What Marketers Need to Understand About AI Right Now: A Big Dogs Network Leadership Conversation with Adam Gordon
The March Big Dogs Network conversation with Adam Gordon focused on the generative AI tools that most professionals are now using on a regular basis, including platforms such as ChatGPT, Claude, Gemini, and NotebookLM. While the broader field of AI includes many other approaches, the discussion centered on the systems that are most visibly reshaping day-to-day work.
From there, the conversation moved beyond tools and into something more foundational. If these systems are going to influence how we think, work, and make decisions, then understanding how they operate and how to interpret their outputs becomes essential.
Understanding What Sits Beneath the Interface
One of the most valuable contributions Adam made was grounding the conversation in how these systems actually work.
Today’s large language models are probabilistic systems. At a high level, they operate by identifying patterns in how words and phrases tend to follow one another. Simply, they break inputs into tokens (each about ¾ of a word) to understand what the user means, then generate outputs (new strings of tokens, then words) based on patterns in language. They are not evaluating truth in a deliberate sense, nor are they reasoning in the same way humans do. Rather, they are non-human reasoning systems that can simulate coherence by predicting what is likely to come next.
Why this matters becomes clear very quickly in practice.
These systems can be remarkably articulate. They can produce responses that are thoughtful, informed, well-structured…and wrong. Do not mistake their fluency for understanding. As Adam noted, they can deliver answers with a high degree of confidence and positivity regardless of whether those answers are correct.
For professionals, particularly those working in areas where accuracy and credibility are essential, this creates a subtle but important shift. A well-written answer has never actually been a reliable indicator of validity, and it’s less so with these LLMs. The bottom line: check everything you care about.
When Confidence Outpaces Accuracy
This dynamic surfaced quickly in the discussion through questions from the group.
If these systems can produce confident but potentially incorrect outputs, how should professionals approach validation?
Adam’s response emphasized the need to step outside the AI environment when accuracy matters. Asking the AI for citations helps, but it is not sufficient. Those sources need to be checked directly. Do they even exist? Are they accessible (e.g., not behind a paywall)? Do they actually support the claims being made?
He also acknowledged something many participants have experienced. Different AI systems may produce different answers to the same question, and even the same system may vary its responses depending on context. This variability reflects how these models operate and is something users need to stay aware of.
The implication is not that these tools lack value. On the contrary, they can significantly accelerate learning and exploration. But they operate at the level of plausibility, not verification. The responsibility for determining what is accurate, current, and relevant remains with us.
From Prompting to Clarity of Thought
A particularly resonant theme in the conversation was the shift from focusing on prompts to focusing on thinking.
Early discussions about generative AI often centered on crafting better prompts. While that remains relevant, Adam reframed the issue in a more fundamental way. The quality of the output is closely tied to the clarity of the intent behind it.
What is the desired outcome?
What defines a strong versus a weak result?
What constraints or boundaries should shape the answer?
What sources or types of information should be prioritized or avoided?
When these elements are clearly defined, the system performs more effectively. When they are not, it fills in the gaps in ways that may appear reasonable but may not align with the original objective.
This point prompted strong agreement from the group. The discipline required to work effectively with these systems mirrors something more familiar. Clear thinking leads to better outcomes, whether the audience is an AI system or a human team.
Moving Beyond Interaction to Execution
The conversation also explored how these systems are evolving from tools we interact with to systems capable of executing more complex tasks.
Adam introduced the distinction between standard models and newer reasoning models. While both rely on similar underlying principles, reasoning models are designed to handle more complex, multi-step processes. They can break tasks into components, iterate through them, and produce more structured outputs.
This shift changes how they are best used.
Rather than guiding every step or overly constraining the system with a predefined role, users are often better served by defining the context upfront. This includes articulating the desired outcome, providing examples of what good and poor outputs look like, specifying relevant data sources, and establishing constraints.
One question from the group focused on the practical implications of this approach. How much time does it take to set up?
Adam’s response was measured. The initial setup can take time, particularly when defining context thoroughly. But that investment creates a reusable structure. Over time, it functions less like a one-off interaction and more like an infrastructure that supports ongoing work. As with any infrastructure, there's an up-front investment that pays off enormously over time.
Practical Realities and Emerging Use Cases
The discussion also included a number of practical questions about how these tools are being used day-to-day.
Adam described working across a small ecosystem of platforms rather than relying on a single solution. Tools such as ChatGPT, Claude, Gemini, and NotebookLM were referenced as examples of the current landscape, each with its own strengths depending on the task. Some are particularly effective for synthesis and writing, others for organizing and working with structured inputs.
He also noted a shift in how search itself is being approached. Rather than relying exclusively on traditional search engines, tools like Perplexity are increasingly being used to synthesize information and accelerate understanding, particularly in early-stage exploration.
This multi-tool approach reflects how many experienced users are operating today. Rather than searching for a single “best” platform, they are developing a working understanding of which tools are most effective in different contexts and moving between them accordingly.
At the same time, he cautioned against over-automation without clear intent. One example discussed involved AI-driven customer service system that was optimized for efficiency, specifically volume of interactions handled, rather than customer experience. The result was unhappy customers at scale.
The broader point is straightforward. AI will optimize for what it is asked to optimize for. If the objective is not well defined, the outcome may be technically successful but strategically misaligned.
Another practical issue raised by the group was the tendency for long conversations with AI systems to degrade in quality over time. Adam acknowledged this as a limitation, noting that as context accumulates, meaning can drift. His recommendation was pragmatic. Reset the conversation when needed and carry forward only the most relevant context.
Where These Systems Add Value, and Where They Do Not
Throughout the conversation, there was a consistent effort to stay grounded in what these systems do well and where their limitations remain.
They are highly effective in areas that benefit from speed, synthesis, and pattern recognition. They can help generate ideas, summarize complex information, and support early-stage exploration.
They are much less reliable when tasks require judgment, contextual understanding, or nuance. They do not inherently understand business implications or human motivations. Those elements still depend on a human's expertise.
For those in marketing and research, this distinction is particularly important. The value of the work does not lie solely in the information gathered, but in how it is interpreted and applied.
The Takeaway
If there was a unifying theme to the conversation, it was this: The advantage does not come from access to AI, but from using it well.
These tools are already widely available. What differentiates professionals is their ability to think clearly, define objectives precisely, and apply judgment consistently.
The conversation was not about replacing expertise. It was about extending it.
As with previous shifts in technology, the fundamentals remain intact. Clarity, rigor, and interpretation continue to matter. What is changing is the environment in which those capabilities are applied, and the speed at which work can now move.
For the Big Dogs Network, the session served as a timely reminder that while the tools are evolving rapidly, the responsibility to use them thoughtfully remains firmly human.
Learn More
The Big Dogs Network connects senior marketing consultants, strategists, and agency leaders experienced with $500M+ organizations to collaborate, share ideas, and elevate the marketing profession. To learn more, contact Ron Snyder at rsnyder@innovations86.com or Kirsty Nunez at kirsty.nunez@q2insights.com.
To connect with Adam Gordon and learn more about his work in AI, marketing strategy, and growth initiatives, contact him at adam@sapients.co or 408.499.6878.
This article is part of a continuing series sharing insights from Big Dogs Network events and conversations with leaders in large-scale marketing organizations.
Article by Kirsty D. Nunez, President and Chief Research Strategist at Q2 Insights, and member of the Big Dogs Network leadership team.
