Edge routing exclusions
Every path the NORG router never handles, on every platform. The lists are the ones NORG's installers actually bind. For why edge routing works the way it does, see the edge routing white paper.
How exclusions work on each platform
An excluded path goes straight to your origin and the router never handles it. Your site is unchanged either way: excluding a path only stops NORG mirroring, classification and visit telemetry for it, and on most platforms it also stops it counting against your plan.
| Platform | How an exclusion works | Where it is set | Saves invocations |
|---|---|---|---|
| Cloudflare Worker | A Workers Route with the Worker set to None. Cloudflare picks the most specific route, so the worker is never invoked for that path. | Set at install from the detected platform. Add your own in Cloudflare (zone, Workers Routes) or ask NORG to set them. | Yes |
| AWS CloudFront | A cache behaviour with no function association. Matching requests never reach NORG code. | Created by the install ahead of the default behaviour. | Yes |
| Vercel middleware | The matcher in your middleware config. Paths it does not match never run the middleware. | Your own middleware.ts. | Yes |
| Fastly Compute | No per-path setting. Every request runs the router, which returns assets and non-agent traffic to your origin before any call to NORG. | Built in. | No |
| Bunny Edge Scripting | No per-path setting. The script runs on cache misses only and passes assets and non-agent traffic straight through. | Built in. | No |
| WordPress plugin | Runs inside WordPress, so static files served by your web server never reach it. It also hands WordPress internals straight back. | Built in. | n/a |
Platform defaults (Cloudflare)
When NORG installs the Cloudflare worker it detects the platform behind your site and binds these paths as no-worker routes.
WordPress (14 patterns)- /wp-content/*
- /wp-includes/*
- /wp-cron.php
- /cdn-cgi/*
- /feed
- /feed/*
- /wp-admin/*
- /wp-login.php
- /xmlrpc.php
- /readme.html
- /license.txt
- /wp-comments-post.php
- /wp-signup.php
- /wp-activate.php
Detected by: WordPress markers on the origin (/wp-content/, /wp-includes/) at install time.
Measured across live WordPress zones, this set carries about 28% of all requests: theme and upload assets (16%), core JS/CSS (4%), wp-cron (3%), feeds, admin and the login, XML-RPC and comment handlers. None of it is agent-facing content. /wp-json/* is left out on purpose because it carries live application traffic on many sites; add it per site if yours does not.
Shopify (Cloudflare O2O) (14 patterns)- /cdn/*
- /.well-known/shopify/*
- /recommendations/*
- /checkouts/*
- /services/*
- /web-pixels*
- /api/*
- /cdn-cgi/*
- /cart.js
- /shopify_pay/*
- /sf_private_access_tokens*
- /private_access_tokens*
- /search/suggest*
- /cart/*
Detected by: A proxied CNAME to shops.myshopify.com on your zone.
Measured on live storefronts, the first ten carry 97-98% of requests (/cdn/* alone about 73%); the last four are background browser calls (access tokens, predictive search, cart AJAX). The bare /cart and /search pages are deliberately not excluded: they share paths with real pages.
Next.js on Vercel (2 patterns)- /_next/*
- /_vercel/*
Detected by: Next.js or Vercel response headers on the origin.
Everything Next.js emits (JS and CSS chunks, image optimisation, fonts) lives under /_next/*, and Vercel analytics beacons under /_vercel/*. Your own public/ folders (/images/*, /fonts/*) vary per site, so add them yourself.
Any other platformSquarespace, Webflow, Wix, a custom build or anything NORG does not recognise gets no default exclusions. Nothing on your site changes; the router simply runs for every path and passes assets and non-agent traffic straight through. Add exclusions for your own asset folders to keep invocations down.
Adding them yourself on CloudflareIn Cloudflare open your zone, then Workers Routes, then Add route. Enter your hostname followed by the pattern, for example www.example.com/wp-content/*, and set Worker to None. Keep your main www.example.com/* route on the NORG worker.
Patterns are path prefixes. Cloudflare does not allow a wildcard in the middle of a path or an extension glob, so /*.js and /blog/*/feed cannot be expressed. Assets that live outside a prefix are still passed straight through by the router itself (see static assets), they just count as an invocation.
CloudFront, Vercel, WordPress, Fastly and Bunny
What each of the other platforms excludes, and how.
AWS CloudFrontThe install adds these as cache behaviours with no function and caching disabled, for every method, so transactional and authenticated paths behave exactly as they did before:
- /api/*
- /wp-json/*
- /wp-admin/*
- /wp-login.php
- /cart
- /cart/*
- /checkout
- /checkout/*
Full detail, including behaviour order and the limit on behaviours per distribution, is in the CloudFront architecture.
Vercel middleware (Next.js)Exclusions are your middleware matcher. This one skips /_next and static files but still matches .txt, .md, .json and .xml, because NORG answers /llms.txt, /agents.md and /.well-known/mcp.json on your domain:
export const config = { matcher: [ '/((?!_next|[^?]*\\.(?:html?|css|js(?!on)|jpe?g|webp|png|gif|svg|ttf|woff2?|ico|csv|docx?|xlsx?|zip|webmanifest)).*)', ], }Variants for other routers are in the Vercel middleware guide.
WordPress pluginThe plugin runs inside WordPress, so files your web server serves directly never reach it. It also hands these straight back to WordPress, along with admin-ajax.php and admin-post.php at any path, ?rest_route= requests, and admin, AJAX, cron, WP-CLI and XML-RPC requests:
- /wp-content/*
- /wp-includes/*
- /wp-cron.php
- /cdn-cgi/*
- /feed
- /feed/*
- /wp-admin/*
- /wp-login.php
- /xmlrpc.php
- /readme.html
- /license.txt
- /wp-comments-post.php
- /wp-signup.php
- /wp-activate.php
- /wp-json
- /wp-json/*
- /wp-admin
A page cache in front of WordPress is a different exclusion (keeping AI agents out of the cache, not paths out of the plugin). See the plugin guide and WP Engine.
Fastly and BunnyNeither has a per-path setting. The router runs, and a browser, a search crawler or a static asset is answered from your origin before NORG is contacted. Bunny only runs it on a cache miss.
On every platform
Three rules hold whichever CDN you use.
Static assets (49 suffixes)On every platform a path ending in one of these is passed straight to your origin, whether or not an exclusion covers it. .xml, .txt and .json are deliberately absent: NORG serves its own files (llms.txt, sitemaps, the MCP manifest) with them.
- .css
- .js
- .mjs
- .map
- .png
- .jpg
- .jpeg
- .gif
- .webp
- .svg
- .ico
- .avif
- .bmp
- .tiff
- .heic
- .woff
- .woff2
- .ttf
- .eot
- .otf
- .mp4
- .webm
- .mov
- .avi
- .mkv
- .m4v
- .ogv
- .mp3
- .wav
- .ogg
- .m4a
- .aac
- .flac
- .weba
- .vtt
- .srt
- .doc
- .docx
- .xls
- .xlsx
- .ppt
- .pptx
- .zip
- .gz
- .rar
- .tar
- .7z
- .webmanifest
These are NORG's own surfaces on your domain. An exclusion covering one would silently switch off MCP or AI discovery, so NORG refuses it, along with the agentic prefix where generated pages live (/ai/* by default).
- /mcp
- /mcp.json
- /.well-known/mcp.json
- /sse
- /.well-known/ai-plugin.json
- /openapi.json
- /openapi.yaml
- /llms.txt
- /llms-full.txt
- /agents.md
- /.well-known/ai-catalog.json
- /tree.json
- /graph.jsonld
- /.norg/*
No default covers a customer portal, member area, booking history or internal dashboard, because the obvious patterns (/account*, /login*) also match real pages like /accounting or /login-help. Name yours explicitly.
It is not a disclosure risk if you do not: the router only ever handles GET requests (plus the MCP POST, which goes to NORG, never to your origin), and NORG only sees what an anonymous public crawler sees. Excluding it keeps signed-in pages on the shortest path and off your invocation bill.
https://norg.ai© 2026 NORG.ai · Help & guides