AI Agents Can't Read B2B Pricing Pages — and That's Sending Buyers to Your Competitors

B2B software companies risk losing sales to competitors as AI shopping agents increasingly fail to read pricing pages. A new study warns that JavaScript-rendered pricing and hidden costs are the core culprits — and the commercial consequences are already playing out in real buying decisions.

A Siteline study tested a Claude AI agent across 100 top B2B software products and found that access errors and unreadable pricing pages forced the agent to rely on third-party sources rather than official company websites. For businesses that have invested heavily in polished pricing pages, the findings expose a critical blind spot in their go-to-market strategy.

---

What the Study Tested and Found

Siteline founder David Kaufman ran a simulated Claude agent through 534 attempts across 20 products in five categories: productivity, developer tools, marketing and sales, customer support, and analytics. The agent's task was straightforward — find monthly prices for all available plans and identify key features.

The results were sobering. At the median, a single pricing research run on Claude Sonnet 4.6 took about 32 seconds and cost approximately $0.24 using three search-or-fetch tool calls. However, Siteline recorded a 4.2x cost difference between the fastest and slowest runs, primarily driven by additional web-search calls when the agent hit dead ends.

Across all runs, only 65% of plans showed readable prices. About 14% of products posted no prices at all and routed the agent to a sales contact form instead. The problem was most concentrated in marketing, sales, and customer support products, where roughly 30% had no public pricing. By contrast, productivity and developer tool categories showed zero instances of hidden pricing.

The data also revealed a sharp content gap when errors occurred. Runs that encountered access errors pulled 58% of their content from third-party sources, compared to just 12% in error-free runs. Those third-party sources — review platforms, comparison sites, and blogs — may carry outdated or incorrect pricing information, introducing meaningful risk into the buyer's decision-making process.

Why This Matters Beyond Pricing Accuracy

This isn't purely a data accuracy problem. When an AI agent retrieves pricing from an unverified third party, it may present that figure to a buyer with no indication that the number is stale, estimated, or simply wrong. A buyer comparing five vendors through an agent could be operating on a fundamentally distorted picture — and the vendor whose page was unreadable won't even know the conversation happened. Understanding the broader risks and challenges of AI in business decision-making helps frame why this technical gap has real revenue consequences.

---

JavaScript Is the Hidden Enemy of AI Agent Visibility

The study points to JavaScript rendering as the single most damaging technical barrier between AI agents and B2B pricing data. Siteline notes that Anthropic and OpenAI agents do not execute JavaScript the way Google does. This means a pricing table that loads perfectly in a browser may appear completely empty to an AI agent crawling the same page.

Thirteen percent of all runs in the study contained internal mentions of JavaScript or rendering trouble that were not even counted as formal errors — suggesting the real scale of the problem is larger than the headline figures indicate.

Real-World Examples From the Study

Real-world examples from the study illustrate the damage clearly:

• Zendesk's pricing page loaded successfully, but its plan comparison table was JavaScript-rendered and therefore unreadable to the agent. The agent pivoted to third-party blogs at five times the cost of a clean run.
• Coda's pricing fetches failed entirely, pushing the agent to external pages.
• Braze's pricing page was inaccessible, and the agent retrieved numbers from G2 and Vendr — two third-party review platforms with no guarantee of current accuracy.
• Databricks represented the most expensive single run in the study at $0.95 per session. Its pay-as-you-go rates were hidden behind a pricing calculator the agent could not access, which triggered a cascade of third-party lookups.

This mirrors a pattern Search Engine Journal previously reported, where a third of top fintech homepages returned little usable content to AI crawlers due to the same JavaScript blind spot. AI crawlers now represent 28% of Googlebot volume according to Vercel data — a share large enough that ignoring agent readiness carries measurable commercial risk.

The Structural Shift Already Underway

The scale of AI crawler activity signals that this is not an emerging edge case. It is an active and growing channel through which buyers are forming vendor shortlists. Companies that treat agent readiness as a future concern are already ceding ground. This shift connects directly to the barriers to AI adoption that B2B organisations face — many of which stem from technical infrastructure decisions made long before AI agents became a procurement reality.

---

What B2B Companies Can Do Right Now

The findings carry direct implications for revenue teams and web developers at B2B software companies. A "Contact Sales" button may feel like a strategic gate, but to an AI agent it is simply a dead end. When an agent cannot find pricing, it will find a competitor who publishes theirs — and recommend them instead.

Technical Fixes That Improve Agent Readability

Siteline recommends several practical fixes that can be implemented without overhauling an entire website:

• Render pricing content server-side so agents can read it on first load without executing JavaScript. This is the single highest-impact change available to most development teams.
• Position key details early in the page's content. Plan names, prices, and differentiating features should appear within the first 15,000 to 20,000 tokens of any page, since agents typically stop processing beyond that threshold.
• Maintain an active pricing page even when a company prefers not to publish specific numbers. Siteline's data includes one case where an agent attempted to visit a pricing URL that did not exist at all and was absent from search results before defaulting to third-party sources. An accessible page with at least structural information gives the agent something to work with.

Strategic Considerations for "Contact Sales" Models

If your company uses a contact-sales model for commercial reasons, consider publishing at minimum a set of starting price ranges or plan tiers. The goal is not necessarily to disclose every pricing variable — it is to ensure the agent has something authoritative to reference rather than defaulting to a competitor's listing or an outdated review post.

The broader shift in how B2B purchasing decisions are being made — with buyers using agents to conduct initial vendor comparisons before any sales interaction — connects to larger changes in how B2B transactions and commercial relationships are evolving. Pricing transparency is becoming a structural prerequisite for being included in a shortlist at all.

On llms.txt Files

Siteline also raises the potential value of llms.txt files as a mechanism for guiding agent behaviour. However, Search Engine Journal has noted that Google's own guidance on llms.txt remains inconsistent, and independent data on its effectiveness is limited. It is worth monitoring developments in this area, but llms.txt should not be treated as a substitute for server-side rendering of core commercial content. For a deeper technical reference on how AI agents interact with web content, the W3C Web Architecture documentation provides useful foundational context.

A Note on the Study's Limitations

It is worth noting that Siteline sells agent analytics and an AI agent readiness tool, giving the company a commercial interest in findings that highlight widespread unpreparedness. The benchmark also tested one model on one task, meaning results may not reflect how every agent handles every product across the entire market. The directional findings are nonetheless consistent with patterns reported independently by other sources, including Search Engine Journal and Vercel's crawler data.

---

As the line between research and purchase continues to blur in B2B sales cycles, the companies that make their pricing legible to machines will have a structural advantage. Buyers who deploy agents to compare plans before ever speaking to sales are no longer a future scenario — they are already here.

How to Act on This Information

• Audit your pricing page for JavaScript-rendered content and confirm it returns readable HTML to a basic web crawler before your next product launch or pricing update.
• If your company uses a contact-sales model, consider publishing at least starting price ranges or plan tiers to prevent agents from recommending competitors with public rates.
• Use this research as a briefing document when making the business case to engineering and marketing leadership for server-side rendering of key commercial pages.