Chris Nackers

ServiceNow

Point Your AI at the ServiceNow Docs Repo

2026-11-06

ServiceNow’s product documentation lives on a website that most AI tools can’t read. That isn’t me guessing. ServiceNow says so in its own repo: the docs site is a JavaScript application that “returns no readable content to LLMs.”

Since Q2 of 2026 there is a fix. ServiceNow publishes the same documentation as a public GitHub repo, in plain Markdown, with an index file written for AI. If you use Claude, Copilot, or anything else to help with ServiceNow work, this is worth an hour of your time, and you can’t really go wrong, just start!

What’s in the repo

The repo is called ServiceNow/ServiceNowDocs. Here’s what ServiceNow says about it, and what I checked myself:

  • One Markdown file per topic. Organized by product area, so an AI can open exactly the page it needs.
  • A branch per release family. Each release family gets its own branch, and brazil is the latest. Per ServiceNow’s own notes, the repo keeps the three most recent released versions, or four while a new release is in early access. The oldest branch is deleted when a new release goes to general availability.
  • An llms.txt index. A table of contents an AI reads first, with a link to every publication.
  • No images. The repo is for AI, not for human eyes. Screenshots stay on the official docs site.
  • Frequent refreshes. The README says “normally at least monthly, sometimes more often”, and its change log backs that up with several entries in Q3 and Q4 of 2026.

Point your AI at it

You have two options. The lazy one is to give your AI a URL. The README’s own example is a prompt like “Read the README at its raw GitHub address and explain AI Control Tower.” The AI follows the links from there.

The better one, if you do a lot of ServiceNow work, is to clone the repo so your AI reads local files. Clone only the branch for your release family. On Windows, run git config core.longpaths true first, or you will hit “filename too long” errors.

Two gotchas I ran into:

  • Don’t send your AI to the docs website. Point it at the repo. The website is the thing it can’t read.
  • Check the branch in every URL. When I looked in Q4 of 2026, the brazil branch’s llms.txt still linked to the australia branch. Same file names, older release. If you want answers for the latest release, make sure the links match the family you asked about.

The part nobody mentions: your copy goes stale

Here’s why I bother writing this up. A cloned repo doesn’t update itself.

In Q3 of 2026 I found my own local copy sitting seven weeks behind, three upstream refreshes out of date. It was hiding something big: the platform-wide rename of Now Assist to ServiceNow Otto. My AI had no idea.

You can see the rename in the files themselves. I counted on the brazil branch in Q4 of 2026. 4,742 files mention “ServiceNow Otto” and 2,948 mention “Now Assist”. 1,893 mention both, and the docs even say things like “ServiceNow Otto (formerly Now Assist)”. An AI working from a stale copy sees only the old name.

How I keep mine current

I added a weekly check that compares my copy to the public repo and reports three things, most important first:

  1. A new release family branch. This is the loudest signal. It means the whole baseline moved, and a file-by-file comparison would miss it.
  2. Vocabulary drift. I keep a list of about thirty terms, count the files that contain each one in the old and new copy, and look for big swings. A term jumping from nearly zero to thousands is a rename or a launch. A term collapsing is a retirement. It’s cheap, and it’s what catches a rebrand.
  3. Files added and deleted per publication. Adds and deletes are real changes. “Modified” counts are mostly noise, because ServiceNow regenerates anchor IDs and bumps the last-updated date on every republish, so tens of thousands of files “change” with no new content.

One habit worth stealing: tag the old copy in Git before you update it. Without the tag, the old files become unreachable and you can never answer “what changed?” again.

A word on the license

The repo carries an Apache License 2.0 notice. GitHub’s own license detector doesn’t recognize it, probably because the file is a short notice and not the full license text. That’s my guess, not something GitHub says. Reading it for your own use is fine. If you plan to redistribute any of it, read the license yourself first.

What to do in the next hour

  1. Open the repo and find the branch for your release family.
  2. Give your AI the llms.txt address and ask it a ServiceNow question you already know the answer to.
  3. If you like what you see, clone that branch.
  4. Put a reminder on your calendar to pull it. Monthly is the minimum, because that’s how often the docs move.

ServiceNow says working from the source docs means less hallucination. I haven’t tested that claim myself, so try it on a topic you know well and see.

Hope this helps someone!