individual advocates responsible for verifying everything that comes back. A tool that pulls from a curated, vetted Library distributes that verification work across a whole research organization, with the individual advocate as the final check. The former offloads responsibility to the user. The latter centralizes verification where it’s most efficient and surfaces the result.
A research tool is only as good as the source it draws from. The Lens is highly useful because the Library is high quality — and the Library is high quality because we keep investing in it.
This is why we don’t see the Lens as a replacement for the Library, nor for the work of the people who build it. Instead, it’s as if we put a new front door on the same building. The building is still being constructed, brick by brick, by humans who care deeply about quality, accuracy, and data. We’re still actively summarizing new studies, still publishing new research, still adding to a broader pool of knowledge. The Lens just makes it faster for advocates to see what’s already there and synthesize it immediately.
Using Our Slingshot Carefully
In our second blog on AI in the movement, we picked up a decade-old metaphor from Faunalytics’ founder Che Green: data is the slingshot in our David-versus-Goliath fight, and AI tools might become an even better slingshot — but using a slingshot carelessly can backfire. We’ve tried to be mindful and methodical in our adoption of AI as an organization, and the Lens is one expression of that. But of course, mindful adoption means being honest about the tradeoffs, so it’s worth naming a few.
The Lens runs on third-party infrastructure. The underlying language model is Anthropic’s Claude Sonnet, and uses a few other bits of software to help make the whole thing run. We chose these carefully, and at present they’re the best fit for the job. However, we’re under no illusion that we control the underlying technology. Pricing will change. Models can and will be deprecated. Capabilities could shift in directions that don’t suit our needs. This means the Lens will be a living work in progress — something which I’m actually quite excited about continuing to develop.
It can still be wrong. Source-grounding and mandatory citations are real mitigations, but they don’t eliminate errors. While testing has been thorough and the quality has been extremely high, it’s entirely possible for the Lens to pull from a less-relevant or older source, paraphrase a finding imprecisely, or miss the nuance of a study’s methodology. The same instinct to verify that we’ve been encouraging readers to bring to general-purpose AI tools applies here too — but with a much shorter path from claim to source.
It has an environmental footprint. While the impact is modest, it isn’t zero, and we don’t want to pretend otherwise. You can read a fuller description of the Lens’ environmental impact on the tool’s landing page; we look forward to assessing this further if Anthropic releases a proper Life Cycle Assessment of its models.
We share these caveats not to undercut the launch, but for the sake of transparency and consistency: as we’ve said before, using AI doesn’t preclude us from having a critical perspective on it, and having a critical perspective doesn’t preclude us from using it. The Lens is one way where we’re trying to model that balance in practice.
How To Use It Well
A few practical suggestions, drawn from how we’ve been using it ourselves during development:
Bring it specific, research-oriented questions. “What arguments are most effective for reducing meat consumption?” will give you a better result than “Tell me about veganism.” The Lens works best when it has something concrete to anchor its search to. You can ask in full sentences, in keyword strings, or with broad concepts — but specificity always helps get you to the finish line more quickly.
Follow the citations. Every linked citation goes back to a Library entry that was vetted by a human. If a claim matters to your work, click through and read the original summary. The Lens is designed to make that easy.
Use it as a starting point, not an endpoint. For methodological questions, study design help, or deeper analysis of a specific topic, our free weekly Office Hours is still the better path. The Lens is excellent at quickly surfacing what’s in the Library; it’s not a substitute for a conversation with a researcher who can help you think through what to do with that information.
Tell us when something doesn’t work. The Lens is in active development, and feedback from real users is how we’re going to make it better. If you get a strange response, a missing citation, or an answer that doesn’t seem grounded in the Library, we want to know. To report bugs and problems, please contact us directly.
Going Forward
The Lens is one experiment in how AI can serve animal advocacy rather than the other way around. It joins a small but growing set of initiatives we’ve been working on: carefully vetted AI summaries of long reports being added to the Library, the work of our internal AI Committee in setting policy and exploring use cases, ongoing collaborations with partners like Stray Dog Institute, and our continued commitment to publishing research that — among other things — keeps the underlying AI ecosystem honest by feeding it human-generated, verified data. We hope it will hold a high bar for what thoughtful AI integration can look like, and serve as a jumping-off point for Faunalytics to explore this territory further, without compromise.
The broader project here is what it’s been for 25 years: making the data driving animal advocacy a sharp, reliable instrument rather than a “good enough” best guess. AI can sharpen that instrument when it’s used carefully, in service of a clear mission, on a foundation of vetted research. It can also blunt it if we let general-purpose tools and unvetted sources shape what we think we know about our own movement. The Lens is one attempt to do the former: to build a tool that strengthens the research infrastructure of animal advocacy rather than displacing it.
We hope you’ll try it out, use it in your work, and let us know what you find. And we hope you’ll continue to join us in thinking carefully — and critically — about how AI fits into the movement we’re all building together.
The Faunalytics Lens was built in collaboration with Kyle Behrend and with the support of Stray Dog Institute.
