We believe you should understand the tool you are trusting with your pet. This page explains, in plain language, how PawDoc's AI works and where its limits are.
1. How it works#
When you submit a photo, short video, or description along with your pet's profile, PawDoc sends that information to third-party AI models — currently Google Gemini and Anthropic Claude — which analyze it and return structured guidance. PawDoc organizes that into a simple urgency signal — an action from get help now through watch and re-check (its lowest level; it never says your pet is "normal" or "fine") — plus general information and a recommendation about seeing a veterinarian.
To keep results as safe and consistent as possible, PawDoc:
- runs hard-coded safety checks for emergency keywords before any AI call, so obvious emergencies are surfaced regardless of what the AI returns;
- constrains the AI to a fixed, structured response format and rejects off-format output;
- uses a low-randomness setting for health analysis; and
- responds with "insufficient information" rather than guessing when confidence is low.
2. What AI can do well — and what it cannot#
AI is good at quickly surfacing patterns and giving you a sensible starting point. But it has real limitations you must keep in mind:
- AI can be wrong. It can miss a serious problem (a falsely low "watch and re-check") or raise a false alarm.
- AI can "hallucinate." It can produce confident, fluent statements that are simply incorrect.
- It is probabilistic. Results are best estimates, not definitive findings.
- It depends entirely on your input. A blurry photo, a partial video, or an incomplete description can produce a wrong or uncertain result. PawDoc cannot see, touch, smell, or test your pet.
- It does not monitor your pet. PawDoc only responds to what you send, when you send it. It does not watch for changes or follow up.
- It is not a veterinarian. It cannot replace professional judgment, an in-person examination, or diagnostic testing.
3. What this means for you#
Use PawDoc as one input to your decision, not the decision itself. If your own judgment says something is wrong, trust it and contact a veterinarian — even if PawDoc's guidance is reassuringly low-urgency. In anything that looks like an emergency, seek care immediately and do not rely on the app (see the Emergency Disclaimer).
4. Honesty about accuracy#
We do not claim that PawDoc is as accurate as a veterinarian, and we do not publish accuracy percentages we cannot substantiate. PawDoc is guidance and triage, not diagnosis. If we ever make a specific performance claim, it will be backed by evidence and clearly scoped.
5. Reporting a problem#
If PawDoc ever returns guidance that seems offensive, unsafe, or badly wrong, please tell us at support@pawdoc.app so we can investigate and improve. Your reports help make the service safer.
6. Model identifiers#
PawDoc refers to AI models by their technical identifiers (for example, claude-sonnet-4-6, gemini-2.0-flash) rather than marketing names, and the specific models may change over time as we improve the service.
7. Related#
See the Veterinary Disclaimer, the Privacy Policy (how your inputs are processed), and the Terms of Service.