Beyond B2B vs B2C: Does Your AI Visibility Strategy Fit Your Business?

The two companies in this article are constructed for illustration. They are not clients, and nothing about them is measured.

Is B2B vs B2C the right first question?

The label identifies the type of customer in the transaction. It does not tell us who asks the question, who uses the product, who influences the choice or who can stop it, and it says nothing about what each of them asks. For visibility in AI answers, the second set of facts matters more, because an assistant answers the person who writes the question, and that person is often not the one who signs.

In our reading, the label also hides differences inside each camp. A small company buying accounting software and a multinational buying a production line are both "B2B", and they have little else in common. A shopper replacing a worn-out kettle and a parent choosing a cream for a child's eczema are both "B2C", and they ask very different things. Our working idea is that two additional questions describe the information demand around a purchase better than the label alone does. How many people take part in the decision? And how many different situations bring them to the category? They are the entry to the four layers described later in the article. All of it is our interpretation, and we mark the places where the evidence supports it and the places where it does not.

How many people decide, and how many situations bring them here?

The first question has numbers behind it, and they vary widely. In a Gartner survey of 632 B2B buyers, fielded in August and September 2024, buying groups ranged from five to 16 people across as many as four functions, and 74% of the buyer teams showed "unhealthy conflict" during the decision[1]. Forrester's 2026 State of Business Buying release puts the typical decision at 13 internal stakeholders and nine external influencers, with the number rising for complex or strategic purchases[2]. A Forrester blog post says 73% of purchases involve three or more departments[3].

FACT

Not every purchase is a committee. A 2015 SiriusDecisions study described three buying scenarios, from independent decisions made by one or two people in a single function to committee decisions with an average of six to ten people[4]. In a 2024 survey of 738 small and mid-sized US businesses by SMB Group, 69% of respondents said they were the sole decision-maker[5]. Both sources are dated or narrow, and neither covers Europe. They are enough to show that the size of the decision unit differs a great deal from one purchase to the next, inside the same label.

The second question, situations, comes from the consumer side. The Ehrenberg-Bass Institute defines category entry points as "all thoughts or motives that buyers may have to buy within a category"[6]. A chocolate bar can be bought as a treat, for the children, as a snack or as a thank-you. The same Institute has argued that the idea also applies to business buyers, and a study of US business insurance customers found that those who linked fewer entry points to their provider were more likely to switch[7]. A buying situation, in other words, is a unit that exists in both camps.

INTERPRETATION

Put together, the two questions give a more useful working map than the B2B/B2C label alone. A mass-market cosmetics brand has one buyer and many situations. An industrial purchase has many people and several jobs to be done. A sole-decision purchase of simple software has one person and one repeated need. We have not found a study that tests this map. It is a way of organising what the sources describe.

Who is asking in B2B?

Gartner describes B2B buying as six "jobs" that buyers cycle through and revisit, not as a straight line: problem identification, solution exploration, requirements building, supplier selection, validation and consensus creation[8]. The same source does not say which job belongs to which job title, and we have found no source that does. A buyer on the pump maker's side may therefore be an engineer checking a specification in March, a procurement manager comparing suppliers in May and a finance director looking for risks in June, or the same person doing all three.

Some of those people are hard to see. LinkedIn's B2B Institute, working with Bain, describes "hidden buyers", process experts in functions such as procurement, finance, legal and operations, who in the data it reports vetoed around half of the vendors on a shortlist[9]. Forrester's release also says procurement professionals are decision-makers in 53% of business buying cycles and engage from the start, not only at the end[2].

FACT

Consensus has a measurable effect. In the same Gartner survey, buying groups that reached consensus were 2.5 times more likely to report a high-quality deal[1]. The decision is a group outcome, and each member of the group may meet the supplier first through a different question put to an AI assistant.

What we do not know is what each role asks. We found no independent study of the questions that engineers, buyers or compliance officers put to AI assistants. That roles ask different questions is a reasonable expectation, and it is an expectation only.

Who, and in which situations, is asking in B2C?

On the consumer side, the picture is less tidy, and the evidence pulls in two directions.

One direction says that people differ. L.E.K. Consulting and Potloc surveyed more than 3,500 facial-skincare consumers in the US, the UK and France and, using machine-learning clustering, produced eight segments. The information sources differ between them: in the report, 65% of the Natural Integrity Seekers segment relied on dermatologists, against 33% of the Effortless Quality Enthusiasts, and 61% of the Effortless Quality Enthusiasts were influenced by online reviews[10]. The segments are the consultancy's model, and another analyst could have drawn different lines.

The other direction says that the lines are less stable than they look. Studies from the Ehrenberg-Bass tradition report that the buyer profiles of directly competing brands seldom differ: 700 brands in more than 60 packaged-goods categories[11], and 50 categories over 25 years[12]. The authors note exceptions, mostly submarkets that are functionally different. For a business, the implication is that a brand usually sells to the same kinds of people as its competitors, and that the useful difference lies in the situations that bring those people to the category.

INTERPRETATION

We read the two together this way. In B2C, the dependable unit is the situation and the question that goes with it, more than the demographic segment. A skincare brand has the same potential customers as its competitors. What varies is whether a person is looking for something for a reaction after a procedure, for a first routine, for a gift, or for a replacement of a product that has been discontinued. The consultancy's segments are still useful as a reminder that these people consult different sources. We have found no evidence for "levels" inside B2C in the sense we use the word for B2B, and we do not claim it.

A regulatory detail matters for the skincare example. Under the common criteria for cosmetic claims in the EU, a claim is assessed from the perspective of the average end user and must be supported by adequate and verifiable evidence[13]. The regulation governs the brand's own claims and not the answer of an unaffiliated assistant, which may still repeat a statement the brand could not lawfully make without evidence. For a brand, the risk in that case is reputational and evidential, and it is not automatically a compliance breach. It becomes a legal question if the brand publishes the statement itself, or if the brand's own pages are where the assistant found it.

What does an AI do with a question?

An assistant with web search does not necessarily submit the user's exact wording as a single query. Google's documentation says its AI features "may use a 'query fan-out' technique, issuing multiple related searches across subtopics and data sources"[14]. Google's head of Search, Liz Reid, said in a 2026 podcast interview that AI Overviews have brought "meaningfully longer queries" and that people now describe their actual need instead of compressing it into keywords[15]. In a preprint based on 171,264 real conversations, researchers at the Max Planck Institute for Software Systems and partners counted, among the turns for which web-query traces were available, an average of 3.07 web queries on ChatGPT, 1.80 on Claude, 4.55 on Grok and 2.63 on DeepSeek[16]. The figures cover only turns in which a search happened, so they do not mean that every prompt triggers several searches. They do show that platforms search in markedly different ways.

Wording affects whether an AI answer appears at all. Pew Research Center, using the browsing data of 900 US adults, found that 8% of one- or two-word Google searches produced an AI summary, against 53% of searches with ten or more words and 60% of searches starting with a question word[17]. A needs-based prompt can also change which brands an assistant names. A preprint testing six language models on five product categories found that prompts describing a need produced different brand lists from prompts naming only the category[18]. The categories did not include beauty, the models were used without web search, and the authors treat the associations as exploratory.

Sources differ across systems. In a SIGIR 2026 paper on 14,212 queries, the average overlap between the sets of sources used by Google Search, Google's AI Overviews and Gemini was between 0.11 and 0.18 on the Jaccard measure[19]. A 2026 single-author preprint that surveys 45 studies on generative engine optimization (GEO) found that no reviewed technique showed "a stable, longitudinal, cross-platform causal effect on organic discoverability or downstream behavior"[20].

FACT

We found no independent evidence on whether the same product question retrieves different sources when it is framed as coming from a procurement manager, an engineer or another member of the buying group. Research on personas in system prompts, such as a study of 162 roles assigned to the model across four model families and 2,410 factual questions, examines a different question, the role given to the assistant, and does not resolve this gap[21].

What does a strategy built on questions look like, and why not one message per person?

Putting the pieces together gives four layers. They are our interpretation of the sources above and the logic of an assistant that answers questions.

The first layer is who takes part. For the pump maker, this is the buying group and the hidden buyers around it. For the skincare brand, it is the buyer and those who shape the purchase: family, pharmacists, dermatologists, reviewers.

The second layer is why they ask. In B2C this is the situation that triggers the search. In B2B it is the job the buyer is doing: understanding the problem, comparing suppliers, validating a choice, or defending it to colleagues.

The third layer is what they ask. People now put their whole need into a question, and an assistant turns that question into several searches. The unit that has to be covered is therefore the set of questions that people in that situation put to an assistant, in their own words and at different lengths, and not a list of search terms. This layer is the most speculative of the four. The change in how people ask is documented. What each role or situation asks is not, and no study shows that covering such question sets is rewarded with citations.

The fourth layer is what the assistants find, and whether the story is the same wherever they look. This depends on which sources answer those questions: the company's own website, trade media, review platforms, regulators' pages, and the ecosystems where they sit. The sources AI draws on differ between ecosystems, and for China we described this in our earlier pieces "Cross-Ecosystem, Not Cross-Border: Why China and Europe Aren't Symmetrical AI Environments" and "The Regulatory Layer Western Companies Don't See — And Why It Shapes What Gets Cited". How the same buyers check an AI answer in Europe and in China is the subject of "From Stuttgart to Shenzhen: Before the First Call, Your Buyers May Already Have Asked AI About You".

Gartner's 2025 release reports that content tailored to the buying group's shared relevance had a 20% positive impact on consensus, while content focused on individual-level relevance had a 59% negative impact on it[1]. These findings do not argue against role-specific information. They suggest that role-specific information should connect back to a shared case the whole group can agree on, because a company that gives the engineer, the buyer and the finance director three incompatible stories risks a group that cannot agree. The layers therefore ask for different questions and one consistent story. Because answers also vary between runs, as we describe in "Why AI Answers Change Between Runs — And What That Means for Measuring Visibility", the story has to be checked repeatedly and across assistants, not once.

Does your strategy fit your business?

For the skincare brand and the pump maker, the answer starts with the two entry questions. How many people take part in a typical purchase, and in how many functions? How many situations bring them to the category? If both answers are one, a short list of questions may be enough. If they are eight and five, a content plan written for "the buyer" speaks to only a fraction of those asking.

The four layers then give the remaining questions. Who, besides the person who pays, shapes the decision? What do they ask in each situation or job? What do the assistants find when they answer, and is it the same story for each role? Does it agree with what we would say ourselves?

INTERPRETATION

What we cannot yet say is what each role and each situation actually asks. That is the gap in the evidence, and it can be closed. A fair test would collect questions from real askers, measure first how much answers vary from run to run, then check whether content built around the questions of one situation or role is cited more often for new questions from the same situation, across several assistants, with B2B and B2C tested separately. We have not run this test. Until someone does, the four layers are a way to ask better questions about a business and not proof of a better result.

Related insights

All insights & research →

About the author

Marin Tintchev is the founder of GeoSynapse and China GEO EU. He has 30+ years in ICT and international business across China and Europe, was educated at Tsinghua University, and is fluent in Mandarin.

Connect on LinkedIn →