The AI Recommendation Trap: Why 73% of B2B Software Comparisons are Controlled by Outsiders

By Global Tech Desk
Published: September 2026

Executive Summary & Main Facts

When enterprise buyers turn to generative artificial intelligence to evaluate software solutions, they expect an objective, authoritative breakdown of how competing platforms stack up. However, a comprehensive new study reveals a startling blind spot for modern SaaS companies: roughly 73% of the sources shaping AI-generated product comparisons come from outside the software vendors themselves.

In a heavily researched analysis of the Applicant Tracking System (ATS) market, the research teams at Foundation and AirOps tracked 240 AI-generated answers derived from 2,695 recorded source presences across major platforms—including ChatGPT, Perplexity, Gemini, Claude, Google AI Mode, and Google AI Overview.

The findings upend traditional digital marketing playbooks. Across six direct head-to-head matchups between major ATS giants (Greenhouse, Ashby, Lever, and Workable), first-party product comparison pages accounted for a meager 27% of visible source presence.

The remaining 73%—dubbed the "Versus Layer"—was supplied by independent and commercial publishers (42.7%), alternative software vendors writing about their rivals (20%), third-party review aggregators (7.4%), and online communities (2.9%).

For software executives and marketing leaders, the implications are profound: if a brand fails to actively construct and optimize the comparative narratives surrounding its software, competitors, publishers, and algorithmic synthesis engines will happily write that narrative for them—often with costly inaccuracies.

Chronology of the Study: How the Data Was Captured

To understand how modern language models build software recommendations, the research initiative bypassed theoretical models and executed a rigorous, empirical test across the industry’s leading AI discovery engines.

Phase 1: Test Design and Execution

The team mapped out 20 distinct ATS (Applicant Tracking System) buying prompts designed to replicate real-world enterprise procurement cycles. These queries ranged from broad, neutral category requests ("What are the best recruiting software options for mid-market teams?") to hyper-specific, direct head-to-head matchups ("Compare Greenhouse versus Ashby").

Phase 2: Multi-Platform Probing

Each of the 20 buying prompts was executed twice across six distinct AI environments:

• ChatGPT (OpenAI)
• Perplexity AI
• Google Gemini
• Anthropic Claude
• Google AI Mode
• Google AI Overview

This process yielded a corpus of 240 distinct AI-generated answers.

Phase 3: Source Forensics

The research team audited every single citation and underlying link referenced by the engines, cataloging 2,695 recorded source presences. By tracing the domain ownership behind every URL, the analysts mapped out precisely who owned the digital assets fueling AI reasoning.

The results exposed a stark reality: when a buyer asks Google’s AI Overview to evaluate Greenhouse’s reporting capabilities against Ashby, the AI might pull its evidence directly from Ashby’s proprietary comparison page. If Greenhouse has neglected to build or optimize its own counter-narrative, the platform is left entirely at the mercy of competitor-authored documentation.

Supporting Data: Dissecting the "Versus Layer"

The 73% external source dominance is not a monolithic block. It is divided across distinct publisher categories, each behaving according to unique algorithmic rules.

Breakdown of the Versus Layer Sources

• Independent & Commercial Publishers: 42.7%
• Competing Software Vendors: 20.0%
• Review & Aggregator Sites: 7.4%
• Community Sources (Forums, Reddit, etc.): 2.9%

A deeper dive into the mechanics of how AI selects these pages revealed four governing patterns:

1. High Domain Authority Is Not a Prerequisite for AI Citation

Traditional Search Engine Optimization (SEO) dictates that high-ranking pages require massive backlink profiles, high Domain Ratings (DR), and thousands of monthly visitors. AI engines, however, operate differently.

Researchers pointed to a striking case study involving Lever. Lever had published comparison pages targeting three of its rivals, as well as an objective page comparing two other competitors: Greenhouse and Ashby. Despite having a URL Rating of zero, just two referring domains, and an estimated eight monthly visits according to Ahrefs, Lever’s third-party comparison page was cited by AI engines in 27 separate answers.

By contrast, Ashby’s own first-party page covering the exact same matchup—boasting a URL Rating of 19, six referring domains, and roughly 20 monthly visits—was only cited 18 times. For LLMs, contextual relevance and semantic clarity heavily outweigh traditional backlink authority.

2. Buyer Intent Shifts Source Reliance

The type of source an AI picks shifts dramatically depending on where the user is in the buying funnel:

• Neutral Category Prompts: When users asked broad, top-of-funnel questions, broad listicles dominated, accounting to 51.1% of surface-weighted source presence, compared to just 12.6% for dedicated comparison pages.
• Vendor-Inclusive Prompts: Once a specific vendor was introduced into the query, dedicated comparison pages surged to 33.8%, while listicles declined to 21.4%.
• Direct Head-to-Head Questions: On the six direct matchups, dedicated comparison pages peaked at 42.1% of source presence.

3. Structural Formatting Dictates Citation Frequency

The study analyzed why certain pages were selected over others and discovered a distinct structural bias in LLM training data. Pages that surfaced buyer fit, decision criteria, trade-offs, and clear recommendations upfront were exponentially more likely to be cited.

For example, a comparison page published by Venture Harbour pitting Workable against Greenhouse quickly outlines target personas, evaluation criteria, and structural trade-offs in its opening paragraphs. That single page surfaced in 13 AI answers. Conversely, Workable’s own corporate comparison page for the exact same matchup buried its insights under top-of-funnel product benefits and generic demo messaging, appearing in only 2 answers.

4. The Competitor Proxy Problem

When a vendor fails to document a specific capability or matchup, generative engines actively seek out proxies. If Company A lacks a dedicated landing page for its rivalry with Company B, LLMs routinely ingest competitor-authored content to describe Company A’s strengths and weaknesses. This creates an environment where go-to-market strategies are actively defined by rival organizations.

Official Responses and Market Shifts: The ChatGPT Anomaly

While platforms like Perplexity, Gemini, and Claude lean heavily on external listicles, independent publishers, and competitor sites, ChatGPT operates on an entirely different architectural playbook.

As the most widely utilized conversational AI platform in the United States, ChatGPT exhibits distinct citation behaviors. When researchers prompted ChatGPT to evaluate Lever versus Workable for a mid-market recruitment team, the engine bypassed independent publisher sites almost entirely.

Instead, ChatGPT drew 77.3% of its visible source presence directly from the vendors’ own product, pricing, and support documentation (with 63.8% originating solely from core product pages). Dedicated comparisons and listicles accounted for a negligible 3.3% combined.

This dichotomy highlights a critical strategic warning for B2B marketers: a unified SEO strategy will fail. Because multi-engine discovery ecosystems process information differently, brands must optimize their web architecture to satisfy both external semantic synthesizers (like Gemini and Perplexity) and direct product-documentation scrapers (like ChatGPT).

Furthermore, recent data from Comscore’s Q2 AI Intelligence Report emphasizes the necessity of a diversified approach. Market share for prompt volume is decentralizing rapidly:

• Google Gemini’s market share surged from 17% to 30% over the past year.
• Anthropic Claude’s market share jumped from 2% to 11%.
• ChatGPT’s market share compressed from 70% down to 50%.

Because these platforms rely on vastly divergent training sets and retrieval-augmented generation (RAG) pipelines, B2B brands can no longer monitor a single search engine to gauge their digital footprint.

Strategic Implications: How B2B Marketers Must Adapt

The era of relying purely on branded search engine optimization (SEO) and gated product brochures is officially over. As conversational search replaces traditional link-and-blue-results pages, go-to-market teams must undergo a fundamental restructuring of their digital asset management.

1. Reclaiming the "Versus Layer"

Software companies must aggressively audit their digital ecosystems to identify every existing comparison page—both internal and external. By utilizing specialized auditing frameworks (such as the Foundation Comparison-Page Audit Template), marketing teams can score matchups based on buyer demand, ownership gaps, and AI source leakage.

2. Rewriting Content Architecture for LLM Parsing

Content creation teams must pivot away from fluffy, top-of-funnel marketing copy on comparison pages. To win AI citations, pages must feature:

• Clear buyer-fit parameters explicitly detailed in the introduction.
• Objective, transparent trade-offs rather than one-sided superlative claims.
• Structured data formatting that allows LLMs to easily extract comparative metrics (e.g., reporting capabilities, pricing tiers, and implementation timelines).

3. Fortifying First-Party Documentation

Given ChatGPT’s heavy reliance on core product and support assets, technical documentation, API guides, help center articles, and pricing transparency pages must be kept scrupulously up to date. If a help article is outdated, the AI will synthesize an inaccurate portrayal of product functionality to prospective enterprise buyers.

Conclusion

As generative AI continues to mediate the relationship between software buyers and vendors, passivity is no longer an option. The data is unambiguous: if a company fails to author its own competitive narrative, third-party publishers, review aggregators, and aggressive competitors will author it instead. In the modern AI-driven marketplace, visibility belongs entirely to those who supply the evidence.