I’ve been diving into the world of ChatGPT lately, and it turns out there’s a fascinating bias toward commercial intent in its fan-out analysis. Let me break down what this means for our content strategy.
Over the course of my tests with 90 ChatGPT prompts, I discovered that commercial prompts lead to web searches a whopping 78.3% of the time, while informational prompts only did so 3.1% of the time.
This discrepancy got me thinking about how to craft our content to increase the chances of being featured in ChatGPT responses.
ChatGPT doesn’t source every response from the same place. Some responses are derived from its training data, while others are based on live web searches. This process, known as query fan-out, involves expanding a prompt into several background searches, and then curating a synthesized response from multiple topics. If our pages don’t fit into these subtopics, we miss out on getting pulled in.
So, our challenge now is more than just ranking well; it’s about ensuring our pages open the door to this fan-out process from the get-go.
In our samples, informational pages fell short. I encourage you to continue reading to uncover the paths the system actually favored.
I conducted tests across three industries: beauty, legaltech/regtech, and IT. The analysis explored prompt intents, the resulting query expansions, and the intents portrayed by those expansions.
The main takeaway is that most queries are aligned with commercial, rather than purely informational, intent.
Why This Question Matters Now and the Role of Query Fan-Outs
Understanding query fan-outs is crucial because it shifts the content creation approach. The system expands a prompt into several background searches, retrieving and synthesizing information from these subtopics.
This behavior triggers parallel web searches connected to the initial prompt, providing opportunities for mentions and citations.
Multi-query expansion is a fundamental design element in today’s search systems. As Google describes AI Mode, it breaks a question into subtopics, searches them simultaneously across sources, and combines the results into one coherent response.
This raises a key strategic question: should we focus more on top-of-funnel educational content or on lower-funnel comparison, shortlist, and recommendation content?
I designed this experiment to address that problem.
We aimed to see where fan-out occurs by intent category across selected industries: informational, commercial, transactional, or branded.
The hypothesis was clear: while informational prompts wouldn’t trigger fan-out, commercial ones would, and would either remain at the same level or move further down the funnel.
ChatGPT fan-outs were observed to align predominantly with commercial intent.
Disclaimer: This analysis reflects observed prompt expansion behavior in ChatGPT. Although Google AI Mode is cited for context to illustrate multi-query expansion as a pattern, it is not evidence of ChatGPT’s architecture.
The Setup: What We Tested
The experiment sampled 90 prompts, focusing heavily on informational intent.
| Prompt intent | Prompts | Share of sample | Prompts with fan-out | Fan-out rate |
| Informational | 65 | 72.2% | 2 | 3.1% |
| Commercial | 23 | 25.6% | 18 | 78.3% |
| Branded | 1 | 1.1% | 0 | 0.0% |
| Transactional | 1 | 1.1% | 0 | 0.0% |
Our sample primarily featured informational prompts, with some commercial and very few branded and transactional ones.
The test was structured around three industries: beauty/personal care, legaltech/regtech, and IT/tech.
The Result: Commercial Prompts Dominated
The findings were clear and conclusive.
Of the 90 prompts, 20 triggered a fan-out. Remarkably, 18 were of commercial intent and only 2 were informational.
Informational prompts accounted for a mere 10% of fan-out triggers (2 out of 20). When they triggered expansion, they were reframed into more evaluative, solution-seeking subqueries.
This indicates that, in this sample, commercial intent overwhelmingly influenced fan-out.
These 20 prompts resulted in 42 fan-out queries, averaging 2.1 per triggered prompt.
Here’s a breakdown of those fan-out queries:
- 39 were commercial.
- 2 were branded.
- 1 was informational.
Even if a prompt led to expansion, it typically transformed into a focus on comparison, feature filtering, shortlist creation, or brand-specific exploration, not broad educational discovery.
Methodology: Our Analytical Approach
Our experiment involved 90 prompts across three industries, mainly informational with fewer commercial prompts, and minimal branded and transactional queries.
The analysis involved:
- Choosing a representative set of prompts.
- Identifying fan-outs.
- Classifying each fan-out by intent.
- Analyzing distribution by prompt metadata.
Our approach followed three key steps:
- Classifying prompts by intent labels.
- Counting prompts that triggered any fan-out.
- Reviewing expansion queries and their intent labels.
This process revealed two distinct perspectives:
- A prompt-level view to determine which prompts instigated fan-out.
- A fan-out-query view to assess the intent of downstream expansions.
This distinction is important: the first identifies prompts that initiate the fan-out path, while the second examines where the system proceeds once engaged.
Interpreting the Results: Fan-Outs Trend Down-Funnel
The clearest takeaway is that, in this instance, fan-outs behave more like decision support rather than topic exploration.
Commercial prompts frequently opened new discovery paths.
Once open, these paths typically remained commercially focused.
The system often expanded into comparisons, feature-based analyses, product listings, and pricing inquiries.
Here are some illustrative examples:
- “Suggest the best accounting software for small business and explain why” expanded to a commercial comparison query on features.
- “What are the top AI document management systems for lawyers?” led to multiple product-centered legaltech queries.
- “What are the best products for skin care?” grew into a shortlist-style inquiry around product categories and reviews.
The rare informational examples expose more about the system’s tendencies than the rules themselves.
- “I need an open-source document management system. What can you suggest?” initially coded informational, shifted to solution recommendations.
- “AI tools for legal research and document automation” also redirected into clearly commercial/evaluative queries.
Ultimately, even broad prompts frequently translate into more focused, commercially driven retrieval paths.
Implications for Our Content Strategy
Let’s not abandon informational content; however, we should recognize that informational content alone doesn’t consistently align with fan-out expansions, at least in this dataset.
If our goal is to shine in AI responses tied to product selection or vendor discovery, we need to strengthen our coverage with content that lines up with these downstream commercial intents.
Consider the following:
- Creating “best-of” and shortlist pages
- Developing thorough comparison pages
- Writing “which tool should I choose” guides
- Feature-led category explainers
- Alternative option pages
- Evaluation-focused FAQs
- Incorporating recommendation passages in broader educational pieces
In practical terms, our content model should integrate both top- and bottom-of-funnel strategies, with strong commercial bridges.
A comprehensive piece can still be beneficial, provided it contains elements that the system can readily transform into decision-support inquiries.
An educational piece that lacks direct references to products, tradeoffs, features, use cases, or selection criteria is less likely to match the system’s fan-out paths.
In short, consider not only answering the obvious inquiries but also forecasting the subsequent evaluative step the system might generate behind the scenes.
Understanding Our Limitations
These results offer direction rather than universal truths.
- 90 prompts highlight a pattern, but don’t establish AI retrieval behavior as a law.
- The prompt mix skews heavily towards informational content, with few branded or transactional samples. The findings don’t signify absence.
- While diverse, the dataset isn’t normalized for brand, style, or use case. Some sectors lean easily into product-discovery language.
- This analysis observed recorded fan-outs rather than controlling for platform-level testing. It reflects what occurred within this set rather than guarantees of ChatGPT’s constant behavior.
- Google’s fan-out description provides context; however, this isn’t a Google AI Mode test. It’s ChatGPT-centric, with strategic—not architectural—takeaways.
Next Steps for Testing
Future versions of this test should further isolate the question while widening the dataset.
A follow-up should map fan-outs to specific content formats.
The aim isn’t solely to affirm that commercial intent triumphs, but to pinpoint which page templates and structures proficiently capture AI-preferred fan-out paths.
Inspired by this post on Search Engine Land.
