AI Assistants Now Beat Search as the Starting Point for Purchase Research
AI assistants beating search as the starting point for purchase research is true and false at the same time, depending on which number gets pulled. McKinsey's November 2025 research found that a plurality of users now rank AI search as their top digital source for buying decisions, ahead of traditional search engines, retailer sites, and review platforms. A Razorfish study says something close to the opposite: a plurality of consumers who recently made a major purchase actually started that journey with a search engine, and only a small slice opened an AI tool first. Google's own Q1 2026 earnings back up the second story: search revenue climbed year-on-year and query volume hit an all-time high, so traditional search is growing right alongside AI adoption rather than getting eaten by it.
There is no contradiction here to explain away. Each survey captures a distinct phase of one sequence, contrasting stated desires with initial actions. Recognizing this difference is central to the essay's thesis.
How preference and behavior diverge
Many people claim they want to begin their research with AI. Habit still puts Google first in line for the click. Therein lies the core tension, and it points less to a statistical anomaly than to a signpost for where behavior is going. McKinsey and Razorfish are not arguing over one fact; each tracks something different, aspiration versus habit-driven clicks, and both findings hold up.
A Bain survey of US consumers muddies this a bit further, finding that a large share of online buyers either mostly start in an AI tool or split time between AI and traditional search. Bain's count folds in people who treat AI and Google as interchangeable, which inflates how big the shift looks. Razorfish's approach, isolating just the first touchpoint, gives the cleaner read on actual behavior.
Ridge Marketing's analysis, pulling from Deloitte, Gartner, AP-NORC, and Pew Research, lands on the same duality: a growing share of consumers now start searches with AI instead of Google, a sharp jump from two years earlier, while Google still handles the overwhelming majority of global search traffic. Both of those facts hold at once.
Consumer mismatches between stated desires and actual habits offer no excuse for dismissing how AI shapes buying decisions. This proves the transition remains underway rather than complete, making the present moment ideal for companies to secure AI prominence ahead of habits aligning with intent.
What AI assistants do before Google gets involved
AI tools and conventional search aren't competitors locked in a battle for one role. These two data sets aren't in conflict; each captures a distinct point in time: stated intent versus observed first action. A Ridge Marketing analysis from May 2026, drawing on Superlines data, found that most sessions on AI search conclude with no click on any website. Readers take in the response, find what they came for, and carry on. In other words, AI is where buyers actually form their impressions of a brand, not where the deal gets done.
Once the split has a name, it is straightforward. AI handles the early question of "what should I even be considering, and who's credible here?" while search engines take over for "who do I actually buy from, and at what price?", so AI owns the exploratory work of building the shortlist. Ridge Marketing's analysis cites a Search Engine Land study showing that, in the research phase, most AI users find AI's responses more useful and easier to understand than standard search results. Users rely on it precisely for that role.
Presented by Chaire TREND(S) from the Université de Toulouse at Fevad's Grande Conférence held in Paris, research on consumers in France and the US showed AI search tools typically spawn further queries instead of steering someone toward a final decision. This shows that artificial intelligence prolongs the information-gathering stage rather than accelerating it. Razorfish research reinforces the point through another lens, revealing that very few shoppers beginning with AI trusted its suggestions enough to purchase.
The buyer may not act the moment an opinion forms, and in that pause the AI response still has influence. If that AI response presents a brand as visible, well positioned, and trustworthy, the buyer is already carrying it forward before search begins. When the answer leaves a brand out, it enters the next, busier stage of evaluation with no place secured and must earn attention from scratch.
A|Brand perception in AI answers is invisible to standard analytics
That early narrowing of options never appears in the dashboards analytics teams typically use. Last-click models record the final Google query or store visit and assign the win to the last channel touched, while the choice may have been steered beforehand by an untracked AI response. Google Analytics offers no visibility into ChatGPT’s brand description, Perplexity’s possible competitor pick during a head-to-head comparison, or Claude’s omission of a brand from its response. By then, that influence has already helped determine what the buyer chooses. Most teams’ existing tools never capture that evidence.
The blind spot is deepened by training-data limits. A company arriving or rebranding past the knowledge boundary is wholly absent from the system, while its internal dashboards reveal no problem, since lacking AI references produce no traceable metric within conventional analytics. Nothing fails, nothing drops; the company simply remains invisible.
Several 2026 B2B analyses show AI-referred visits converting better than organic when they arrive, with the premium highest on deliberative, research-intensive buying; those sessions carry outsized value, so the unseen upstream influence matters all the more. TechCrunch reported Similarweb data showing AI platform referral visits exceeded one billion monthly by mid-2025, a sharp rise on the year before. Leaving such a sizable channel unmeasured leaves a genuine gap in brand strategy, far from a trivial oversight.
Categories and buyers already inside the AI research layer
The urgency varies by category, so treating it as universal overstates the case. Stated preferences most closely match behavior in rational, utilitarian categories, while purchases driven by hopes, style, or longing show the largest divide. In November 2025, McKinsey found that many shoppers already use AI-based search when deciding what to buy in sectors such as gadgets, food, trips, health, clothing, cosmetics, and banking.
Fashion is split. Strategy&’s 2026 Fashion Retail Outlook says AI agents are appearing earliest in footwear, sportswear items, jeans, coats, and basic pieces, since shoppers in those areas tend to decide on price and practical value. But only some consumers believe a shopping aide with agency can capture their personal style.
Luxury falls between those poles, and it resists a clean read either way. The BCG and Altagamma True-Luxury Global Consumer Insights research, polling luxury shoppers in multiple markets, shows most already turn to AI when researching, comparing, and assessing luxury goods and experiences. Confidence at the point of purchase still trails, since the desire and aspiration behind a luxury buy resists algorithmic delivery in ways a simple spec-by-spec matchup does not.
The starkest finding in the entire dataset is how sharply age groups diverge. According to Yahoo and YouGov figures highlighted by Ridge Marketing, nearly all Gen Z adults have tried AI chatbots and many rely on them several times daily for search. Younger buyers who will drive spending for decades now treat conversational tools as their primary way to investigate products. Sectors currently dependent on purely logical purchasing decisions are witnessing the future that awaits all markets.
How AI models decide which brands to name
Search platforms order web pages. AI models instead piece together the most likely shared view, producing responses probabilistically and giving weight to signals that recur with authority throughout their data. Brands gain an advantage when many credible sources reinforce them, rather than relying on a single polished landing page with no broader support.
Aggarwal and colleagues authored the foundational study, a KDD 2024 paper whose authors spanned IIT Delhi, Georgia Tech, Princeton, and Allen Institute for AI. They found that sources gain real visibility when generative engines produce answers, provided the material includes figures, expert quotations, and cited references. That mechanism operates quite differently from keyword stuffing or tallying backlinks.
The signals also don't transfer evenly across platforms, which is where this gets genuinely tricky for anyone trying to plan around it. ChatGPT rewards consistent mentions in high-authority publications, something researchers call institutional echo. Gemini rewards accurate, real-time product data, or data integrity. Claude rewards substantive technical material like whitepapers, or technical depth. Optimizing for one of those doesn't carry over to the others automatically.
Which engine is used can redefine a brand's apparent visibility. Rocketblue's tracked prompt set showed Claude calling out brands far more frequently, Google AI Overviews doing so much less often, and ChatGPT falling between them. The same query can make a brand look prominent in one system yet absent from another.
Tailoring responses to individual users introduces yet another source of fluctuation. When a 2026 analysis tested ChatGPT alongside Gemini, Claude, plus Perplexity using diverse personas, identical prompts yielded brand shortlists diverging by 16.0 percentage points based on past conversations. Rather than chasing a single static metric, AI brand visibility spans an array of platforms, models, and audiences.
After examining tens of thousands of answers from AI systems, Peec AI found that how a prompt is worded rarely shifts which brands ChatGPT, Gemini, Perplexity, Google AI Mode, and Google AI Overviews recommend. What really moves the needle is a brand people consistently trust and see as the answer to their specific need, rather than any slick wording in the prompt itself.
A|Sentiment and mention frequency data as indicators of brand risk
Across the AI research layer, most brands face a larger danger: being left out entirely. In rocketblue's analysis of AI responses that named brands, mentions mostly came across as neutral, with a meaningful positive share and only rare negative framing.
This shifts what matters most. The real goal is simply being named at all, since an AI response that omits a company entirely leaves no room for positive or negative spin. Such a brand simply never enters the discussion.
Pessimistic framing rarely appears, yet its structural origins make it especially harmful when it does. Because RAG pipelines surface the most visible external commentary on a company, hostile Trustpilot threads or abrasive Reddit posts can dictate the precise tone of the generated reply. Consequently, managing reviews shapes AI narratives about a company rather than merely boosting conversions.
Presence and sentiment aren't contradicting measures; they capture different moments, the stated intent versus the observed first action. One brand may be mentioned all the time with framing that stays neutral or positive, while another barely surfaces and reads negatively on the rare occasions it appears. The two situations call for different fixes, and a single check-in catches neither.
Why one-off manual prompting is not a monitoring strategy
A one-time ChatGPT check on a brand name may feel like diligence, but it is not. AI outputs change with the phrasing of the prompt, prior chat context, and the model version in use, which makes one manual query on one day only one sample. It cannot tell whether brand visibility truly changed or the answer simply varied as usual.
Shifts in how a brand appears to AI happen whenever training sets expire or retrieval layers refresh, yet web analytics and SEO dashboards show nothing. Detecting such shifts requires repeatedly querying identical prompts at regular intervals across models. A brand's share of voice, meaning how often it appears versus rivals in pertinent AI answers, offers a truly valuable metric. Such a figure materializes only when identical queries run on a schedule through several models, tallying references, source attributions, and tone uniformly each time. Checking by hand yields neither the baseline total nor any pattern over time.
Profound noted that Ramp’s AI visibility surged within one month, a shift that would have stayed hidden absent continuous monitoring spanning engines and prompts throughout. Catching it does not take a complicated setup. It only takes measuring AI visibility on a schedule instead of checking it on a whim.
What systematic AI visibility monitoring looks like in practice
A working monitoring setup executes an organized prompt sequence on multiple AI systems simultaneously, evaluating the output to track brand references, citation habits, sentiment, and voice share longitudinally. Such repetition, not one fortunate or unfortunate search, yields the longitudinal patterns that hand-typed queries never could.
This commercial segment emerged during 2026, and a set of named vendors today address separate pieces of the work. After pulling in $58.5 million backed by Sequoia, NVIDIA, Kleiner Perkins, and Khosla Ventures, Profound earned G2 Winter 2026 leader status and says clients such as Ramp have logged major AI visibility gains within weeks. SE Ranking constructs historical comparisons spanning ChatGPT, AI Mode, and Google AI Overviews. Ryze AI bills itself as a hands-off platform that keeps watch on share of voice day and night. Otterly devotes itself entirely to tracking how brands show up in AI. By mid-2026 Writesonic had recast itself as an "AI Search Growth Engine," monitoring brand visibility on ChatGPT, Claude, Gemini, Grok, Perplexity, Microsoft Copilot, and Google AI Overviews.
Lettertrace offers another way to tackle the problem, with an open-source, self-hostable design, bring-your-own-API-key setup, and no cost from start to finish. It monitors how often a brand appears, its share of voice, audience sentiment, and competitor comparisons, while avoiding the ongoing fees associated with commercial platforms. Teams that need to steer their prompt library and inspect the underlying raw data get something materially different from the earlier dashboard tools.
This category has emerged now that the need it addresses has moved beyond theory. People now compare options inside AI responses, form brand impressions from them, and can see that pre-search experience only through monitoring tools, whether bought or run in-house,.