The most striking observation from a recent major AI conference was not about a new model capability or a breakthrough benchmark. It was about people.
Across sessions spanning healthcare, government, education, and enterprise, the same theme surfaced repeatedly: organizations have access to AI tools. Many are already using them. But confidence in what those tools are actually doing, and trust in the organizations behind them, is lagging far behind adoption rates.
One presenter described it as a "trust paradox": willingness to use AI-enabled technologies is outpacing genuine trust in their capabilities. In research, we would call this a signal worth paying attention to.
The trust paradox has direct implications for market research. When people use tools they do not fully understand or believe in, they use them differently than they would otherwise. They may anchor on AI-generated outputs without interrogating them. They may dismiss findings that conflict with their intuitions, not because the findings are wrong, but because they do not trust the system that produced them. They may use AI at the top of their workflows and then abandon rigour at the point where human judgment actually matters.
The result is not necessarily faster, cheaper research. It can be faster, cheaper-looking research that carries uncertainty, the kind that only becomes visible when a business decision goes wrong.
This is the challenge the industry is grappling with, and it is one we have thought carefully about in designing how AI works inside the itracks platform.
Several sessions at the conference explored how large language models are evolving from systems that produce words to systems that take actions, what researchers are calling the shift "from words to action." This progression matters for research technology in ways that are not always obvious.
Early AI tools in research were essentially pattern-matchers: they could identify frequently occurring themes in transcripts, flag sentiment shifts, or surface keywords. Useful, but limited. The newer generation of AI reasoning capabilities, shaped by reinforcement learning and more sophisticated training approaches, is beginning to enable something different: genuine interpretive inference, the ability to surface what participants meant, not just what they said.
This is significant because qualitative research has always been an exercise in interpretation. A focus group is not just a list of statements. It is a conversation with texture, with hesitation, with what is left unsaid. For AI to add real value in this space, it needs to operate at that level, not as a replacement for trained human moderators and analysts, but as a tool that extends their reach and sharpens their perception.
The itracks Analysis Assistant is built around this understanding. It surfaces themes, detects sentiment, and generates highlight reels but it is designed to work alongside moderators and researchers, not to replace the judgment that makes qualitative research valuable in the first place.
A philosopher of technology at the conference laid out three criteria for evaluating whether an AI company is actually trustworthy, not just trustworthy-sounding. They are worth quoting almost in full because they translate directly to what we believe research clients should be demanding from their technology partners:
Transparency: Are they honest about what their system does, where it works well, and where it does not? Do they communicate clearly about data handling, model limitations, and how outputs should be interpreted?
Consistency: Do they apply the same ethical commitments across all clients and contexts? Are their products reliable, or do they perform well in demos and degrade in the field?
Responsiveness: When researchers raise concerns about outputs, do they listen? When evidence demands updating, do they update?
Applied to research technology, these criteria have practical implications. Transparency means clear documentation of what the AI analysis actually measures and what it does not. Consistency means the same quality of output whether you are running three sessions or three hundred. Responsiveness means a platform that improves based on real research practitioner feedback, not just internal assumptions about what researchers need.
These are standards we hold ourselves to. They are also standards we encourage our clients to apply when evaluating any AI tool they integrate into their research workflows.
One of the most repeated phrases across the conference in sessions on education, enterprise AI adoption, and technology strategy was some variation of: "We combine our expertise with emerging technology to amplify humanity, not replace it."
In research, this principle is not optional. The entire value proposition of qualitative research rests on human judgment: the ability to design studies that capture nuance, to build rapport with participants so they share what they actually think, to interpret data in light of context that no algorithm currently has access to.
AI tools that try to automate this away do not just devalue the research, they undermine the business decisions made based on it. AI tools that augment it, by handling logistics, accelerating analysis, and flagging patterns for human review, make researchers more effective and research more credible.
The itracks platform was designed with this distinction at its core. Recruiting, scheduling, session management, transcription, initial analysis—all can be handled or significantly accelerated by the platform. But the itracks Realtime environment puts live, human-led conversation at the center of the research experience, with AI operating in support rather than in command.
itracks Fusion, which bridges quantitative and qualitative research in a single workflow, reflects the same philosophy: quantitative data surfaces patterns worth exploring; qualitative conversation reveals what those patterns mean. Neither replaces the other; the intelligence is in the combination.
A session on organizational AI adoption described a pattern that will be familiar to anyone managing research teams right now: some people sprint ahead, adopting every new tool immediately. Others ignore AI entirely. And some are frozen — unsure where to start or what to trust.
This fragmentation creates real operational risk. Research quality becomes inconsistent not because the tools are bad, but because different people in the same organization are using them in fundamentally different ways — or not using them at all.
The recommendation from that session was to define a clear AI vision and pair it with adoption tactics tailored to each group. For research organizations, that means: defining which parts of the workflow AI should touch and which it should not, establishing shared standards for how AI-assisted analysis is reviewed and validated, and creating the internal confidence that comes from seeing AI work well in practice rather than just hearing that it might.
itracks supports this through our Learning and Enablement services, designed not just to train researchers on the platform, but to help them develop a coherent approach to AI-assisted research that they can actually believe in and explain to their clients.
The conference left us with a renewed conviction about where the real differentiation lies in research technology right now. It is not the organization with the most AI features— it is the one whose clients can confidently explain what those features do, trust the outputs they produce, and make better business decisions as a result.
That is the product we are building. And the conversations happening in about AI right now suggest the rest of the industry is beginning to catch up to what good research technology has always known: intelligence without trust is just noise.