You're Being Told Your Buyers Are Researching Through AI. Should You Believe It?

You’re being told that your customers research through AI before they ever contact you. That they compare suppliers and form shortlists in conversations you can’t see, and that you may be losing business without knowing it.

Ask how much of this is actually happening, and the answers get vaguer. Your analytics show a trickle of visits from AI platforms. A vendor shows you a set of prompts where competitors appear and you don’t. Neither tells you what your buyers did last quarter.

You are being asked to respond to a change in buying behavior without the measurement you would normally expect before spending money. So what should you believe? Enough to act on, and no more than the evidence supports.

“Researching Through AI” Is Several Claims

The claim usually arrives as one sentence. It contains at least six propositions: buyers use AI to answer questions about products and services; they use it to find suppliers; they use it to compare suppliers; they use it to narrow a shortlist; they make the decision mostly inside AI; and businesses are losing opportunities as a result without realizing it.

Each step is a larger claim than the one before, and evidence for an early step doesn’t carry the later ones. A survey showing that buyers ask AI questions says nothing yet about who made their shortlist, let alone about your revenue.

“Entirely” deserves particular suspicion. Research for a considered purchase usually runs across search, supplier websites, colleagues, review sites, industry publications and conversations with salespeople. AI can be one of those places, or several of them. The evidence describes a mix, not a migration.

What the Research Supports

The adoption evidence is solid. Pew Research Center’s nationally representative survey of 5,119 U.S. adults, conducted in February 2026, found that 44% use AI chatbots, with searching for information the most common use.

Influence on purchasing is where the evidence thins. In Gartner’s consumer surveys published in May 2026, willingness to let AI make a purchase decision topped out at 11% across the categories tested, and 54% of those who had used AI while shopping said they had to double-check all the information it provided. On the business side, Gartner found 45% of 645 B2B buyers had used generative AI during a recent purchase, while 69% wanted to validate what AI told them with a salesperson.

6sense’s 2026 Buyer Experience Report combines three surveys of buyers involved in purchases worth at least $25,000 a year: completed purchases (3,596 journeys), evaluations that ended without a purchase (1,011), and a separate survey on AI. In the AI survey, 81% of buyers used a large language model somewhere in the journey, mostly in the middle stages to sort, summarize and validate what they already knew. Only 26% used one to find new vendors, and 25.2% said LLMs significantly influenced their perception of vendors. The report’s journey data asks a different question: when buyers rated what shaped their vendor preference before contacting a seller, AI tools sat mid-table at 21.2%, behind existing knowledge (34.4%) and internal discussions (31.8%).

The report also provides evidence about how much of the buying process occurs before seller contact. In the completed purchases it studied, buyers first contacted a seller, on average, 59% of the way through a 10.3-month cycle, and the vendor they favored at that point went on to win 81.5% of the time. Buyers do decide a great deal before you hear from them. In this sample, AI was one influence on that private stage, not the main one.

Other studies find more. G2’s 2026 survey of 1,076 B2B software decision makers, conducted in March, found 51% now start software research with an AI chatbot more often than with Google. Its strongest findings concern selection itself. Asked whether an AI chatbot influenced the vendor they ultimately chose, 33% said it led them to a vendor they hadn’t considered and 36% to one they were already familiar with, which G2 presents together as 69% choosing a different vendor than expected. Another 22% said it reinforced the vendor they expected, and 7% said it had no influence. Forrester’s Buyers’ Journey Survey, 2025, discussed in January, found 94% of business buyers used AI in their buying process and that twice as many named generative AI or conversational search as a more meaningful or important source of information than named any other source.

Those findings deserve to be taken seriously: they go beyond adoption to influence on who won, as buyers themselves attribute it, which is not independent evidence of cause. They also describe particular populations. G2 surveyed software buyers, where AI tools are adopted early and where G2, as a review platform, has a direct interest in how vendors are found. TrustRadius’s survey of 1,862 technology buyers found AI tools still outside the top five resources buyers used most. 6sense’s surveys span software, services and physical goods, but it too sells to the revenue teams it is advising, and its authors caution that the results don’t establish that every market follows the same pattern.

Part of the divergence is plausibly in the questions. Using AI somewhere, naming it a meaningful source, rating it an influence on preference and saying it changed the final choice are different things to ask, of different buyers in different categories.

All of these measures are self-reported, and none observes a purchase being decided. They establish that AI is a real part of B2B research and that some buyers credit it with changing their choice, not how large that effect is in your category.

What Reaches You, and What Doesn’t

Imagine a finance director replacing a payroll system. She asks an AI assistant to explain the options, assess providers against her requirements and narrow them to two. Then she searches for one provider by name and books a demo.

That provider sees a branded search and an inquiry. The others she eliminated see nothing. That part of the problem is familiar. The less familiar part is how little the fragments that do reach a business reveal.

Five things are easy to run together: the buyer’s private conversation, the query a platform reports, the answer the system generated, the sources it retrieved or cited, and what the buyer did next. Seeing one doesn’t show you the others.

Platform reporting makes this concrete. Search Console’s standard performance report lists queries, and Google says that when a user asks a follow-up question in AI Mode, “they are essentially performing a new query”, with the resulting impressions and clicks counted against it. A query a publisher sees may therefore be one turn of a longer conversation. That doesn’t mean every turn appears, and nothing in the report links one turn to the next. Its separate generative AI report shows impressions in AI Overviews and AI Mode by page, country, device and date, not what was asked or answered.

Bing Webmaster Tools reports grounding queries, which Microsoft describes as the key phrases “the AI used when retrieving content”, drawn from a sample. Those are the system’s search terms, not the buyer’s words.

So the last query a business sees is not necessarily the buyer’s last question. Some queries carry a great deal of context, because the buyer typed it into the question. What a business can’t see is the rest of the conversation and whatever else the system drew on to answer it.

What Monitoring Can and Can’t Establish

AI visibility monitoring puts a defined set of prompts to AI systems and records whether a business is mentioned, cited, compared or left out. That is useful: it finds factual errors, missing evidence, inconsistent descriptions and competitors who are represented more clearly.

These are controlled observations. The prompts are chosen by whoever runs the test, and neither the business nor the monitoring provider can see how often real buyers ask them. Answers also vary from run to run: SparkToro found less than a 1 in 100 chance that ChatGPT or Google’s AI would give the same list of brands in any two of 100 identical requests. Monitoring can tell you how you would fare if buyers asked questions like these. Its value depends on how realistic the questions are, and a visibility score is not a measure of market share, preference or revenue.

Website analytics have the opposite limitation. They identify some visits from AI platforms, but a buyer who copies a name, searches for it later or types the address leaves no AI trace, and Google includes AI Mode and AI Overviews clicks within its ordinary search traffic. An AI referral doesn’t prove the platform caused a sale, either.

Customer evidence fills some of the gap. Lead-source questions, sales conversations, win-loss interviews and CRM notes can reveal whether buyers arrive with comparisons formed elsewhere, though they suffer from recall and from what buyers think sellers want to hear.

The useful distinction is between observing an output, observing a behavior and demonstrating a commercial consequence. Most measurement reaches the first or second. Few businesses can reach the third with confidence, and a citation is not the business outcome.

Two Confident Conclusions From Thin Evidence

The skeptical executive says: our analytics show almost no traffic from AI, so AI isn’t influencing our customers. That doesn’t follow. Influence on a later branded search, direct visit or referral leaves no AI label.

The vendor says: you don’t appear in these AI answers, so you are losing customers. That doesn’t follow either. Absence from selected test prompts doesn’t show that real buyers asked those questions, saw those answers or would have bought.

Both observations are worth having. Each becomes a problem when it is presented as proving more than it does. Between a weakness and a lost sale sit three separate links: evidence of a possible representation problem, evidence that buyers encounter it, and evidence that it changes what they decide. A business can have good grounds for the first without being able to establish the other two.

What Evidence Would Justify Acting

Different evidence justifies different sizes of decision.

The first decision is whether the possibility warrants investigation. The research above provides good reason to ask that question, even if the answer differs from one market to another. Answering it costs a few customer conversations, some questions for the sales team and tests built around what a qualified buyer in your market would realistically ask.

The second decision is whether to fix what that work finds, and it needs evidence about your own representation rather than about the market. A factual error deserves attention, though its materiality and whether it recurs across systems and repeated tests may be worth checking first. An omission from a comparison where you clearly belong, or a description that favors a competitor on a point you can document, justifies targeted work without knowing how often the comparison is made. Corrections like these are usually modest and reversible.

The third decision is a substantial, continuing investment, and that needs stronger evidence specific to your market: customers who say AI shaped their shortlist, sales teams hearing comparisons they didn’t create, and representation gaps on the scenarios that carry the most revenue. Even then it won’t amount to revenue attribution, but it can be enough when exclusion from early consideration would be expensive and the pattern holds across several sources.

A reasonable executive could also conclude that none of this is a priority yet. That can be a sound decision, provided someone looked.

Better platform reporting won’t remove the underlying problem. Knowing that a brand appeared in an AI response, even with every impression counted, doesn’t establish whether it influenced a real buyer’s decision.

Belief Is Not a Measurement Strategy

There is credible evidence that buyers use AI during purchasing research. In 6sense’s research they used it more often to sort and validate than to discover, while G2’s software buyers credit it with changing their choice. How much that matters varies by market, and no business can fully see its influence on individual decisions. Neither point removes the need to look.

You don’t have to accept every claim about AI changing how buyers behave. You also shouldn’t dismiss the possibility because your analytics can’t show you the conversations. The useful question isn’t whether you believe what you’re being told. It’s what evidence would justify acting on it, and whether you have gathered any.