The AI Search Paradox: When LLMs Recommend Your Product Using Your Competitor’s Content

By The Tech & Marketing Desk

In the modern landscape of B2B software evaluation, prospective buyers rarely rely solely on traditional search engines. Instead, they increasingly turn to conversational AI platforms and search generative experiences—such as Google AI Overview, ChatGPT, Perplexity, Claude, and Gemini—to narrow down shortlists and weigh technical trade-offs.

However, a groundbreaking new study by Foundation Labs reveals a startling vulnerability in how these artificial intelligence systems construct their answers. When asked to evaluate specific software products, generative AI frequently bypasses first-party vendor resources entirely, pulling its evidence, trade-off analysis, and product criticisms from competitor comparison pages and independent review hubs.

The findings expose a critical blind spot in modern Generative Engine Optimization (GEO). Companies that fail to proactively build and optimize their "Versus Layer"—the digital ecosystem of comparison, review, and category pages—risk letting rivals define their strengths, weaknesses, and market positioning within the world’s most powerful AI systems.

Main Facts: The "Versus Layer" Blind Spot

The research, which focused on the Applicant Tracking Systems (ATS) software category, evaluated six major AI surfaces across 20 distinct buying prompts. Researchers tested four leading ATS vendors: Greenhouse, Ashby, Lever, and Workable.

The headline findings underscore a major disconnect between enterprise marketing efforts and AI citation mechanics:

• The Evidence Gap: Across all direct two-product comparison prompts, the four tested ATS vendors supplied a mere 27% of the visible evidence utilized by AI surfaces. The remaining 73% originated from third-party sources.
• The Competitor Citation Phenomenon: In multiple instances, AI models cited a competitor’s dedicated comparison page to explain a rival product’s weaknesses or suitability for specific mid-market use cases.
• Format Dominance: Across the entire study, listicles and comparison pages accounted for 58.7% of surface-weighted source presence. For direct head-to-head questions, dedicated comparison pages spiked to 42.1% of all visible citations.
• Platform Divergence: While platforms like Perplexity, Gemini, and Google AI Overview relied heavily on external comparison pages and listicles, ChatGPT uniquely skewed toward first-party product documentation, support pages, and pricing guides.

Chronology of the Study: How Foundation Labs Audited the AI Surfaces

To understand how AI systems assemble purchasing recommendations, Foundation Labs structured a controlled, repeatable audit.

Phase 1: Category Selection and Baseline Audits

Researchers chose the ATS category because it presents a clean testing ground: software buyers evaluate these tools against standardized criteria, multiple credible players compete for overlapping demographics, and first-party comparison coverage varies wildly.

Before running a single prompt, the research team audited the foundational web presence of the four brands (Greenhouse, Ashby, Lever, and Workable) across six possible direct matchups:

• Lever had established dedicated comparison pages for all three of its potential matchups.
• Ashby and Workable maintained coverage for two matchups each.
• Greenhouse—one of the category’s market leaders—offered zero dedicated first-party comparison pages for any of its direct matchups.

Phase 2: Prompt Engineering and Execution

The team developed 20 structured buying prompts designed to mimic real-world mid-market procurement journeys.

• Direct Matchups (6 prompts): Head-to-head comparisons, such as asking AI to weigh Greenhouse against Ashby for an in-house recruiting team based on workflows, CRM capabilities, reporting, integrations, and scalability.
• Alternative Queries (4 prompts): Requests for viable alternatives to a named platform.
• Neutral Category Questions (10 prompts): Broad discovery prompts focusing on scalability, administration, analytics, and candidate experience.

These 20 prompts were executed twice across six AI surfaces—ChatGPT, Perplexity, Gemini, Claude, Google AI Mode, and Google AI Overview—yielding 240 distinct AI-generated answers and 2,695 individual source presences.

Supporting Data: Dissecting the Source Mix and Search Metrics

The data collected from the 240 generated answers revealed that traditional Search Engine Optimization (SEO) metrics do not always correlate with AI citation success.

Traffic Does Not Equal AI Visibility

One of the study’s most striking revelations centered around the performance of low-traffic comparison URLs. For instance, Lever’s "Greenhouse vs. Ashby" guide appeared in 27 answers across 14 prompts, despite Ahrefs estimating its organic search traffic at roughly eight monthly visits, a URL Rating (UR) of 0, and only two referring domains.

Conversely, Ashby’s competing first-party page boasted 20 monthly visits, a UR of 19, and six referring domains—yet it appeared in nine fewer answers than Lever’s page. Similarly, Venture Harbour’s "Workable vs. Greenhouse" page drove an estimated 11 monthly visits with a UR of 4.5, yet successfully surfaced across 13 answers spanning five distinct AI surfaces.

How AI Reaches for Different Pages Across Buyer Stages

The data demonstrates that AI models dynamically shift their source material as a buyer moves from broad discovery to narrow evaluation:

1. Neutral Discovery: When AI must build a shortlist from scratch, broad category listicles dominate. For example, Greenhouse’s "13 Best ATS Software" article appeared in 38 answers, heavily driving top-of-funnel consideration. Listicles accounted for 51.1% of source presence on neutral prompts.
2. Shortlist Consideration: As soon as a brand name enters the query, dedicated comparison pages rise to 33.8% of source presence.
3. Head-to-Head Decisions: On direct matchups, dedicated comparison pages capture 42.1% of the citation share, as models hunt for granular feature breakdowns and explicit verdicts.

Official Responses & Industry Insights: The Platform Split

Different large language models possess drastically different retrieval-augmented generation (RAG) behaviors. While five of the six tested AI surfaces relied heavily on third-party comparison guides and review aggregators, ChatGPT stood apart by leaning aggressively into first-party documentation.

Across ChatGPT tests, product pages, help documentation, and support articles accounted for 77.3% of visible inline sources (with product pages alone making up 63.8%). Dedicated comparisons and listicles accounted for a negligible 3.3% on OpenAI’s platform.

In stark contrast, comparison pages and listicles made up 89.4% of source presence on Gemini, 72% on Google AI Overview, 69.2% on Google AI Mode, 59.2% on Perplexity, and 59.1% on Claude.

Commenting on the structural fragmentation across platforms, Ross Simmonds, CEO of Foundation Marketing, noted:

"Every platform is going to be different. Every LLM shows links differently and will pull your content differently."

This variance means B2B marketers cannot rely on a one-size-fits-all content strategy. Dominating AI search requires a bifurcated approach: satisfying third-party evaluation aggregators for Google and Perplexity while ensuring deep, pristine product documentation is indexed for ChatGPT.

Implications: Strategic Takeaways for B2B Marketers

The foundational takeaway from the ATS study is direct: If you do not supply the evidence for how your product compares to competitors, someone else will.

When brands fail to publish objective, nuanced comparison pages, AI engines pull narratives from competitor sites, third-party review platforms, or niche blogs. Consequently, an opposing vendor’s comparison page may be cited by Google AI Overview to explain your product’s reporting limitations or workflow rigidities.

Actionable Framework: Building the Versus Layer

To capture and control the Versus Layer, B2B marketing teams should execute a phased optimization strategy:

1. Prioritize Based on Evidence Leakage

Marketers must map out high-intent comparative keywords and evaluate their current coverage against AI citation behavior. By calculating an Evidence Leakage Score—weighing search demand, first-party coverage gaps, and competitor citation frequency—teams can identify which comparison pages to build first.

2. Structure Content for Instant Extraction

AI models favor pages that clearly delineate decision criteria, buyer fit, integration capabilities, and feature trade-offs early in the document. Content that buries comparative verdicts behind generic marketing fluff or gated demo CTAs consistently loses citation share to structured, objective breakdowns.

3. Audit Competitor Comparison Narratives

If a competitor’s guide is frequently cited when discussing your product, analyze their claims regarding your limitations. Publish updated, authoritative first-party evidence that directly addresses those decision points, clarifying your true capabilities and edge cases. Retesting the queries will reveal whether your first-party asset successfully supplants the outside source.

The Bottom Line

As generative search continues to cannibalize traditional organic traffic, software buyers will rely increasingly on AI-synthesized recommendations. Brands that fail to populate the Versus Layer with robust, structured, and objective comparative data are effectively flying blind—handing the pen to their competitors to write the stories that influence modern B2B purchasing decisions.