How We Build Our Tools – The Herpetology Collective

The short version

What answers you

Rules we wrote ahead of time, applied to information gathered from named sources. Nothing is invented while you wait.

Where AI helped

In the workshop: writing code, organizing information, and drafting some drawings. We say so openly.

What we promise

If we ever add a feature where AI writes what you read, it will be clearly labeled, and it will not carry the “No AI-generated answers” note.

What “no AI-generated answers” means

Many websites now pass your question to an AI chatbot (a program that writes a likely-sounding reply, one word at a time). Those replies can be useful, and they can also be confidently wrong, because the program is predicting text rather than looking something up.

Our tools do not work that way. Each one is ordinary software: a fixed set of steps that looks things up in information we prepared in advance. When the Shed Identifier asks whether the scales under the tail are in one row or two, your answer removes the snakes that do not match. When What Lives Near Me takes your ZIP code, it finds your county and lists the species on record there. No AI model is consulted along the way.

Programmers call this kind of software deterministic (the same question, asked of the same information, gives the same result). You do not need the word. The point is that a result here can be traced back to a rule and a source, and it can be checked and corrected.

Where the information comes from

It depends on the tool, and each tool page names its own sources. In general:

Published references, such as field guides and the standard names lists used by herpetologists (people who study reptiles and amphibians).

Scientific records, such as museum specimens and observations shared through GBIF and iNaturalist.

Our own courses and textbook, written by Micha Petty.

Public information from agencies, such as park details and weather forecasts.

Micha decides what each tool should do, which sources it draws on, and how its rules work.

Where AI did help

We are a very small team, and we use AI assistants to build things we could not otherwise afford to build. AI has helped with:

Writing the software. Most of the code was written by an AI assistant, working to Micha’s design and decisions, and then tested.

Organizing information. Sorting and cross-checking large sets of records and notes so they could be loaded into a tool.

Some drawings. The line drawings in a few tools and games, such as Scales & Scutes and Decode the Description, were drafted with AI. We are working on improving them. Photographs are by real photographers and are credited where they appear.

That work happened before the tool reached you. Once a tool is published, what it gives you comes from its rules and its sources.

A few honest details

Results can change over time. Some tools read fresh information, such as tonight’s forecast in Crossing Watch or new sightings on a map. The rules stay the same; the information is newer. We also correct mistakes as we find them.

Games shuffle. A game picks its questions at random from a set we prepared, so no two rounds are alike. Nothing is being written as you play.

Outside sources have their own methods. Records from places like iNaturalist and GBIF are gathered and checked by those organizations in their own ways, which can include automated help. We name those sources so you can look for yourself.

Reading names aloud. The Name Decoder uses the voice already built into your phone or computer.

No AI is not the same as no mistakes. A rule can be wrong, and a record can be out of date. Treat every result as a well-sourced starting point, and never rely on an online tool alone to decide whether an animal is safe to handle. If something looks wrong, please tell us.