Purpose
For two decades, businesses optimized for search. Over time, social platforms became an additional layer for discovery and research. That model is now expanding.
Customers use AI systems such as ChatGPT, Claude, and AI Overviews to research, compare options, and think through decisions. What people actually ask LLMs — prompt-level data — is not publicly visible, and interactions are highly personal and conversational.
As a result, most observable signals about AI-driven customer behavior come from academic studies, institutional reports, and platform analyses. That evidence is fragmented, uses different methodologies, and is published across formats. For time-constrained marketers, separating signal from opinion becomes unnecessarily difficult.
This report synthesizes findings from fifteen curated, first-party research sources including Anthropic, Harvard Business Review, McKinsey, Measure Protocol, MIT, and NBER. The purpose is to consolidate credible evidence, remove interpretive noise, and give marketers reliable insights they can absorb quickly.
What this research can and cannot tell you
This synthesis shows how AI is shaping research and decision-making at a behavioral level. It does not show how your brand performs inside AI answers.
It cannot tell you:
- Whether your website is being crawled or indexed by AI bots
- How often your brand appears in AI-generated responses
- Whether your pages align with what users ask LLMs
- Which of your content formats are surfaced or ignored
Answering those questions requires direct observation of AI crawler activity, citation patterns, and bot-level interactions with your site. It requires brand-specific analysis — the kind Kachi runs on your domain.
The signals that matter
1. AI is becoming the default research tool
McKinsey research shows that 44% of AI search users now prefer AI-powered search as their primary information source, compared to 31% who prefer traditional search engines.
The study surveyed 1,927 U.S. consumers across age groups, including non-technical users. Methodology and sampling are disclosed and representative.
McKinsey further projects that up to $750 billion in U.S. revenue could be influenced by AI-powered search journeys by 2028. For marketers, this suggests a near-term behavioral shift rather than a long-term speculative one.
2. People use AI for every decision, big or small
Anthropic, Measure Protocol, and NBER analyzed real user interactions across large datasets — including 142,965+ AI conversations and longitudinal usage trends since the launch of ChatGPT. Three patterns are consistent across studies.
A. Personal use dominates, and it shapes purchase journeys
- Personal usage grew from 53% in mid-2024 to approximately 71% by mid-2025 in ChatGPT alone.
- The top categories of usage account for roughly 80% of interactions: practical guidance (28%), information seeking (27%), writing and assistance (25%).
This matters because purchase journeys often begin in personal, low-stakes contexts. Product discovery is embedded inside everyday questions rather than explicit "buy" searches.
B. Commercial intent is measurable and meaningful
Semantic classification of 142,965+ AI conversations shows:
- 21.6% include commercial intent
- 14.4% reflect research-phase intent
- 7.1% indicate high commercial intent (comparisons, evaluations, purchase signals)
High-intent prompts often include phrases such as "best", "compare", "alternatives", or "worth it". These are late-funnel moments that historically occurred on search engines or review sites. A significant portion of high-intent conversations occur in categories like health, wellness, consumer services, and technology — categories where users seek guidance before making a decision.
C. The verification loop between AI and traditional search
Research from Measure Protocol and NBER identifies a consistent behavioral loop:
- 26.4% of users use AI and traditional search together
- 41% of searches occur within five minutes of an AI interaction
- 32% of all searches are directly related to recent AI conversations
AI search is not fully replacing traditional search. It is reshaping the sequence of discovery and validation.
3. Brand websites are underrepresented in AI answer sources
One of the most overlooked findings is the distribution of sources used in AI-generated answers. McKinsey and related research suggest that brand-owned websites typically represent only 5% to 10% of the sources surfaced in AI responses. The majority of citations come from:
- Publishers and editorial content
- User-generated content
- Forums and reviews
- Affiliates and third-party aggregators
Two implications. First, strong website SEO alone does not guarantee visibility in AI-generated answers. Second, third-party narratives about a brand may influence AI recommendations more than the brand's own messaging. In several categories, market leaders were inconsistently represented in AI responses despite strong traditional search presence.
4. Most brands aren't tracking AI search yet
Despite rapid platform adoption, measurement maturity remains low. McKinsey's CMO research indicates that only 16% of brands systematically track performance in AI search. At the same time, AI-generated summaries already appear in roughly half of Google searches and are expected to increase further.
This creates a structural lag: consumer behavior is shifting faster than marketing analytics frameworks.
5. Early AI visibility creates durable advantage
The data does not suggest immediate displacement of traditional search. It suggests coexistence and gradual reallocation of attention.
OpenRouter's research on early adoption patterns shows a retention advantage for early-visible sources in AI systems. Content that becomes part of training and citation loops tends to persist in responses over time.
About the research base
This report synthesizes 15 primary studies from Anthropic, Harvard Business Review, McKinsey, Measure Protocol, MIT, NBER, OpenRouter, and related research institutions. Combined research inputs include:
- 142,965+ AI conversations analyzed
- Longitudinal usage studies from 2022 to 2025
- Large-scale consumer surveys with disclosed methodologies
- Cross-platform behavioral datasets
Findings were included only where multiple independent sources showed directional alignment.