Sample report • Our own domain • 22 October 2026
What an AI assistant says about AgentReadyWebsite
This is a real AI Answer Audit, run on us. We asked one assistant 13 questions a prospect might ask about this business, three times each, and graded all 39 answers against our actual prices, services, and contact details. The misses are left in.
Assistant in this run: Grok. We asked Grok's engine directly through the xAI API, with web search on. The same audit runs across ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode. This page will grow a column for each as those runs are added.
The numbers
| Assistant | Facts stated when asked by name | Contradicted a price | Named us when asked for recommendations | Cited agentreadywebsite.io |
|---|---|---|---|---|
| Grok | 93% (28 of 30 answers) | 0 | 11% (1 of 9 answers) | 80% (31 of 39 answers) |
13 prompts: what we do, how to contact us, the price of each of the three packages, whether we offer each of three named services, two FAQ-style questions, and three “best X” recommendation prompts. Three runs per prompt because the same question gets a different answer each time.
Three findings
Finding 1 • Accurate when asked by name
Every price question came back right, all nine times: $199 for Basic, $599 for Medium, $1,399 for Complex, each labelled one-time, with the page ranges and the 12 months of Domain Watch. “One-time purchase or subscription?” was answered correctly three times. Every one of those answers cited our own homepage, FAQ, or About page.
“$599 one-time. According to the official AgentReadyWebsite.io site, the Medium package is a one-time payment of $599. It covers websites with 51–500 pages… and it includes the agent-readiness package (typically 13 files) plus 12 months of Domain Watch.” (Grok, run 1 of 3, citing agentreadywebsite.io and agentreadywebsite.io/faq)
What it tells us: when the facts are in visible text on the pages the assistant opens, it repeats them correctly. This is the layer a website controls.
Finding 2 • Our email address was hidden from the assistant
“How can I contact AgentReadyWebsite?” was answered with our X handle and a link to the About page. In two of three runs the email came back as the literal string [email protected]. That is what Cloudflare’s email obfuscation feature serves to a non-browser fetcher. A human sees the address. The assistant did not.
“Email: [email protected]. A human reads every message. This is the primary way to reach them…” (Grok, run 1 of 3, citing agentreadywebsite.io/about)
What we are fixing: the address now appears in the footer of every page and in the Organization structured data, and we are turning off the obfuscation feature for this domain so automated readers see it. The audit exists to find this kind of gap. We could not see it. The machine could.
Finding 3 • Asked for a recommendation, it named us once in nine tries
For “Best service that makes a small business website clear to AI agents,” “Who should I hire for…,” and “Top-rated… providers,” Grok recommended Wix, Scrunch AI, WebCare Pro, and AgentFi, and cited Moz, kite.ai, mavek.ai, and bestaeotools.com. None of those nine answers cited our site. We appeared once, as “another option.”
What it tells us: recommendations are built from third-party pages such as reviews and comparison posts. No file on our own site changes what those pages say about us, and we will not tell you otherwise. An audit that only measured the by-name questions would have scored us 93% and hidden this. Both numbers are the report.
How this was measured
We asked each assistant’s engine directly through its API, with web search turned on. The app a customer uses adds its own instructions and that person’s history, so its wording can differ from what we recorded. Google AI Mode has no API from Google, so we read the AI Mode page through SerpApi. For each run we record the full answer text, every cited URL, and the token and tool usage. We grade by matching strings against the business profile. A fact the assistant never raised is marked not stated. Contradicted is reserved for a different price appearing where ours did not.
The grader missed one thing worth saying out loud. It marked “What does AgentReadyWebsite do?” as partly stated in two runs because the answer described the package without using the exact phrase “agent-readiness package.” A human reading the answer would call it correct. We publish the grade the tool gave.
Find out what AI says about your business
Pick an assistant on the homepage and we will ask it one question about you, free. The full audit asks six assistants up to 18 questions, three times each, and comes with the fix list.