AI is Coming

After more than 25 years of hand-reviewing every trail report (over 130,000 as of December 2025), it is time for change. In the fall of 2024, we began working on an artificial intelligence (AI) model to review trail reports. The initial step was to develop our prompting data (instructions on how to review), which in itself was a difficult process as it forced us to attempt to define on paper a lot of information that has been more a philosophy in the past.

We’ve been testing the report review system internally with several AI models (like Claude, ChatGPT and Gemini) for the past year. During testing last season, the feeling was the AI was a bit too inconsistent in its analysis, sometimes providing excellent results and other terms failing horribly. But over the last six months, newer AI models have rolled out that have improved the results significantly. While we are still tweaking the model, we now feel it is reasonably accurate with most reports.

We are finally ready to move this tool from the back end to the front end, providing real-time feedback on the trail reports. The trail report form will now contain a ‘Review Report’ button.

This will submit the report to the AI model for review (the data itself is anonymized – no user identification is sent with the report data). Within 3-5 seconds, the model should return three pieces of feedback:

• Review: A short review of the report, typically highlighting the positive points and possible trouble issues. The review should help to guide the reporter on areas that could be improved, and what might cause issues with the report being posted.
• Grade: A grade on the report, indicating how well it achieves the goal of providing detailed, clear information on trail conditions. Grades of D or F could indicate the report may be held up for more information or clarification.
• Flag: A flag will indicate there is an issue the system feels needs to be addressed. A flagged report will require a human to review it further.

The goal is to eventually develop a model that we can trust such that most reports could be approved real-time without the need for human review.

For now, the review process on the user end is completely optional. At any point during the reporting process, the user can push the ‘Submit Report’ and bypass the review interactions. Humans will still be looking over the reports before going online, either way.

Jan 25 update:  All reports will go through an initial review when submitted.   If issues are detected that will require further review, the submitter will be presented with some feedback and an opportunity to address the issues.  Pressing the Submit button will bypass any further feedback.

This is an evolving process. We’d love for most reporters to test drive the review process to “pre-check” their reports and improve the reporting system. And we’d like to hear about any issues and concerns that might arise. Send feedback to: