AI Agents Hit a Wall: B2B Software Sites Unprepared for the Autonomous Buyer

San Francisco, CA – A groundbreaking new report reveals significant friction for artificial intelligence agents attempting to navigate the digital landscape of B2B software products. Commissioned by Siteline, a company specializing in AI agent analytics, the study found that a simulated Claude agent frequently encountered access errors, unreadable content, and hidden pricing, forcing it to abandon official brand websites in favor of potentially stale or incorrect third-party sources. This critical oversight by B2B vendors threatens to impede the burgeoning trend of AI-driven purchasing, raising alarms across industries.

The comprehensive benchmark, detailed in a report by Siteline founder David Kaufman, involved deploying a sophisticated Claude agent to investigate 100 top B2B software products across five key categories. Over 534 attempts, the agent was tasked with uncovering monthly pricing for all available plans and highlighting main features. The findings illuminate a crucial blind spot in current B2B digital strategies: many websites are fundamentally unprepared for the autonomous agents increasingly being deployed by prospective buyers. This data reaches into the critical funnel stage that much traditional agent-visibility coverage misses – after a buyer has identified a product of interest and dispatches an AI agent to gather specific pricing and feature comparisons.

Siteline, with its commercial interest in agent analytics and AI agent readiness tools, positions these findings not merely as an academic exercise but as a vital warning to the B2B sector. The report underscores a looming paradigm shift in how businesses research and procure software, urging companies to adapt their online presence or risk becoming invisible to a new generation of AI-powered decision-makers.

The Dawn of Agent-Driven Discovery: A Chronology of Challenges

The evolution of the B2B buying journey has been relentless, shifting from direct sales interactions to extensive self-service research. In recent years, the proliferation of generative AI has introduced a new, formidable player into this ecosystem: the autonomous AI agent. These agents are designed to mimic human research, sifting through vast amounts of information to extract precise data points, compare offerings, and even recommend solutions.

Siteline’s initiative can be viewed as a timely intervention, anticipating the inevitable future where AI agents become a standard "first touchpoint" for B2B procurement. Recognizing this impending shift, David Kaufman and his team embarked on a mission to quantitatively assess the readiness of the B2B web. The methodology was meticulously designed to simulate a real-world scenario: a buyer, having identified potential software solutions, deploys an AI agent to perform the grunt work of pricing and feature comparison.

The simulated Claude agent, specifically leveraging the Sonnet 4.6 model, was programmed with a clear mandate: access designated B2B software websites, identify all pricing plans, retrieve their monthly costs, and extract key features. This involved navigating complex web structures, parsing content, and making logical deductions, much like a human researcher. From the outset of the testing phase, initial observations hinted at significant friction. The agent frequently encountered unexpected barriers, ranging from seemingly minor technical glitches to outright data black holes. This foreshadowed the comprehensive data that would eventually paint a stark picture of unpreparedness across a substantial portion of the B2B software market. The study’s "chronology" is less about historical dates and more about the unfolding narrative of an agent’s struggle in a digital environment not built for its intelligence.

Unpacking the Data: Supporting Evidence and Deep Dives

The Siteline report meticulously details the agent’s performance, the obstacles it faced, and the resulting compromises in data accuracy and cost-efficiency. The findings present a compelling case for immediate action from B2B software providers.

The Agent’s Mandate and Performance Metrics

The core task set for Siteline’s Claude agent was unambiguous: access the websites of 100 leading B2B software products, retrieve comprehensive details of all available plans and their monthly prices, and identify main features. Beyond just data extraction, the study also measured the operational costs incurred by the agent, specifically in terms of tokens consumed and tool calls made during each attempt.

At the median, a single run using the Claude Sonnet 4.6 model took approximately 32 seconds and cost about $0.24, involving three search-or-fetch tool calls. However, these median figures mask significant variances. Siteline’s data reveals a stark 2.2x difference in time and a 4.2x difference in cost between the fastest tenth of runs and the slowest. This considerable disparity was primarily attributed to an increased number of web-search calls, indicating the agent struggling to find information directly and resorting to broader, more resource-intensive searches. For instance, a more efficient website like Linear, which parsed four plans in a single fetch, demonstrated superior performance, costing only about $0.11 per run. This highlights a clear correlation: optimized website structure and transparent data presentation directly translate to lower operational costs and faster data retrieval for AI agents.

The Pervasive Problem of Access Errors

One of the most critical findings of the report was the prevalence of access errors. Approximately 30% of all agent runs encountered at least one error during the process of fetching or searching a site. A significant portion of these, roughly a quarter of all errors, were outright access denials. These denials stemmed from various issues, including sophisticated bot-blocking mechanisms or pages rendered in a way that made them unreadable for the AI agent.

While most retries proved successful, allowing the agent to eventually recover and proceed, a concerning 5% of all runs resulted in the agent abandoning the brand’s official site entirely. In these cases, the agent was forced to seek information from third-party sources. This reliance on external sites, such as review platforms, industry blogs, or reseller pages, introduces a significant risk. As the report notes, information on these third-party sites can be stale, outdated, or outright incorrect, compromising the accuracy of the data presented to the end buyer.

This issue creates a steep "content gap." Runs that faced access errors pulled a staggering 58% of their content from third-party sources, a stark contrast to the mere 12% retrieved from external sites during error-free runs. This discrepancy underscores the critical impact of website accessibility on data integrity for AI agents.

Technical Deep Dive: The JavaScript Blind Spot

A major technical culprit behind these access errors is the way many modern websites are built. The report highlights that AI models like those from Anthropic (Claude) and OpenAI do not inherently run JavaScript, unlike Google’s crawlers. This difference is crucial. While Google’s advanced crawlers, which Vercel data cited by SEJ suggests constitute 28% of Googlebot’s traffic, are adept at rendering JavaScript-heavy pages, many AI agents operate with a more limited rendering capability. Siteline’s data noted that 13% of runs internally flagged JavaScript or rendering troubles, even if not formally counted as an "error." This aligns with previous SEJ findings that a third of top fintech homepages returned little content, revealing a widespread JavaScript blind spot for AI agents.

The report provides compelling illustrative case studies:

• Zendesk: The agent successfully loaded Zendesk’s pricing page, but the crucial plan table was rendered using JavaScript. This made the table unreadable for the agent, which then had to rely on third-party blogs. This fallback was not only less reliable but also five times more expensive in terms of operational cost than a direct fetch would have been.
• Coda: For Coda, the pricing fetches failed outright. Consequently, the agent was directed to gather information from various third-party pages, again raising concerns about data freshness and accuracy.
• Braze: The agent was completely unable to access Braze’s pricing page. As a result, it obtained pricing figures from popular review and vendor platforms like G2 and Vendr, bypassing the official source entirely.

These examples vividly demonstrate how modern web development practices, while enhancing user experience for humans, inadvertently create barriers for AI agents, pushing them away from authoritative sources.

One particularly straightforward issue highlighted by the report involved a case where the agent attempted to use a non-existent pricing URL that wasn’t even present in search results, subsequently relying on third parties. Siteline offers a simple, yet potent recommendation: companies should maintain an active, accessible pricing page, even if it explicitly states "Contact Sales" rather than displaying specific numbers. This ensures the agent finds an authoritative source, preventing it from straying to potentially inaccurate external sites.

The Enigma of Hidden Pricing

Beyond technical access issues, the content itself presented another significant hurdle: hidden pricing. Across all runs, only 65% of plans displayed readable prices directly on the page. A considerable 14% of plans offered no explicit prices, instead routing the agent to a "Contact Sales" prompt.

A notable disparity emerged across product categories. Approximately 30% of marketing, sales, and customer support products completely omitted pricing information, pushing buyers towards a sales inquiry. In stark contrast, productivity and developer tools categories had zero instances of hidden pricing, consistently providing transparent rates. This suggests that some sectors are more inclined to gate pricing information, a strategy that proves counterproductive in the age of AI agents.

A "Contact Sales" button, while a standard lead generation tactic for human buyers, represents a dead end for an autonomous AI agent tasked with direct price comparison. Siteline’s report suggests that such a barrier could inadvertently lead agents to recommend competitors who offer transparent, publicly available rates. This implies a significant risk of losing potential leads to more forthcoming rivals.

Specific examples underscore this challenge:

• FullStory: The agent found FullStory’s page to be devoid of explicit prices, forcing it into an unproductive loop.
• Databricks: This product proved to be the most expensive to investigate, costing $0.95 per run. Its pay-as-you-go rates were hidden behind an inaccessible calculator, which further redirected the agent to third-party sources. The high cost here is not just monetary but also in terms of the agent’s efficiency and the buyer’s budget for research.

Hidden costs and opaque pricing structures, while perhaps intended to encourage direct engagement, are fundamentally at odds with the operational logic of AI agents. These agents are built for efficiency and direct comparison, and any barrier to readily available information will invariably lead them elsewhere, often to less reliable sources or competing products.

Industry Reactions and Proactive Measures

While the Siteline report primarily focuses on the technical challenges and findings, its implications resonate deeply within the B2B software industry and among AI developers. As of the report’s initial publication, specific official responses from the 100 tested companies have yet to emerge publicly. However, the findings serve as a silent alarm, underscoring a critical need for self-assessment and strategic adaptation across the sector.

From the perspective of B2B vendors, the report acts as a powerful call to action. Companies must acknowledge that the traditional website design principles, often optimized for human visual scanning and manual form completion, are no longer sufficient. Internal audits of digital properties, particularly pricing and features pages, are becoming an imperative. This involves evaluating how effectively their sites can be parsed by automated systems, not just human eyes. There’s a growing necessity for the development of "agent-friendly" design guidelines that prioritize machine readability without compromising human user experience. This could involve, for instance, structuring data in predictable formats, using semantic HTML, and clearly labeling key information.

For AI developers like Anthropic (creators of Claude) and OpenAI, these challenges highlight ongoing areas for agent improvement. While current agents might struggle with dynamic JavaScript rendering, future iterations will likely incorporate more sophisticated web browsing capabilities. However, even with advanced agents, the onus remains on website owners to present information clearly. The report indirectly pushes AI developers to continue refining their agents’ error recovery mechanisms and their ability to discern authoritative information from less reliable sources.

The search engine giant, Google, has long championed web standards and accessibility. Its guidance on llms.txt (a proposed standard for controlling AI crawler access, similar to robots.txt) is an attempt to provide tools for website owners to manage AI interaction. However, as the SEJ article noted, Google’s guidance on llms.txt can vary depending on which product you ask, indicating a lack of clear, unified direction and suggesting its value remains uncertain for independent data collection. This highlights the nascent stage of AI-specific web protocols and the need for more consistent industry standards.

In response to this emerging landscape, a new market category is rapidly forming: "agent readiness" tools. Siteline itself is a vendor in this space, offering analytics and tools to assess how well a website performs for AI agents. Cloudflare, a major web infrastructure company, also released its own agent-readiness scanner, a development covered by SEJ. These tools provide objective scoring and insights, allowing companies to benchmark their site’s "agent-friendliness" and identify areas for improvement. While the current benchmark assesses one specific model on a single task, it sets a precedent for how all agents might interact with various products, establishing a foundational metric for future comparisons.

The Road Ahead: Strategic Implications and Future Outlook

The Siteline report is more than just a technical assessment; it’s a strategic forecast for the future of B2B commerce. As more and more buyers, particularly within enterprise settings, adopt AI agents to conduct preliminary research and compare plans before engaging human sales representatives, the imperative for B2B companies to adapt their digital storefronts becomes undeniable.

Key Recommendations Revisited

The findings offer clear, actionable recommendations for B2B software vendors:

1. Server-Side Rendering: Critical information, especially pricing and key features, should be rendered server-side. This ensures that the content is directly available in the HTML when the page is loaded, making it immediately readable for AI agents that may not execute client-side JavaScript.
2. Prioritize Key Details Early: AI agents typically pull only the first 15,000 to 20,000 tokens of content from a page. Therefore, essential information such as core features, pricing tiers, and unique selling propositions must be highlighted early in the page structure.
3. Maintain Active Pricing Pages: Even if a company’s pricing model requires a "Contact Sales" interaction, maintaining an active, accessible pricing page is crucial. This page should clearly state that pricing is custom or requires consultation, preventing agents from being misled or forced to rely on potentially outdated third-party sources. A non-existent or inaccessible pricing page is a dead end that actively pushes agents (and their human masters) away.

The "Agent Readiness" Ecosystem

The rise of agent-readiness scoring, pioneered by companies like Siteline and Cloudflare, signifies the emergence of a new set of metrics for website performance. The question for the industry now is whether these agent-readiness measures will coalesce around shared, standardized metrics, or diverge across various vendor scorecards, each focusing on different signals and proprietary methodologies. A unified standard would benefit the entire ecosystem, providing clear benchmarks for B2B companies and consistent data for AI agents.

Competitive Advantage and Ethical Considerations

In this evolving landscape, companies that embrace transparency and optimize their sites for AI agent readability will gain a significant competitive advantage. Clear, easily discoverable plan details will enable agents to represent a product confidently and accurately to their human principals, streamlining the pre-sales funnel. Conversely, sites that remain opaque or technically challenging risk being overlooked by AI agents, effectively becoming invisible in the new digital marketplace. This could lead to a substantial loss of qualified leads and market share.

Beyond competitive dynamics, the report also touches upon ethical considerations. The reliance on potentially stale or incorrect third-party data raises concerns about data accuracy and the potential for AI agents to inadvertently propagate misinformation or recommend suboptimal solutions based on flawed data. Ensuring that AI agents access authoritative, up-to-date information is not just a technical challenge but an ethical imperative for responsible AI deployment in business.

Conclusion

The Siteline report serves as a stark reminder that the digital transformation journey is far from over. As AI agents become increasingly sophisticated and integrated into the B2B purchasing process, the onus is on software vendors to adapt their digital presence. The days of solely designing for human eyes are waning; the future demands websites that are equally intelligible to autonomous AI. Companies that proactively address issues of access, rendering, and pricing transparency will not only optimize their sales funnels but also solidify their position as leaders in the age of intelligent automation. Ignoring these signals risks rendering a company’s offerings effectively invisible to the autonomous buyers of tomorrow.