If you have spent any time navigating the discourse within SEO and AEO (Answer Engine Optimization) Slack channels over the past twelve months, you have likely encountered a pervasive piece of advice: "Be on Reddit. Be on YouTube. Go where the community already trusts."
This recommendation is grounded in the reality of the early-stage buyer’s journey. When a prospective customer is still struggling to define a category or diagnose a problem, they look for human consensus. However, a new, granular analysis suggests that this advice loses its potency the moment a buyer transitions from "learning" to "evaluating." When the search shifts from "What is a CRM?" to "Pipedrive vs. HubSpot for sales-led companies," the reliance on community forums like Reddit drops precipitously.
In a recent study conducted by the content agency Ten Speed, researchers analyzed how modern AI models—ChatGPT, Perplexity, Claude, and Gemini—source information when users are in the late-stage evaluation phase. The findings offer a sobering correction to the "community-first" narrative, revealing that when the checkbook is about to open, AI models overwhelmingly prefer brand-controlled, proprietary assets over third-party discussion threads.
The Methodology: What Was Actually Tracked
The study, authored by Ten Speed’s Nelson Brassell, sought to bypass anecdotal evidence by focusing on citation data. Using Peec AI, a specialized tool that monitors how LLMs cite URLs in their responses, the team constructed a test set of 170 prompts. These prompts were designed to mimic the specific, high-intent queries of a B2B buyer—questions focused on specific vendor comparisons, technical integrations, and compliance standards like SOC 2 reporting.
To ensure the data reflected a broad market rather than just the niche of the agency’s client list, the prompts covered diverse sectors including fintech, physical security, hospitality, and IT automation. By pulling data across 7,387 individual citation appearances, the study provides a high-resolution snapshot of how AI models prioritize information when a buyer is actively building a vendor shortlist.
The Data Breakdown: Where AI Finds Its Answers
The results provide a clear hierarchy of content value. Contrary to the popular belief that AI favors "neutral" community discussions, the data demonstrates a strong preference for authoritative, brand-owned domains.
The Hierarchy of Citations
- Product Pages: 24.1% (The dominant source)
- Articles (Blogs, News, PR): 17.4%
- Comparison Pages: ~13%
- Listicles: ~13%
- How-to Guides: Just under 9%
- Homepages: 7.8%
- Directory Profiles (G2, Capterra): 7.2%
- Community (Reddit, YouTube, Forums): 4.2%
When aggregated, brand-controllable content—assets that marketing teams write, own, and maintain—accounted for a staggering 88.3% of all citations provided by the AI models. Most notably, Reddit and YouTube, despite their perceived dominance in the current SEO zeitgeist, accounted for a combined 4.2% of citations, with YouTube contributing barely 1%.
The standout performer was "Comparison Content." While comparison-based prompts made up only 20% of the total prompt set, they generated 27% of the citations. This represents a 1.33x return on investment, suggesting that many B2B marketing teams are significantly underutilizing comparison assets, treating them as defensive "battlecards" rather than primary drivers of visibility.
The "Six Questions" Audit: Testing the Credibility
In journalism, data is only as strong as its verification. Upon reviewing the initial findings, it became clear that several discrepancies required clarification. A series of six pointed questions were posed to Ten Speed to stress-test the validity of the report.
Chronology of the Audit
- The Denominator Discrepancy: The initial report contained a contradiction where one chart cited 220 total prompts while another used 170. Ten Speed confirmed 170 was the correct figure and committed to updating the visual, noting it was a labeling error rather than a data-integrity issue.
- Category Tagging: There was inconsistency in how "comparison" content was categorized across different charts. After inquiry, the team reconciled the data, confirming the 34-prompt count that supported the study’s headline.
- Sample Size Transparency: When asked about the number of distinct client brands and verticals, Ten Speed declined to provide a specific count to protect the confidentiality of their clients. While this prevents external researchers from judging the breadth of the sample, it was accepted as a reasonable boundary for a private agency study.
- Platform Disaggregation: The study did not separate data by specific AI model (e.g., how ChatGPT compares to Perplexity). Ten Speed openly admitted this was a data gap, acknowledging that models behave differently at the bottom of the funnel.
- Statistical Rigor: When asked if the 88/4 split—the cornerstone of the study—was statistically tested, the answer was "no." It was presented as a descriptive pattern. This distinction is vital; it means the study is a snapshot of specific behavior, not a universal law of AI search.
- Actionable Intent: Finally, the team confirmed they lacked data linking these citations to actual pipeline metrics like demo requests or closed deals.
This transparency regarding the study’s limitations is, ironically, what gives the data its credibility. Rather than presenting the report as an immutable industry benchmark, Ten Speed maintained a degree of professional restraint, admitting that their findings are descriptive, time-bound, and specific to the B2B SaaS landscape.
Implications for Future Content Strategy
For marketers, these findings do not invalidate the importance of community building; rather, they demand a more strategic allocation of resources based on the funnel stage. If the objective is to capture the buyer at the point of decision, the content strategy must pivot accordingly.
1. Optimize for the "Cold Read"
Product pages and homepages are the first point of contact for an AI model interpreting a vendor’s value proposition. These pages must be written with extreme clarity. Clever, metaphorical, or "creative" copy often confuses LLMs. Instead, focus on plain language that answers three questions: What does this do? Who is it for? What does it integrate with?
2. Move Beyond the "Battlecard"
Comparison content is currently yielding a 1.33x return on visibility. Marketers should move beyond creating "Competitor A vs. Us" pages and begin building out broader, more comprehensive comparison assets. If the AI models are disproportionately citing comparison content, you must ensure your brand is present in the comparison hubs that the models are most likely to crawl.
3. Treat Directories as Structural Content
At 7.2% of total citations, platforms like G2 and Capterra serve as foundational data for AI. These should not be treated as mere review inboxes. Ensure that your company’s category tags, feature lists, and integration data are updated with the same rigor you would apply to your own website. AI models rely on this structured data to build shortlists when users ask for "options in X category."
4. Apply Skepticism to AI Research
Perhaps the most important takeaway is the practice of verification. Every AI visibility statistic shared in a boardroom should be interrogated. What is the denominator? Is this an average across platforms? Is the finding descriptive or statistically significant?
Conclusion
The era of AI search is still in its infancy, and the "rules" of visibility are changing faster than the tools themselves. This study serves as a crucial reminder that while the SEO community often chases the latest trend—be it community forums or social signals—the fundamental logic of search remains anchored in the availability of clear, authoritative information.
A study that survives a rigorous audit and admits to its own gaps is far more valuable than one that promises a "silver bullet." By focusing on high-intent, brand-owned assets and maintaining a healthy skepticism toward industry benchmarks, B2B marketers can ensure their brand remains a primary source of truth for the AI models that are increasingly guiding the modern buyer.
