Siteline Report evaluated Claude Agent in the top 100 B2B software products. It found access errors and hidden values that caused the agent to visit third-party sites to find information they couldn’t find on official pages.
The report, by Siteline founder David Kaufman, involved using an agent made by Claude with 534 attempts to find monthly prices for all plans and highlight key features.
Data reaches the most missed part of the funnel in agent visibility, after the buyer gets to know your product and sends the agent to check out prices and features.
Siteline sells agent analytics and an AI agent readiness tool, so it has a commercial interest in finding out that many sites are not optimized for agents.
What the Siteline Agent Is Asked To Do
Siteline tested 20 products in five categories: productivity, developer tools, marketing and sales, customer support, and analytics. Check if the agent can access the site, find plans and prices, and determine the cost of tokens and tool calls.
At home, running on Sonnet 4.6 took about 32 seconds and $0.24, with three calls to search or download tools. The site line shows a 2.2x time and 4.2x cost difference between the ten fastest and slowest runs, mostly due to web search calls. Linear was very efficient, splitting four programs into one download for about $0.11 on the company’s site.
Access Errors Push Agent to External Groups
About 30% of runs face at least one error to download or search the site, about a quarter of those errors are denial of access to bot blocking or unreadable pages. Many try and recover, but 5% leave the product site because of third-party sources, a comment report that may or may not be correct.
The errors caused a steep content gap, with faulty access pulling 58% of content from third-party sources, compared to 12% of error-free runs. Site line notes Anthropic and OpenAI do not use JavaScript, unlike Google. SEJ covered Vercel data showing AI searches accounted for 28% of Googlebot volume. Siteline’s data shows 13% of runs dealt with an internal JavaScript error or rendering problem, which could be counted as errors. SEJ also highlighted that a third of the top fintech homepages returned minimal content, revealing JavaScript’s blind spot.
Examples show this: Zendesk’s pricing page was loaded, but the schedule table was rendered in JavaScript, according to the report, making it unreadable for the agent, who then relied on third-party blogs at five times the cost of Linear. Retrieving the Coda price failed, leading the agent to use third-party pages. The Braze agent was unable to access the pricing page and got the numbers from G2 and Vendr.
One case suggests a straightforward option. The site line says that the agent used a URL for a missing price, which was not in the search results, and depended on third parties. The report recommends keeping the pricing page active, even if it doesn’t show prices.
Where Price Pages Came In Brief
Across all runs, 65% of systems displayed readable values, while 14% did not send values and were forwarded to a sales contact. About 30% of marketing, sales, and customer support products had no pricing, as opposed to zero in manufacturing and developer tools
The ‘Sell to Contact’ button is a dead end for buyer’s agents comparing prices. Siteline suggests that it may allow agents to recommend competitors at public prices. Testing revealed problems: FullStory’s page had no prices, and Databricks was overpriced at $0.95 per run, with pay-as-you-go prices hidden behind an inaccessible counter, redirecting third parties.
Why This Matters
A blueprint table loaded on the client side may appear empty to agents, even if it looks complete in the browser. Siteline’s fixes, covered by SEJ, include offering server-side pricing and features and highlighting important details early, as agents typically only withdraw the first 15,000 to 20,000 tokens. Siteline also recommends llms.txt, but Google’s guidance varies, and its value remains uncertain for independent data.
Siteline is among the vendors that offer agent readiness scores, a category SEJ tested when Cloudflare released its scanner. The benchmark tests one model in one operation, benchmarking agent access to specific sites, not how all agents handle the entire product.
Looking Forward
Since many buyers have agents compare plans before selling, sites with clear, readable plan details at first glance help agents represent with confidence. The question is whether measures of agent readiness will be based on shared metrics or split between dealer scorecards, each focusing on different signals.
Featured Image: Vasiliev Alexandr/Shutterstock
