Cloudflare is adding new tools to its dashboard to address an audience whose presence on the web is growing: AI agents that discover information, compare options, and may act on behalf of the user. The company is combining its previous work on Agent Readiness with a new tool called Answer Engine Optimization (AEO), enabling website owners to determine whether agents can use their websites and whether intelligent assistants recommend them to users.
Cloudflare says that fewer than half of HTML page requests now come from humans, while the remaining share comes from machines, noting that not all of these requests represent agents acting on behalf of people. The company believes that discoverability is no longer limited to appearing on search results pages, but now also includes agents’ ability to find, read, trust, and then recommend content.
Agent Readiness Checks
The Diagnostics section represents the technical check within Agent Readiness. It examines the website in the way an agent might read it, verifies that the agent is allowed to enter and can discover content, then retrieves a machine-readable version and looks for interfaces it can invoke.
The checks include robots.txt files, XML sitemaps, response headers, a Markdown version of the content, and published data for authentication and tools. The results are combined into a single assessment ranging from “Not ready” to “Already agent-ready,” with each check classified as passed, failed, or neutral, accompanied by a note explaining its importance and a record showing the request and response that were inspected.
Cloudflare divides the suggested improvements into levels that include quick fundamentals, such as a readable robots.txt file, an XML sitemap, AI crawler rules, and clean Markdown; the technical foundation, such as Content Signals, an API index, link headers, and agent login instructions; and advanced integrations, such as OAuth discovery, MCP and A2A cards, a skills index, Web Bot Auth, and WebMCP.
The section also displays emerging commerce standards for agent-mediated payments, including x402, ACP, Universal Commerce Protocol, and AP2. Cloudflare explains that this information is currently advisory and does not affect the assessment result. The suggestions also include next steps: settings supported by Cloudflare are linked to an “Set up in Cloudflare” option, while a prompt can be copied that suggests what a coding agent needs to build the other improvements.
Measuring Intelligent Assistant Recommendations
The AEO section focuses on the following question: when a customer asks an intelligent assistant a question within the website’s category, does the assistant recommend that website or a competitor? Cloudflare infers the website’s domain and category, then tests prominent assistants, namely Anthropic’s Claude and OpenAI’s GPT, using prompts that simulate recommendation questions, product comparisons, and general advice.
The tool displays several metrics, including Citation Rate, the percentage of answers that cite the website as a source; Prominence, which measures the website’s presence and position in the answer; Mention Rate, which indicates how many times the brand name is mentioned whether or not the website is included as a source; and Share of Voice, which compares the website’s share of citations with those of its competitors.
Before evaluating a particular website, Cloudflare builds a baseline for each industry and category by sending prompts that do not specify a particular brand, then recording the cited websites, their positions, and the extent of their presence. It reuses this data for calculations within the same category, enabling results to load immediately and reducing repeated model calls, while also helping extract an Industry Fit score to measure how prominently the website appears alongside its actual competitors.
Activity Data and Early Access
To account for differences in model responses, Cloudflare uses the AI Gateway to send prompts multiple times through different models, then analyzes the response texts and cited sources. It uses Workers AI when an evaluation based on judgment is needed, alongside direct textual analysis instead of having the model evaluate its own output.
The tool also displays AI operator activity, including crawling and referrals by operator, such as OpenAI and Google, while showing errors encountered by requests, including 403 and 404. Cloudflare says this data allows website owners to rescan and measure the impact of changes on the questions that bring them business. Agent Readiness is accessible from the Overview tab in the dashboard, while the company allows users to request early access to AEO Visibility.