The Customer Journey Has a New Middleman

Recently, I had a fairly ordinary customer-service issue with a Walmart grocery order. I had expected to qualify for free delivery, but a delivery fee was charged. The issue itself was trivial. What interested me was how little of the eventual resolution required my direct involvement.

My AI assistant had access to the relevant emails, reminded me that the issue remained unresolved and helped manage the follow-up until Walmart refunded the charge as an exception.

There was nothing particularly sophisticated about the individual actions. The significance was in the role the AI played between me and the company.

Walmart had communicated with me through conventional digital channels, but an intermediary was increasingly deciding what deserved my attention, interpreting those communications and helping determine what happened next.

I used that example recently while speaking at the Future of Pharma Marketing & Sales Summit in Toronto because it illustrates a broader shift that I think marketers need to understand.

For years, we have invested heavily in designing digital customer journeys. We consider the appropriate channels, content, sequencing, experience and measurement. AI is now introducing another participant into that system: one that may interpret our information before the intended customer interacts with it directly.

That has important implications for pharma.

Digital marketing has always involved intermediaries. Search engines influence discovery. Social platforms determine distribution. Publishers provide context and credibility. Recommendation systems shape what appears in front of a user.

Generative AI is different because it does not simply rank or distribute information. It can interpret it.

A user can ask a question and receive a synthesized answer assembled from multiple sources. They can ask an assistant to summarize communications, compare options or identify the information most relevant to a particular task.

The original content may still be important, but the customer does not necessarily need to consume it in its original form.

This changes the role of the intermediary. The search engine primarily helped a user decide where to go. An AI assistant can increasingly attempt to answer the question before the user goes anywhere.

For marketers, that means visibility alone is no longer enough. We also need to understand how information is represented once another system has interpreted it.

The emergence of AI-generated answers has understandably created significant interest in generative engine optimization.

There is a legitimate question here. If customers increasingly use AI systems to research topics, organizations need to understand whether their information is discoverable and whether authoritative sources are being used.

But I think the conversation becomes too narrow when GEO is treated simply as the next version of SEO.

The objective cannot just be to increase the frequency with which a company or brand appears in generated responses.

Particularly in healthcare, representation matters as much as visibility.

If an AI system references information from a source, we need to consider whether the resulting answer reflects that source accurately. Important qualifications may be lost. Geographic context may be confused. Information intended for one audience may be interpreted in another context.

These risks exist because generative systems do not simply retrieve information. They reconstruct it.

The question is not only whether our content can be found. It is whether the information is sufficiently clear, well-supported and appropriately structured that it remains accurate when interpreted outside its original presentation.

This is particularly relevant for email, where marketing organizations have spent years improving communications through better segmentation, personalization, design and content strategy.

AI assistants increasingly give recipients another option: rather than reviewing every communication individually, they can ask an assistant to identify what is relevant. I already use AI this way myself, but it means the communication may be compressed before it reaches the person.

A detailed email can become a short summary, several communications may be combined into one response, and information that appeared prominently in the original design may receive less emphasis in the generated version.

This becomes especially important when the meaning of a communication depends on context. A summary of an event invitation is useful if it preserves the information someone needs to decide whether to attend, while a summary of more consequential healthcare information needs to retain the appropriate qualifications and supporting context.

The original communication can be completely accurate while the recipient still receives an incomplete representation of it. That is a consequence of introducing another interpretive layer into the communication process, and marketers need to understand that layer rather than assuming the original experience remains intact.

The stakes are obviously higher when the information concerns health. AI is already becoming part of how patients and healthcare professionals navigate complex information environments, whether that means understanding terminology, preparing for an appointment, summarizing research or organizing large volumes of material.

There is considerable potential value in these behaviours, but also an obvious requirement for accuracy.

Pharma already operates within a communication environment where audience, evidence, indication, regulatory requirements and appropriate context matter enormously. AI does not remove those considerations; it creates another environment in which they need to be respected.

This is why I am cautious about simplistic calls to make all pharmaceutical content more accessible to AI systems. Some information is public and some is intended specifically for healthcare professionals. Authentication and access restrictions exist for legitimate reasons and should not be viewed as technical barriers that need to be circumvented in pursuit of AI visibility.

Public information should be accurate, current and understandable, while HCP-directed environments should continue to respect the relevant access requirements. Organizations should understand where their information is appearing in AI-generated responses without assuming they can or should control every answer.

This is as much a governance question as a marketing one.

The traditional model of a website assumes that someone arrives on the page after marketing activity generates interest, and the website then provides the intended experience. That behaviour will continue, but websites increasingly have another role: they provide information that may contribute to an answer delivered somewhere else.

A customer may learn something derived from an organization's website without ever visiting it, while an AI research tool may use several sources to create an overview before a user decides which original material deserves closer inspection.

This means organizations should think about their digital content both as an experience and as a source of structured information.

Fortunately, many of the principles that improve one also improve the other. Clear language, strong information architecture, current content, evidence, citations and appropriate context all matter.

There is a tendency in emerging technology to assume that a completely new technical discipline must accompany every behavioural shift. Sometimes it does, but I suspect a significant part of preparing content for an AI-mediated environment is simply becoming much better at fundamentals we should already value.

If the core point of a page is difficult for a human reader to identify, an AI system has no special obligation to interpret it correctly.

This shift also complicates digital measurement because our current analytics systems are strongest when the customer interacts directly with environments we control or can observe.

An AI intermediary introduces activity that may occur outside those observable environments. A customer could receive a useful answer informed by an organization's content without visiting the original source, or they could receive an inaccurate answer and never reach the source that would have corrected it.

From the organization's analytics, both interactions may look identical: nothing happened.

That creates an uncomfortable measurement gap. Existing digital metrics still describe important parts of customer behaviour, but they may describe a decreasing proportion of the information journey.

The absence of a visit does not necessarily mean the information had no influence, while appearing within an AI-generated answer does not demonstrate that the interaction was useful, accurate or trusted.

We will need better ways to understand this environment without pretending that everything important can be reduced to a single visibility score.

I do not think organizations need to begin with a large transformation programme. A more practical first step is simply to understand the current environment.

Take a small set of questions that customers genuinely ask based on existing insights, search behaviour or recurring information needs, then ask those questions across the major AI platforms and assess the responses in the appropriate geographic and audience context.

Look at which sources are being used, whether the information is current and whether important details remain intact.

For a Canadian organization, this should include an assessment of whether Canadian information is being distinguished appropriately from US and global sources. Where relevant, the same exercise should be conducted in French rather than assuming identical performance across languages.

The objective should not be to create a flattering benchmark; it should be to understand what a customer might realistically encounter today.

That exercise alone can reveal useful information about digital content quality, source authority and areas where existing materials could be clearer.

One-off testing is useful, but the environment changes too quickly for a single audit to remain meaningful for long because models, search integrations, available sources and user behaviours all change.

Organizations therefore need a lightweight way to monitor important questions over time. A defined set of priority questions, repeated periodically across major platforms, can provide a useful view of how the information environment is evolving.

Material changes can then be investigated. If an answer becomes inaccurate, teams can assess whether the underlying source has changed, whether another source is being preferred or whether the model is interpreting the available information differently.

That is a more useful operating model than treating GEO as a one-time optimization exercise because it turns the question from 'How do we rank in AI?' into 'How is the information environment our customers encounter changing?'

Monitoring also creates an organizational question: what happens when somebody discovers an important inaccurate AI-generated response?

Marketing may identify the problem, but marketing may not be the appropriate function to determine whether the information is clinically significant, whether the underlying source requires modification or whether any external action should be taken.

Organizations need a clear pathway for evaluating these situations. That does not mean every incorrect AI response requires escalation, but where the information is material, teams should know who assesses it and what evidence should be captured.

The same cross-functional disciplines that already support responsible pharmaceutical communication remain relevant here. AI introduces a new surface; it does not eliminate the need for medical, regulatory, legal, privacy and technology expertise where those perspectives are appropriate.

The most useful lesson from my Walmart experience was not that AI is replacing the customer journey. The customer journey was never really the objective; the objective was to help a customer accomplish something.

In my case, I wanted to resolve a small billing issue, and the company's preferred digital journey was secondary to achieving that outcome efficiently.

Healthcare communication is considerably more complex, but the underlying principle still holds. Patients and healthcare professionals do not need more content simply because organizations are capable of producing it; they need useful, accurate and appropriately contextualized information.

AI is becoming another mechanism through which that information may reach them. That creates new questions for marketers, but it does not fundamentally change the standard we should be aiming for.

The information still needs to be useful and trustworthy, and in healthcare it needs to retain the context required to interpret it appropriately.

The significant change is that we may no longer control the interface through which the customer encounters it.

For years, digital marketing has focused heavily on designing the journey between our content and the customer. The next challenge is understanding what happens when AI becomes part of that journey too.