Back to Blogs
Drew Neisser is the founder of CMO Huddles and a globally recognized authority on B2B marketing. He’s an AdAge columnist, LinkedIn TopVoice, leading CMO coach, podcast host & friend of penguins everywhere.
AI agents are beginning to research vendors, compare alternatives, investigate pricing, and test buying experiences on behalf of B2B decision-makers. Drawing on four CMO Super Huddle workshops led by Pepper co-founder Kishan Panpalia, this article explores what happens when an AI agent effectively joins the buying committee and why CMOs must rethink the information their websites make accessible.
For years, B2B marketers have designed websites for two audiences: human buyers and search-engine crawlers. The human needed a persuasive experience. The crawler needed enough structure and relevance to help the page get discovered.
A third audience has arrived, and it behaves differently from both.
AI agents can research a category, assemble a shortlist, compare vendors, investigate pricing, evaluate implementation requirements, and report their findings to a buyer. Increasingly, the first visitor from a buying committee may not be a person at all.
The bot is becoming part of the buying committee.
That possibility dominated four workshops led by Kishan Panpalia, co-founder of Pepper, at the CMO Super Huddle. Kishan described the evolution succinctly: digital discovery moved from humans using search engines, to LLM crawlers assembling answers, and now toward agents that can take action.
This is still an emerging behavior, and no CMO should rebuild the entire website around a prediction. Yet the implications are already concrete. When an agent cannot find a price range, migration plan, security answer, technical detail, or implementation path, it may downgrade the vendor before a human ever reaches the website.
Disclosure: Pepper was a Founding Sponsor of the CMO Super Huddle. This article draws on Kishan's workshop framework and the questions raised by participating CMOs. Kishan Panpalia and Liza Adams are quoted by name; other participants are quoted anonymously. Quotes have been lightly edited for clarity.
Before CMOs spend more money trying to improve AI visibility, they should understand how uncertain the underlying measurement remains. Platforms such as Profound, Pepper, and hundreds of newer monitoring tools can track how a brand performs across a selected set of prompts. They cannot definitively tell a CMO how frequently commercially relevant buyers are asking those questions because the major LLM providers do not release complete query-volume analytics.
Kishan was unusually candid about this limitation. “Not Profound, not us. Don't trust anyone with that,” he told one workshop. He explained that monitoring providers buy overlapping pools of third-party chat-session data, then combine those samples with proxies such as Google search volume, paid-media patterns, and social-listening signals.
The result is informed guesswork packaged with impressive precision. A platform recommends the prompts a company should monitor and then grades the company against those same prompts. Kishan described the conflict memorably: “The heads are mine, the tails are mine, and the coin is mine.”
That does not make every monitoring platform worthless. It makes the role of the platform narrower than many dashboards imply. CMOs can use these tools to observe performance across a prompt set they have independently validated, compare answers over time, and identify the sources LLMs cite. They should be far more skeptical when a vendor claims to know exactly which prompts matter most or how often buyers use them.
Kishan recommended starting with primary evidence: sales-call recordings, customer interviews, support questions, win-loss research, and the recurring concerns raised by the buying committee. A CMO in the room summarized the principle perfectly: “Trust your primary research.”
This matters because optimizing for the wrong prompts can produce an attractive visibility score without influencing a single meaningful purchase. A company might dominate broad educational questions while remaining absent from the comparisons, pricing questions, migration concerns, and technical evaluations that determine the shortlist.
Precision on a dashboard does not turn a proxy into demand.
B2B buying committees already include end users, functional leaders, finance, procurement, security, IT, legal, and executive sponsors. Each person enters the process with different questions, and the research rarely follows a clean path from awareness to demo request.
Kishan shared the example of a hospital supply-chain manager asking an LLM to recommend revenue-cycle-management vendors for a smaller hospital. The manager forwards the shortlist to a CFO, who passes it to procurement. Procurement searches the recommended brands directly, visits their websites, and continues the evaluation.
The marketing team may see branded search or direct traffic without realizing an AI system shaped the consideration set. More important, the agent has already influenced which vendors receive attention. The company absent from the initial answer may never know it lost.
Kishan's research with roughly 4,000 procurement leaders reinforced the point. He said 93% reported using an LLM at least once during a buying process, most often for category shortlists, direct vendor comparisons, and searches for less expensive alternatives.
Those are not casual educational queries. They are consideration and procurement questions with commercial consequences.
The next change is behavioral. A conventional search engine returns links and waits for the person to decide what happens next. An agent can receive a goal, visit multiple websites, gather information, complete comparisons, and return with a recommendation.
Liza Adams demonstrated this distinction during one of the workshops. She instructed an AI agent to act as the CMO of a global company evaluating project-management software. After the agent produced a shortlist and scored the vendors, she sent it to each website to investigate pricing, migration, and webinar registration.
The resulting report showed where the agent found answers, encountered friction, and reached a dead end. Liza explained that several sites performed reasonably well near the top of the journey but struggled as the questions became more specific. When she asked how a three-to-six-month migration would work, the websites often provided little more than a form asking for an email address.
A human buyer may tolerate that interruption, open another tab, ask a colleague, or book a sales call. An agent tasked with comparing several vendors can simply record “answer unavailable” and move on.
A gated answer can become a missing answer.
Most B2B websites are designed to create conversion moments. Pricing is withheld to encourage a sales conversation. Research is gated to capture a lead. Technical documentation may live in a separate product system. Security and migration information often appear only after a buyer enters an active sales process.
That structure assumes a human will accept the exchange. AI agents introduce a different calculus: retrieve the answer quickly, compare it with alternatives, and complete the assigned task. If another vendor makes the necessary information easier to obtain, accessibility itself can influence the shortlist.
One CMO in the room raised the tension directly. Their company had recently placed technical documentation behind a firewall because of legal, intellectual-property, security, and brand concerns. Those concerns were legitimate, but the solution also made useful product information unavailable to the systems helping buyers conduct research.
Kishan suggested creating public summary pages containing approved information, with deeper documentation remaining protected. The summary can answer the essential technical questions, establish the relevant entities, and direct qualified humans toward the complete resource.
This does not require publishing trade secrets or pretending every enterprise product has a simple list price. It requires deciding which information an informed buyer needs before agreeing to a conversation and making an approved version retrievable.
Pricing exposes the challenge particularly well. Many enterprise companies avoid publishing prices because contracts depend on users, usage, integrations, service levels, implementation requirements, and negotiated terms.
Agents will still try to answer pricing questions. Kishan argued that if the company provides no useful context, an agent may turn to review platforms, analyst reports, community discussions, or competitors. As I observed during the workshop, the choice is increasingly between defining the range yourself and allowing someone else to define it for you.
A useful pricing page can explain the variables, typical packaging, representative ranges, qualification criteria, or the information required to produce an estimate. It can acknowledge complexity without creating an information vacuum.
The same principle applies to implementation and migration. A company may not be able to promise an exact timeline, but it can explain the phases, dependencies, resources, common risks, and questions that determine the plan.
“Talk to sales” is a CTA. It is not an answer.
Agent readiness reaches beyond content. Liza's experiment showed that agents can evaluate the mechanics of a website: whether they can locate information, navigate registration, access a resource, and progress toward an outcome.
Kishan described similar tests in which agents attempted to shop or complete tasks on websites, then logged the point of failure. A collapsed product detail, inaccessible form, JavaScript-dependent document, or unstructured page can prevent an agent from retrieving information that is technically present.
This creates a new dimension of customer experience. CMOs have long studied the paths taken by human visitors. They now need to understand what a delegated agent experiences when researching on the human's behalf.
One practical starting point is to give an agent a realistic buying assignment rather than asking it to “audit the website.” Define the buyer, company size, use case, constraints, budget context, security needs, migration timeline, and required integrations. Then ask the agent to compare your company with several competitors and document where it found or failed to find each answer.
The result will not be a definitive score. It will reveal where an increasingly plausible buyer journey breaks.
Becoming accessible to agents does not mean publishing more content about everything. Kishan shared that Pepper removed 293 of roughly 400 blog posts so the company could concentrate on a clearer market idea. He reported that its AI citations increased approximately sixfold over the following eight weeks “because we were known for something.”
The lesson extends beyond citation performance. Agents synthesize information across the website and the wider market. A sprawling archive filled with outdated positions, loosely related topics, and inconsistent product claims makes the brand harder to understand.
A focused company gives both humans and machines a clearer answer to four questions: What does this company do? Who is it for? What problem does it solve? Why should it make the shortlist?
More content creates more signals. Focus makes those signals coherent.
AI agents do not limit their research to company-controlled pages. They consult review sites, analyst research, YouTube, Reddit, LinkedIn, structured databases, customer commentary, and other third-party sources.
Kishan urged CMOs to think about the entire buying committee's questions and the sources agents are likely to trust for each one. Product documentation may answer a technical question, while peer communities or review platforms may carry more weight when the agent evaluates implementation difficulty or customer satisfaction.
YouTube deserves particular attention. Kishan said he has seen highly relevant videos with modest audiences outrank videos with far more views because LLMs can use the transcript, description, and metadata to answer a specific question. The content does not have to become a viral hit to become useful evidence.
This should change how CMOs evaluate authority. Owned content establishes the company's case, while credible third parties help an agent validate it. The goal is a consistent, supportable story wherever the agent looks.
This is not a request for another generic website audit. It is a leadership conversation about whether the digital buying experience reflects how decisions are beginning to happen.
The most useful next step is simple: give an AI agent the kind of assignment a serious buyer might delegate. Ask it to create a shortlist, compare the finalists, investigate pricing and migration, identify security concerns, and recommend the best option for a defined company and use case.
Study the result with your web, product marketing, content, product, engineering, security, and legal leaders. The gaps will cross organizational boundaries, which is precisely why the CMO needs to lead the conversation.
AI agents may not control most B2B purchases today, but they are already influencing discovery and evaluation. CMOs who wait for perfect attribution or universal adoption risk learning about the change after their brand has quietly disappeared from the shortlist.
For companies that need help understanding how agents and LLMs discover, evaluate, and represent their brands, visit Pepper. Pepper was a Founding Sponsor of the CMO Super Huddle, and Kishan's workshops gave participating CMOs a candid view of both the opportunity and the limits of today's measurement tools.
An AI buying agent is software that performs research or actions on behalf of a buyer. It can investigate a category, compare vendors, retrieve pricing and implementation information, test website experiences, and summarize recommendations based on the buyer's requirements.
A search engine returns links, while an LLM crawler retrieves information to help generate an answer. An agent can pursue a goal through multiple steps, such as visiting several vendor websites, collecting specific facts, testing forms, comparing the results, and reporting a recommendation.
They do not need to publish a universal list price. They should consider providing useful pricing context, including packaging, variables, representative ranges, or the information needed to produce an estimate. Otherwise, agents may rely on less accurate third-party sources.
Create public summary pages containing approved findings and essential context, then direct human buyers to the complete gated or protected resource. Legal, security, product, and marketing teams should agree on what can be made retrievable.
Give an agent a realistic buying assignment with a defined company, use case, budget, technical requirements, migration timeline, and comparison set. Ask it to document the sources used, unanswered questions, points of friction, and reasons behind its final recommendation.
Not definitively. Most monitoring providers rely on samples of third-party chat sessions and proxies such as search volume and social-listening data because the major LLMs do not release complete query-volume analytics. Use these platforms to monitor independently validated prompts, and use customer research, sales conversations, support questions, and buying-committee evidence to decide which prompts matter.