The rapid integration of generative artificial intelligence into the search landscape has fundamentally disrupted how consumers find information online. However, a groundbreaking joint study by SEO platform Semrush and prominent search strategist Kevin Indig reveals that the disruption is far more complex than previously understood.
According to the research, ChatGPT’s "high-reasoning" mode—which utilizes advanced chain-of-thought processing to tackle complex queries—behaves as an entirely different search engine compared to its standard, "minimal-reasoning" counterpart. The study found that when ChatGPT is allowed to "think" deeper, it cites completely different domains, executes nearly five times as many web searches, and drastically deprioritizes user-generated content (UGC) platforms like Reddit.
For digital marketers, brand managers, and SEO professionals, these findings signal a critical paradigm shift: visibility in basic AI search results does not guarantee presence when users ask complex, high-intent buying questions.
Main Facts: The Bifurcation of AI Search
The Semrush and Kevin Indig study analyzed how ChatGPT’s search behaviors diverge when shifting from a fast, minimal-reasoning mode to an analytical, high-reasoning mode. The data paints a picture of two distinct search surfaces operating under the same interface:
- Minimal Source Overlap: Only 25.6% of cited domains overlapped between the minimal and high-reasoning modes for the exact same prompts. In other words, nearly three-quarters of the sources cited by ChatGPT change depending on the reasoning mode utilized.
- Exponential Search Activity: High-reasoning mode executed 1,130 web searches across the test set, compared to just 245 searches performed by the minimal-reasoning mode—an increase of over 360%.
- Greater Citation Density: The overall citation rate rose from 50% in minimal-reasoning mode to 68% in high-reasoning mode. Furthermore, the average number of citations per response increased from 2.6 to 4.5.
- The Marginalization of Social Media and UGC: Forums and user-generated review sites saw their visibility cut in half. Reddit’s citation share plummeted from 15% in minimal-reasoning mode to just 7% in high-reasoning mode. General UGC and review platforms dropped from 14.3% to 6%.
- Comparison Queries Drive Massive "Fan-Out" Searches: When faced with comparison prompts, the high-reasoning model executed an average of 24 sub-queries (fan-out searches) per prompt, compared to only 5.5 for the minimal-reasoning model.
Chronology: The Evolution toward Agentic Search
To understand why this divergence is occurring, it is necessary to trace the rapid evolution of search technology over the last several years.
[Traditional Search] ──> [Semantic & RAG Search] ──> [Minimal-Reasoning AI] ──> [High-Reasoning Agentic Search]
(Blue Links) (Featured Snippets) (Quick LLM Answers) (Multi-Step Deep Reasoning)Phase 1: The Keyword Era (Pre-2020)
For over two decades, search engines relied primarily on index-based keyword matching, semantic analysis, and PageRank algorithms. Marketers optimized websites for specific keywords to win a spot on the coveted first page of Google.
Phase 2: Retrieval-Augmented Generation (2023)
With the launch of ChatGPT and Google’s Search Generative Experience (SGE), search engines began integrating Retrieval-Augmented Generation (RAG). In this phase, the AI reads the top search results and synthesizes a direct answer for the user, citing its sources via inline links. This is what the study characterizes as "minimal reasoning"—a fast, superficial synthesis of the first few search results.
Phase 3: The Dawn of Agentic, High-Reasoning Search (Late 2024–Present)
With the introduction of OpenAI’s o1 and o3-mini models, AI search entered the era of deep reasoning. Instead of executing a single search and summarizing the top three links, high-reasoning models break a prompt down into a series of logical sub-steps.
The AI acts like an autonomous research assistant: it searches the web, analyzes the results, identifies information gaps, executes secondary and tertiary searches to fill those gaps, and then cross-references its findings before generating a final, highly structured response.
Supporting Data: A Deep Dive into the Semrush Study
The empirical foundation of the study conducted by Semrush and Kevin Indig rests on a rigorous methodology designed to simulate real-world B2B and B2C buyer journeys.
Methodology
The researchers developed a test set of 100 highly targeted prompts mapped across 20 distinct buyer journeys. These journeys spanned four major commercial verticals:
- B2B SaaS
- Finance
- Consumer Technology
- Health and Lifestyle
Each prompt was run twice under identical conditions: once in ChatGPT’s minimal-reasoning mode and once in its high-reasoning mode. The researchers tracked citation rates, the identity of the cited sources, and the number of "fan-out" queries (sub-searches executed by the AI behind the scenes).
The "Fan-Out" Effect in Comparison Prompts
The difference between the two modes became most pronounced during the "comparison" stage of the buyer’s journey—when a user asks the AI to compare multiple products, services, or software solutions.
| Metric | Minimal-Reasoning Mode | High-Reasoning Mode | Increase (%) |
|---|---|---|---|
| Average Sub-Queries (Searches) per Prompt | 5.5 | 24.0 | +336% |
| Average Citations per Response | 5.8 | 9.8 | +69% |
In minimal-reasoning mode, the AI would typically search for a query like "best CRM software for startups" and summarize a couple of listicles.
In high-reasoning mode, the AI executed up to 24 distinct searches. It searched for individual product documentations, pricing pages, user reviews, and feature comparisons, culminating in a highly comprehensive matrix of 9.8 citations per response.
[User Prompt: Compare CRM A vs CRM B]
│
▼ (High-Reasoning Mode)
┌───────────────┼───────────────┐
▼ ▼ ▼
[Search Pricing] [Search Docs] [Search Reviews] ==> (24 Sub-Queries)
└───────────────┼───────────────┘
│
▼
[Synthesized Output with 9.8 Citations]The Collapse of User-Generated Content (UGC)
One of the most surprising findings of the study was the sharp decline in visibility for Reddit and other community-driven review sites when high reasoning was active.
Under minimal-reasoning conditions, ChatGPT frequently relied on Reddit (15% citation share) and other UGC platforms (14.3% citation share) to quickly pull user opinions. However, when the high-reasoning engine was engaged, Reddit’s citation share fell to 7%, and general UGC/review sites dropped to 6%.
This suggests that when the AI has the "time to think," it deprioritizes casual forum discussions in favor of primary sources, official documentation, authoritative industry analyses, and structured data.
Industry-Specific Citation Dynamics
The impact of high reasoning on brand visibility was not uniform across all sectors. The finance vertical, characterized by highly regulated and data-sensitive queries, experienced the most dramatic shift.
Because financial queries require high accuracy and verification of real-time regulatory compliance, ChatGPT’s high-reasoning mode aggressively expanded its search parameter in this sector. The AI favored government databases, established financial institutions, and recognized market analysis platforms over secondary blogs, resulting in the largest citation lift of any vertical tested.
Expert Perspectives and Context
The study’s co-author, Kevin Indig, has emphasized that these findings require a complete reassessment of how search marketers measure "AI visibility."
"Your content may appear in fast ChatGPT answers but disappear when users ask more complex questions," Indig noted. He explained that high-reasoning models do not merely read the web; they evaluate the credibility of information.
AI industry analysts point out that this behavior aligns perfectly with OpenAI’s stated goals for its reasoning models. By giving the LLM a dedicated "thinking" budget before it outputs a response, the model can cross-reference facts, spot contradictions in superficial blog posts, and seek out authoritative documentation. This systematic verification process naturally weeds out lower-quality affiliate blogs and forum hearsay, replacing them with authoritative, primary-source domains.
Implications: How Marketers Must Adapt to High-Reasoning Search
The bifurcation of AI search into low-reasoning and high-reasoning surfaces introduces a new layer of complexity to search engine optimization. To maintain brand visibility in an era dominated by agentic AI search, marketers must pivot their strategies in several key areas.
1. Optimize for the "Fan-Out" Search Path
Because high-reasoning AI conducts dozens of sub-queries to verify claims, brands must ensure that their product details are consistent across the entire web ecosystem. The AI will look at your official site, your API documentation, your GitHub repositories, your press releases, and third-party industry directories. If there are discrepancies in pricing, features, or technical specifications, the reasoning model may exclude your brand due to a lack of verifiable consensus.
2. Prioritize Authority and Original Research Over UGC
While optimizing for Reddit and Quora has been a popular "shortcut" for SEOs over the past two years, the Semrush study proves that this strategy fails when users ask high-value, deep-reasoning questions. Brands must reinvest in:
- In-depth, primary-source whitepapers.
- Comprehensive, technically accurate product documentation.
- Original research, proprietary data, and case studies.
3. Build a "Brand Graph" That AI Can Easily Parse
To be cited during a 24-query comparison search, your website must be highly crawlable and structured for machine consumption. Implement robust Schema markup, maintain clean XML sitemaps, and present product data in structured tables. The easier it is for an AI agent to extract facts from your page during a split-second sub-query, the more likely your brand will be cited in the final high-reasoning output.
4. Understand Your Visibility Across the Buyer’s Journey
Marketers can no longer rely on a single ranking report to measure search success. A brand must track its AI visibility across different query types.
A brand might dominate the "minimal-reasoning" informational queries (e.g., "What is B2B SaaS billing?") but be completely absent from "high-reasoning" transactional queries (e.g., "Compare the API latency and enterprise security features of Billing Platform X and Billing Platform Y"). Ensuring visibility at the high-reasoning stage is where actual conversions and revenue are generated.
Conclusion
The Semrush and Kevin Indig study serves as a wake-up call for the digital marketing industry. ChatGPT is no longer just a passive answering machine; it is evolving into an active, analytical researcher. As high-reasoning AI models become the default interface for consumers making complex purchasing decisions, survival in the search landscape will belong to brands that prioritize deep authority, technical clarity, and comprehensive digital footprints over quick, surface-level optimization.
