AI Agents Struggle To Read B2B Pricing

A Siteline report examined a Claude agent on 100 prime B2B software program merchandise. It discovered entry errors and hidden pricing induced the agent to go to third-party websites for information it couldn’t discover on official pages.

The report, by Siteline founder David Kaufman, concerned operating a simulated Claude agent via 534 makes an attempt to find month-to-month costs of all plans and spotlight most important options.

The information reaches the funnel stage most agent-visibility protection misses, after a purchaser is aware of your product and sends an agent to test pricing and options.

Siteline sells agent analytics and an AI agent readiness software, so it has a industrial curiosity within the discovering that many websites aren’t ready for brokers.

What Siteline’s Agent Was Requested To Do

Siteline examined 20 merchandise in 5 classes: productiveness, developer instruments, advertising and gross sales, buyer assist, and analytics. It checked if the agent may entry the positioning, retrieve plans and costs, and decide prices in tokens and gear calls.

On the median, a run on Sonnet 4.6 took about 32 seconds and $0.24, with three search-or-fetch software calls. Siteline exhibits a 2.2x time and 4.2x value distinction between the quickest tenth of runs and the slowest, primarily as a consequence of web-search calls. Linear was extra environment friendly, parsing 4 plans in a single fetch for about $0.11 on the corporate’s web site.

Entry Errors Pushed The Agent Towards Third Events

About 30% of runs face no less than one error fetching or looking out a web site, with roughly 1 / 4 of these errors being entry denials from bot blocking or unreadable pages. Most retries recuperate, however 5% abandon the model web site for third-party sources, which the report notes could be stale or incorrect.

Errors induced a steep content material hole, with access-error runs pulling 58% of content material from third-party sources, in comparison with 12% in error-free runs. Siteline notes Anthropic and OpenAI don’t run JavaScript, not like Google. SEJ lined Vercel data showing AI crawlers make up 28% of Googlebot volume. Siteline’s information exhibits 13% of runs carried inner mentions of JavaScript or rendering hassle, not counted as errors. SEJ additionally highlighted that a third of top fintech homepages returned little content, revealing a JavaScript blind spot.

The examples present this: Zendesk’s pricing web page loaded, however the plan desk was JavaScript-rendered, in line with the report, making it unreadable for the agent, who then relied on third-party blogs at 5 occasions the Linear value. Coda’s pricing fetches failed, main the agent to make use of third-party pages. Braze’s agent couldn’t entry the pricing web page and obtained numbers from G2 and Vendr.

One case proposes an easy choice. Siteline states that the agent used a non-existent pricing URL, which was not in search outcomes, then relied on third events. The report recommends preserving a pricing web page lively, even when it doesn’t present costs.

The place Pricing Pages Got here Up Brief

Throughout the runs, 65% of plans confirmed readable costs, whereas 14% posted no costs and routed to a gross sales contact. About 30% of selling, gross sales, and buyer assist merchandise had no costs, not like zero in productiveness and developer instruments

A ‘Contact Gross sales’ button is a lifeless finish for purchaser’s brokers evaluating costs. Siteline suggests it might let brokers suggest rivals with public charges. Checks reveal points: FullStory’s web page lacked costs, and Databricks was the most costly at $0.95 per run, with pay-as-you-go charges hidden behind an inaccessible calculator, redirecting to 3rd events.

Why This Issues

A client-side loaded plan desk could appear empty to brokers, even when it appears full within the browser. Siteline’s fixes, lined by SEJ, embrace rendering pricing and options server-side and highlighting key particulars early, as brokers usually pull solely the primary 15,000 to twenty,000 tokens. Siteline additionally recommends llms.txt, however Google’s guidance varies, and its worth stays unsure with impartial information.

Siteline is amongst distributors providing agent-readiness scoring, a class SEJ examined when Cloudflare released its own scanner. The benchmark assesses one mannequin on a single activity, measuring agent entry to particular websites, not how all brokers deal with each product.

Trying Forward

As extra patrons have brokers examine plans earlier than partaking gross sales, websites with clear, readable plan particulars on first view assist brokers symbolize confidently. The query is whether or not agent-readiness measures will depend on shared metrics or diverge throughout vendor scorecards, every specializing in totally different indicators.

Featured Picture: Vasilyev Alexandr/Shutterstock