Quickstart

Five steps, from an account to a card on one section of a live site, on a Cloudflare Worker. Steps 1 to 3 happen in the console and are the same on every edge; on Vercel or ESI, do them, then follow your page in Integrate. You need:

a doubleshift account where you are an owner: only an owner creates API keys;
a site served through Cloudflare, and a Wrangler project for the Worker (wrangler.jsonc or wrangler.toml) that you can deploy on that zone;
one public HTML article in the section you start with (/articles/ below), not behind a login or a consent wall.

In the console, open Sites & fragments and add your bare domain, host only: https://your-domain.example/news becomes your-domain.example. See Sites and fragments.

Route a section to a fragment

Open the site and create a fragment: a name and a language. It starts active. Under Routing rules, enter your section’s path, /articles/, choose prefix, select that fragment, then Add rule. Pages under /articles/ now resolve to it; no default fragment is needed. See How delivery works.For step 5 to show a card, at least one active card must be eligible on that fragment. A card with no fragment selected in its eligibility is eligible everywhere. See Cards.

Create a delivery key

Open Settings → API keys and create a key. It is shown once: copy it now. Note your account id (acc_…), shown in the console: it is the partnerId below. The key stays server-side, in a Worker secret; see API keys.

Deploy the Worker with the edge library

Get the library, version 1.1.4 or later, and check its SHA-256:

curl -sO https://api.shftd2.com/edge/doubleshift-edge.v1.js
shasum -a 256 doubleshift-edge.v1.js
# v1.1.4 — expected: d4db890cbf27a2da1977d3ba76e0585ab3bf7cc78111eaddc71019527114e9d1

Save doubleshift-edge.v1.js next to your Worker entry file and commit it, replacing any older copy: before 1.1.4, scope is ignored and every page is processed. Then store the key as a Worker secret, out of the code:

npx wrangler secret put DOUBLESHIFT_API_KEY

In the console, open the site, then the fragment’s Integration page. The routing rules that target this fragment are selected, and the Worker code there carries their paths in scope. Use it as a new Worker, or add its two lines to a Worker you already run:

worker.js — standalone

import { enrichWithDoubleshift, prepareOriginRequest } from "./doubleshift-edge.v1.js";

export default {
  async fetch(request, env, ctx) {
    const response = await fetch(prepareOriginRequest(request)); // your origin, as today
    return enrichWithDoubleshift(request, response, {
      apiKey: env.DOUBLESHIFT_API_KEY,
      partnerId: "<your account id>",
      scope: { include: ["/articles/*"] },
    }, ctx);
  },
};

Inside an existing Worker

import { enrichWithDoubleshift, prepareOriginRequest } from "./doubleshift-edge.v1.js";
// ... where you fetch and return the page:
const response = await fetch(prepareOriginRequest(request)); // instead of fetch(request)
return enrichWithDoubleshift(request, response, { apiKey: env.DOUBLESHIFT_API_KEY, partnerId: "<your account id>", scope: { include: ["/articles/*"] } }, ctx);

Replace <your account id>, and /articles/* with the paths the console shows; outside them, the Worker leaves the page alone. To cover the whole site, choose Every page of your-domain.example in the console: the Worker then has no scope. Patterns and limits: scope.prepareOriginRequest drops If-None-Match and If-Modified-Since from an AI agent’s GET; every other request is returned untouched. Pass response as fetch() returned it: do not read its body, and do not compress it yourself. Nothing changes in your templates: the card goes right before the first <article>, and a page without one is served intact.Give the Worker a route on the hostname you will test, or it never sees your pages: in its Wrangler configuration, or in the Cloudflare dashboard under the Worker’s Settings → Domains & Routes. Keep the route broad and let scope pick the paths (routes and scope).

wrangler.jsonc

"routes": [{ "pattern": "your-domain.example/*", "zone_name": "your-domain.example" }]

Then deploy:

npx wrangler deploy

Check humans, agents and scope

Request your article as a human, then as an AI agent, then a page outside scope as an AI agent:

# Expect 0: no card for humans.
curl -s https://your-domain.example/articles/some-article | grep -c 'abc-card'

# Expect 1: one card for agents.
curl -s -A "Mozilla/5.0 (compatible; GPTBot/1.0)" \
  https://your-domain.example/articles/some-article | grep -c 'abc-card'

# Expect 0: the Worker leaves this page alone.
curl -s -A "Mozilla/5.0 (compatible; GPTBot/1.0)" \
  https://your-domain.example/about | grep -c 'abc-card'

Human 0, agent 1, outside 0 is a first presence check: the card reaches agents only, on your section only. It does not check where the card sits, the brand line or your cache. Run Verify and troubleshoot before you call the page live; any other result has a known cause and fix there.

Next: Cloudflare Worker covers routes and scope, several sites, delivery logs, updating the library and watching its records.

Start

Configure

Operate

Updates