AI Agents Struggle To Read B2B Pricing, Report Finds

A Siteline report tested a Claude supplier connected 100 apical B2B package products. It recovered entree errors and hidden pricing caused the supplier to sojourn third-party sites for info it couldn’t find connected charismatic pages.

The report, by Siteline laminitis David Kaufman, progressive moving a simulated Claude supplier done 534 attempts to observe monthly prices of each plans and item main features.

The information reaches the chimney shape astir agent-visibility sum misses, aft a purchaser knows your merchandise and sends an supplier to cheque pricing and features.

Siteline sells supplier analytics and an AI supplier readiness tool, truthful it has a commercialized liking successful the uncovering that galore sites aren’t prepared for agents.

What Siteline’s Agent Was Asked To Do

Siteline tested 20 products successful 5 categories: productivity, developer tools, trading and sales, customer support, and analytics. It checked if the supplier could entree the site, retrieve plans and prices, and find costs successful tokens and instrumentality calls.

At the median, a tally connected Sonnet 4.6 took astir 32 seconds and $0.24, pinch 3 search-or-fetch instrumentality calls. Siteline shows a 2.2x clip and 4.2x costs quality betwixt the fastest tenth of runs and the slowest, chiefly owed to web-search calls. Linear was much efficient, parsing 4 plans successful 1 fetch for astir $0.11 connected the company’s site.

Access Errors Pushed The Agent Toward Third Parties

About 30% of runs look astatine slightest 1 correction fetching aliases searching a site, pinch astir a 4th of those errors being entree denials from bot blocking aliases unreadable pages. Most retries recover, but 5% wantonness the marque tract for third-party sources, which the study notes tin beryllium old aliases incorrect.

Errors caused a steep contented gap, pinch access-error runs pulling 58% of contented from third-party sources, compared to 12% successful error-free runs. Siteline notes Anthropic and OpenAI don’t tally JavaScript, dissimilar Google. SEJ covered Vercel information showing AI crawlers dress up 28% of Googlebot volume. Siteline’s information shows 13% of runs carried soul mentions of JavaScript aliases rendering trouble, not counted arsenic errors. SEJ besides highlighted that a 3rd of apical fintech homepages returned small content, revealing a JavaScript unsighted spot.

The examples show this: Zendesk’s pricing page loaded, but the scheme array was JavaScript-rendered, according to the report, making it unreadable for the agent, who past relied connected third-party blogs astatine 5 times the Linear cost. Coda’s pricing fetches failed, starring the supplier to usage third-party pages. Braze’s supplier couldn’t entree the pricing page and obtained numbers from G2 and Vendr.

One lawsuit proposes a straightforward option. Siteline states that the supplier utilized a non-existent pricing URL, which was not successful hunt results, past depended connected 3rd parties. The study recommends keeping a pricing page active, moreover if it doesn’t show prices.

Where Pricing Pages Came Up Short

Across the runs, 65% of plans showed readable prices, while 14% posted nary prices and routed to a income contact. About 30% of marketing, sales, and customer support products had nary prices, dissimilar zero successful productivity and developer tools

A ‘Contact Sales’ fastener is simply a dormant extremity for buyer’s agents comparing prices. Siteline suggests it whitethorn fto agents urge competitors pinch nationalist rates. Tests uncover issues: FullStory’s page lacked prices, and Databricks was the astir costly astatine $0.95 per run, pinch pay-as-you-go rates hidden down an inaccessible calculator, redirecting to 3rd parties.

Why This Matters

A client-side loaded scheme array whitethorn look quiet to agents, moreover if it looks complete successful the browser. Siteline’s fixes, covered by SEJ, see rendering pricing and features server-side and highlighting cardinal specifications early, arsenic agents typically propulsion only the first 15,000 to 20,000 tokens. Siteline besides recommends llms.txt, but Google’s guidance varies, and its worth remains uncertain pinch independent data.

Siteline is among vendors offering agent-readiness scoring, a class SEJ examined when Cloudflare released its ain scanner. The benchmark assesses 1 exemplary connected a azygous task, measuring supplier entree to circumstantial sites, not really each agents grip each product.

Looking Ahead

As much buyers person agents comparison plans earlier engaging sales, sites pinch clear, readable scheme specifications connected first position thief agents correspond confidently. The mobility is whether agent-readiness measures will trust connected shared metrics aliases diverge crossed vendor scorecards, each focusing connected different signals.

Featured Image: Vasilyev Alexandr/Shutterstock