Optimize at Edge

This page provides a detailed overview on how to deliver optimizations at the CDN edge without any authoring changes. It covers the onboarding process, the available optimization opportunities and how to auto-optimize at edge.

What is Optimize at Edge?

Optimize at Edge is an edge-based deployment capability in LLM Optimizer that serves AI friendly changes to LLM user agents. In the current context, “Edge” means that the optimization is applied at the CDN layer. Because it delivers optimizations at the CDN layer, no authoring changes in the Content Management System (CMS) are required so your origin CMS remains unchanged. This separation lets you improve LLM visibility without altering your existing publishing workflows. It targets only agentic traffic and does not impact either human users or SEO bots. When LLM Optimizer detects opportunities to optimize a page, users can deploy fixes directly at the CDN edge.

Optimize at Edge is a faster, leaner alternative to traditional fixes that demand complex engineering efforts. As mentioned, once you complete a one-time setup, no platform changes or long development cycles are required to apply the changes. You can publish improvements in minutes, without requiring developer engagement. It is a no-code way to optimize your website for AI agents.

Optimize at Edge is designed for business users in marketing, SEO, content and digital strategy teams. It can enable business users to complete the full journey in LLM Optimizer: identifying opportunities, understanding suggestions, and easily deploying the fixes. With Optimize at Edge, users can preview the changes, deploy them quickly at the CDN edge and validate that the optimizations are live. Performance can be tracked in the LLM Optimizer ecosystem.

Key benefits

Which opportunities are supported with Optimize at Edge?

Opportunities that can improve the agentic web experience are supported with Optimize at Edge. Learn more about each opportunity both in the Opportunities Dashboard page and the opportunities section in the current page.

Onboarding

You should reach out to either your Adobe account team or FDE team to start the onboarding process. Your IT or CDN team is also required to complete the pre-requisites and setup process. Additionally, you can also contact our team at llmo-at-edge@adobe.com for further onboarding assistance.

Pre-requisites to onboard to Optimize at Edge:

Requirements for your IT/CDN team:

To guide the setup process, presented below, are sample configuration for a number of CDN setups. These examples should be adapted to your actual live configuration. We recommend applying changes in the lower environments first.

curl -svo page.html https://frescopa.coffee/about-us --header "user-agent: chatgpt-user"
< HTTP/2 200
< x-tokowaka-request-id: 50fce12d-0519-4fc6-af78-d928785c1b85

kind: "CDN"
version: "1"
data:
  # Origin selectors to route to Tokowaka backend
  originSelectors:
    rules:
      - name: route-to-tokowaka-backend
        when:
          allOf:
            - reqHeader: x-tokowaka-request
              exists: false # avoid loops when requests comes from Tokowaka
            - reqHeader: user-agent
              matches: "(?i)(Tokowaka-AI|ChatGPT-User|GPTBot|OAI-SearchBot|PerplexityBot|Perplexity-User)" # routed user agents
            - reqProperty: domain
              equals: "example.com" # routed domain
            - reqProperty: originalPath
              matches: '(/[^./]+|\.html|/)$' # routed extensions, with .html extension or without extension
            - anyOf:
              - { reqProperty: originalPath, in: [ "/page.html" ] } # routed pages, exact path matching
              - { reqProperty: originalPath, like: "/dir/*" } # routed pages, wildcard path matching
        action:
          type: selectOrigin
          originName: tokowaka-backend
    origins:
      - name: tokowaka-backend
        domain: "edge.tokowaka.now"

curl -svo page.html https://www.example.com/page.html --header "user-agent: chatgpt-user"
< HTTP/2 200
< x-tokowaka-request-id: 50fce12d-0519-4fc6-af78-d928785c1b85

unset req.http.x-tokowaka-url;
unset req.http.x-tokowaka-config;
unset req.http.x-tokowaka-api-key;

if (!req.http.x-tokowaka-request
    && req.http.user-agent ~ "(?i)(Tokowaka-AI|ChatGPT-User|GPTBot|OAI-SearchBot|PerplexityBot|Perplexity-User)") {
  set req.http.x-fowarded-host = req.http.host; # required for identifying the original host
  set req.http.x-tokowaka-url = req.url; # required for identifying the original url
  set req.http.x-tokowaka-config = "LLMCLIENT=true"; # required for cache key
  set req.http.x-tokowaka-api-key = "<YOUR API KEY>"; # required for identifying the client
  set req.backend = F_Tokowaka;
}

if (req.http.x-tokowaka-config) {
  set req.hash += "tokowaka";
  set req.hash += req.http.x-tokowaka-config;
}

if (req.http.x-tokowaka-config && resp.status >= 400) {
  set req.http.x-tokowaka-request = "failover";
  set req.backend = F_Default_Origin;
  restart;
}

if (!req.http.x-tokowaka-config && req.http.x-tokowaka-request == "failover") {
  set resp.http.x-tokowaka-fo = "1";
}

Opportunities

Presented in the following table are opportunities that can improve the agentic web experience and are supported with Optimize at Edge.

Additional Tools

The Adobe LLM Optimizer: Is your webpage citable? Chrome extension lets you see exactly how much of your webpage content LLMs can access and what stays hidden. Designed as a free, standalone diagnostic tool, it requires no product license or setup.

With a single-click, you can evaluate any site’s machine readability. You can view a side-by-side comparison of what AI agents see versus what human users see, and estimate how much content could be recovered by using LLM Optimizer. See the Can AI read your website? page for more information.

Opportunities detailed

In the sections that follow, you can view additional details for each opportunity that is supported with Optimize at Edge.

Recover Content Visibility

This opportunity flags pages where key content is hidden for AI agents due to client-side rendering. For each identified page, it shows you exactly which content is missing from the AI agent view, highlights visibility gaps, and enables you to directly apply changes to recover the hidden content. When you deploy this opportunity with Optimize at Edge, a pre-rendered, AI-optimized version of the page is served to LLM user agents so they can access the full context without executing Javascript.
This ensures the page is first fully visible to AI agents. Additional enhancements are applied on top of that pre-rendered HTML.

Optimize Headings for LLMs

This opportunity detects pages where the heading structure makes it hard for AI agents to understand the page due to empty, duplicate, missing or ambiguous headings. For each affected page, the opportunity surfaces the suboptimal headings and recommends a clearer hierarchy. When deployed with Optimize at Edge, the improved headings are applied in the HTML served to agentic traffic. This helps machine readability while leaving your human facing layout the same.

Add LLM-Friendly Summaries

This opportunity identifies pages that can benefit from concise summaries to help LLMs quickly understand what the page content is about. For each page, the opportunity detects where a summary is most needed and creates AI-generated summaries either at the page level or section level. When you deploy with Optimize at Edge these summaries are inserted into the HTML that AI agents retrieve, improving the chances of having your content be described more accurately.

Add Relevant FAQs

This opportunity flags pages where additional Q&A content could better match the user intent and prompts in AI-driven discovery. For each page, it proposes AI-generated FAQ blocks tied to user intent and content on the page. With Optimize at Edge, these FAQs are injected into the HTML, making your page more AI-friendly and increasing the likelihood that AI answers directly reflect your guidance.

Simplify Complex Content

This opportunity finds pages with long, complex paragraphs that can reduce AI comprehension. For each page that exceeds the readability thresholds, it creates AI-generated content that is simpler and easier to scan while preserving the original meaning. When deployed at the edge, the simplified content delivered to agentic traffic helps LLMs interpret and summarize your content more faithfully.

Auto-Optimize at Edge

For each opportunity, you can preview, edit, deploy, view live, and roll back the optimizations at the edge.

Preview

Preview lets you see the impact of a suggestion before it goes live. It surfaces a side-by-side difference between the current page and the AI-optimized version expected after applying the suggestion. This view uses the same Optimize at Edge logic that will power live traffic, but in an isolated preview mode. This does not impact live traffic as it is a read-only simulation for review.

Edit

Edit allows you to refine or rewrite altogether the auto-generated suggestion before deploying it. Instead of accepting the suggestion, you maintain full control through the edit workflow. The view displays proposed changes in a structured editor, where you can modify the text to better match your original intent. The edited version will then be served to AI agents once deployed.

Deploy

Deploy publishes the selected suggestions so the optimized experiences can be served from the edge to AI agents. If the CDN is fully routed, all pages in the domain usually go live with the new changes within minutes. If routing has been configured for select paths only, only the allowlisted pages go live with the optimizations.

View Live

View Live lets you verify that the optimization is live and behaving as expected for agentic traffic, a view that would otherwise be hard to access. You can view the live page under Fixed Suggestions, which renders the page as shown to AI agents.

Rollback

Rollback safely reverts a previously deployed optimization. The AI-only version of the page is typically returned to its previous state within minutes, making it safe to experiment with optimizations when needed.

Frequently Asked Questions

Q. What kind of LLMs do you target with Optimize at Edge?

The list of user agents to target is defined by you during the onboarding process.

Q. What happens if I’m not onboarded to Optimize at Edge yet?

If you click Deploy optimizations before completing the required setup, nothing will be applied to your site. Instead, a pop-up dialog prompts you to contact our team at llmo-at-edge@adobe.com for onboarding assistance. Until onboarding is complete, you can still explore the detected opportunities and suggestions, but the one-click deployment workflow will remain inactive.

Q: What happens when the content is updated at source?

We serve the optimized version of the page from cache as long as the underlying source page hasn’t changed. However, when the source does change, our system automatically refreshes so AI agents always receive the most up-to-date content. This is because we use low cache time to live (TTL) settings (by order of minutes) so that any content update on your site triggers a new optimization within that window.

Q. Is Optimize at Edge only for sites using Adobe Edge Delivery Service (EDS)?

No. Optimize at Edge is CDN-agnostic and works with any front-end architecture, not just the ones deployed on Adobe’s EDS Stack.

Q. How is Optimize at Edge pre-rendering different from traditional server-side rendering (SSR)?

Both solve different problems and can work together. Traditional SSR renders server-side content but doesn’t include content loaded later in the browser. Optimize at Edge pre-rendering captures the page after JavaScript and client-side data has loaded, producing the fully assembled version at the CDN edge. SSR focuses on improving the human experience and Optimize at Edge improves the web experience for LLMs.