A line of site report tested an agent Claude on 100 best B2B software products. He discovered that access errors and hidden prices were causing the agent to visit third-party sites for information that he could not find on the official pages.
The report, written by Siteline founder David Kaufman, involved running a simulated Claude agent through 534 attempts to discover monthly prices for all plans and highlight key features.
Data reaches the stage of the funnel that most agent visibility coverage misses, once a buyer knows about your product and sends an agent to check pricing and features.
Siteline sells agent analytics and an AI agent preparation tool. It therefore has a business interest to discover that many sites are not prepared for agents.
What the Siteline agent was asked to do
Siteline tested 20 products across five categories: productivity, developer tools, marketing and sales, customer support and analytics. It checked whether the agent could access the site, retrieve plans and prices, and determine costs in tokens and tool calls.
On a median, a run on Sonnet 4.6 took about 32 seconds and $0.24, with three calls to the search or recovery tool. Siteline shows a 2.2x time and 4.2x cost difference between the fastest and slowest execution tenth, primarily due to web search calls. Linear was more efficient, analyzing four shots in one go for about $0.11 on the company’s site.
Access errors pushed the agent to third parties
About 30% of runs experience at least one error when retrieving or searching for a site, with about a quarter of those errors being access denials due to bot blocking or unreadable pages. Most attempts are scraped, but 5% abandon the brand’s site for third-party sources, which the report says may be outdated or incorrect.
The errors caused a significant gap in content, with runs with access errors pulling 58% of content from third-party sources, compared to 12% for runs without errors. Siteline notes that Anthropic and OpenAI don’t run JavaScript, while Google does. SEJ covered Vercel Data Shows AI Crawlers Make Up 28% of Googlebot Volume. Siteline data shows that 13% of runs had internal mentions of JavaScript or rendering issues, not counted as errors. SEJ also pointed out that a third of top fintech landing pages returned little contentrevealing a JavaScript blind spot.
The examples show this: Zendesk’s pricing page loaded, but the plan table was rendered in JavaScript, according to the report, making it unreadable for the agent, who then relied on third-party blogs at a cost five times the linear cost. Coda’s price retrievals failed, leading the agent to use third-party pages. Braze’s agent was unable to access the pricing page and got the numbers from G2 and Vendr.
One case offers a simple option. Siteline says the agent used a non-existent pricing URL, which was not included in search results, and therefore relied on third parties. The report recommends keeping a pricing page active, even if it doesn’t display prices.
Where pricing pages fell short
During the rounds, 65% of plans displayed readable prices, while 14% displayed no prices and were routed to a sales contact. About 30% of marketing, sales, and customer support products were priceless, while productivity and development tools were.
A “Contact Sales” button is a dead end for buyer’s agents comparing prices. Siteline suggests letting agents recommend competitors offering public rates. Testing reveals issues: FullStory’s page lacked pricing and Databricks was the most expensive at $0.95 per run, with pay-per-view pricing hidden behind an inaccessible calculator, redirecting to third parties.
Why it matters
A plan table loaded on the client side may appear empty to agents, even if it appears complete in the browser. Siteline fixes, covered by SEJ, include server-side rendering of prices and functionality and early highlighting of key details, as agents typically only mine the first 15,000 to 20,000 tokens. Siteline also recommends llms.txt, but Google advice variesand its value remains uncertain with independent data.
Siteline is among the vendors offering Agent Readiness Rating, a category reviewed by SEJ when Cloudflare released its own scanner. The benchmark evaluates a model on a single task, measuring agent access to specific sites, not how all agents handle each product.
Looking to the future
As more buyers ask their agents to compare plans before committing to a sale, sites with clear, readable plan details at first glance help agents represent with confidence. The question is whether agent readiness metrics will build on shared metrics or diverge across vendor dashboards, with each focusing on different signals.
Featured Image: Vasilyev Alexander/Shutterstock
