When sales has to correct AI

The sales conversation is starting later and later. Before speaking with a salesperson, a buyer may have researched the problem they want to solve, compared alternatives, reviewed vendors, and developed an initial hypothesis about what they need. The difference is that an increasing portion of that research can now take place using generative artificial intelligence tools.

In B2B, Gartner found that 45% of surveyed buyers had used GenAI during a recent purchase, primarily to obtain information about vendors and products. The same study found that 69% preferred to validate AI-generated insights with sales representatives. The study surveyed 645 B2B buyers between August and September 2025.

The phenomenon also appears in B2C. Adobe analyzed more than one trillion visits to U.S. retail websites and surveyed more than 5,000 consumers. In August 2025, it reported that 38% of surveyed consumers had used GenAI for online shopping and that 52% planned to do so. Among its uses, 53% reported product research and 40% recommendations. Adobe also recorded a 4,700% year-over-year increase in GenAI-referred traffic to retail websites during July 2025.

Forrester found even broader adoption in its 2025 Buyers’ Journey Survey: 94% of surveyed business buyers reported using AI during their buying process. The same research indicates that buyers are validating AI results through other sources, including peers, product experts, industry analysts, and their internal and external buying networks.

The relevant point for Marketing & Sales is not simply that buyers use AI. It is that this research can reach the sales conversation before the salesperson does, and when a person arrives with previously researched information, the conversation no longer necessarily begins with identifying the problem. It may already have begun with a conclusion in mind.

Maximizing sales through Customer Experience

The problem arises when a conclusion needs to be revisited

An AI-generated response can be useful while also containing incorrect, imprecise, or incomplete information.

NIST defines “confabulation” in its Generative AI Risk Profile as the production of erroneous or false content presented with confidence.

OpenAI also describes hallucinations as plausible but false statements generated by language models. Its analysis explains that certain training and evaluation processes can favor producing an answer rather than acknowledging uncertainty, even when the model does not have a sufficient basis to answer correctly.

This establishes a concrete possibility in a sales interaction: a buyer may use AI for research and arrive with a statement that requires validation. That does not mean the buyer necessarily believes everything they receive from AI. A study published in Frontiers in Computer Science compared participants’ trust and reliance on recommendations from an LLM with those relating to AI-generated snippets. The experiment, with 199 participants, found no significant difference in reported trust between the two sources. When participants were informed that a recommendation was incorrect, their trust decreased.

The commercial question, therefore, is not to assume that the buyer is wrong because they used AI. Nor is it to assume that the information they bring is correct because it comes from a tool that can present its responses in a structured and plausible way. The question is to determine what information needs to be validated.

That change is important because it places a new task within the sales conversation: when the salesperson identifies an incorrect, incomplete, or out-of-context premise, they need to be able to correct it without losing the purpose of the conversation.

Inbound with AI to accelerate repetitive sales cycles

The buyer can arrive convinced before arriving informed

Research on belief updating helps explain why this situation requires care. Sanna and Lagnado studied how people update their beliefs when they receive contradictory information or a retraction, also considering the reliability of the source. Across four experiments, they found that sources considered reliable helped participants discount initial claims more effectively. Source reliability incorporated both perceived reliability and elements related to trust and expertise.

The implication for the sales conversation is concrete: when a buyer arrives with information that needs correction, the salesperson’s work does not end with presenting a different fact. They must also make that new information sufficiently credible for it to modify the premise with which the conversation began.

This matters especially because prior research may have been part of a broader sequence. The buyer did not necessarily consult a single source. Gartner found that B2B buyers used an average of seven information sources during the buying process. Therefore, AI-generated information does not necessarily replace other sources. It can be incorporated into them. The buyer may arrive with an interpretation built from several sources, including an AI tool. If one of those premises needs correction, the salesperson must identify it and work through it before continuing to build a commercial recommendation.

Gartner provides a particularly relevant finding here: although 67% of surveyed B2B buyers preferred an experience without sales involvement and 70% preferred a completely digital or self-service process, 69% also said they preferred to validate AI-generated insights with sales representatives.

The figures do not describe a contradiction that needs to be resolved. They describe two needs that can coexist: the buyer may want to research and move forward digitally on their own while also turning to sales when they need to validate what they found. And that is where a specific function appears within the sales interaction.

Correcting information is part of the conversation, not the objective of the conversation

An effective correction is not simply a matter of replacing one statement with another. Research on information correction shows that corrections can improve the accuracy of beliefs, although their effects are not always permanent and do not necessarily produce equivalent changes in attitudes or behaviors. Porter and Wood’s review concludes that factual corrections improve belief accuracy across countries, beliefs, and demographic groups, and that backfire effects are uncommon. Prike and colleagues’ review also identifies conditions that can support effective correction, including detailed corrections and alternative explanations. Within a sales conversation, this means that identifying an incorrect statement is only the beginning. Sales’ purpose remains to advance understanding of the problem and determine whether an appropriate solution exists. The correction has value because it allows that conversation to continue on a more accurate basis.

If the salesperson identifies that the buyer is starting from an incorrect premise and simply contradicts it, the interaction can become a discussion about the source of the information. If, instead, they identify what needs validation, explain the difference, and establish the information relevant to the buyer’s situation, the conversation can return to the problem that originally needs to be solved.

The evidence does not allow us to claim that this practice increases a specific conversion rate, reduces the sales cycle, or automatically produces greater trust. What it does establish is that belief updating depends, among other elements, on the reliability attributed to the source, and that corrections can improve belief accuracy. That limitation is important.

The challenge for Marketing & Sales is not to prove that AI is wrong. It is to be prepared for a situation in which AI-generated information needs to be validated or corrected within a sales interaction.

B2B: Sales receives a new kind of prior information

In B2B, the change has a particularly visible consequence: the buyer may use GenAI to research vendors and products before involving sales. Gartner found that, among B2B buyers who used GenAI during a recent purchase, its primary use was related to obtaining information about vendors and products. The sales conversation, then, can begin after a stage of research that the salesperson did not witness, and that changes the starting point, because the salesperson may receive questions, objections, or statements whose origin is not found in traditional sales materials. The buyer may be looking to confirm a conclusion they have already formed during their research.

In this context, Gartner’s finding on validation becomes particularly relevant: 69% of surveyed B2B buyers prefer to validate AI-generated insights with sales representatives. Sales can thus become a source of validation within a process that has already begun before the commercial contact. But the need for validation is not exclusive to AI. Forrester notes that business buyers are validating AI results through other sources, including peers, product experts, analysts, and buying networks. Therefore, the opportunity for sales is not to become the sole arbiter of information. It is to be able to participate credibly when the buyer needs to determine what information applies to their situation.

Your next visitor isn't human: get your website ready for AI

B2C: research takes place before the interaction with the brand

In B2C, the dynamic appears at a different scale and with a different type of buying process. Adobe reported that 38% of surveyed consumers had used GenAI for online shopping and that 53% planned to do so. Product research was the most frequently reported use, at 53%, followed by product recommendations, at 40%. In March 2025, Adobe had already reported that 39% of surveyed consumers had used GenAI for online shopping and that 53% planned to do so that year. In that research, 55% said they used GenAI to research products and 47% to receive recommendations.

Deloitte also found growth in the use of GenAI as a shopping tool: in a survey of 4,000 U.S. consumers conducted for the 2025 holiday season, 33% said they planned to use GenAI for their shopping, more than double the previous year. Reported uses included finding prices, summarizing reviews, and creating personalized shopping lists.

The consequence for Marketing & Sales is similar in principle: part of the evaluation process can take place before the person interacts directly with the brand. If the AI-generated information is incomplete or incorrect in a particular case, the commercial organization may encounter a consumer who already has a prior expectation, comparison, or interpretation.

The difference from B2B lies in the nature of the interaction. In many B2C processes, there is no individual salesperson who can participate in every decision. The responsibility of Marketing & Sales may therefore fall on the different touchpoints that allow the consumer to validate information and move forward with their decision.

The available evidence does not allow this to be turned into a specific correction methodology for B2C. It does allow us to identify the change in context: GenAI already participates in purchase research and can influence the information with which consumers arrive at an interaction with the brand.

The question for Marketing & Sales is changing

The phenomenon does not require assuming that AI is replacing sales or that buyers are abandoning other sources of information. The available data show something more specific: buyers are already using AI in their buying process. In B2B, Gartner found that 45% of surveyed buyers had used it in a recent purchase and that 69% preferred to validate their insights with sales. In B2C, Adobe found significant use of GenAI for product research and recommendations. Forrester reported that 94% of business buyers used AI during the buying process and that these results were being cross-checked against other sources. At the same time, models can generate false or erroneous information presented in a plausible way. NIST and OpenAI document this possibility. And research on belief updating indicates that source reliability matters when a person receives information that contradicts a prior belief.

Therefore, the relevant question for Marketing & Sales is not whether AI is good or bad for sales, but “What should the commercial organization do when a buyer arrives with AI-generated information and part of that information needs to be validated, corrected, or contextualized?”

The answer begins before the correction. Marketing and Sales need to recognize that the research process may have begun outside their channels and that the buyer may arrive with an interpretation formed before the conversation. When that interpretation is correct, the conversation can move forward. When it needs to be reviewed, the salesperson or the corresponding point of contact must be able to identify it.

Correction should not become the ultimate objective either. Its function is to allow the conversation to continue with information appropriate to the buyer’s specific problem.

That is the fundamental change: the commercial organization is not necessarily involved in the initial construction of the information the buyer uses. In certain cases, it participates in its validation. And when that information needs to be corrected, Marketing & Sales’ ability to do so becomes part of the quality of the commercial interaction.