A.I.: The Good, The Bad, The Ugly

This article is an adaptation of a presentation given to Cox Research, reproduced here with their permission.

Artificial Intelligence is no longer a future concept in market research—it is now embedded infrastructure. What was once experimental has crossed what many describe as the “curiosity threshold,” with global AI investment reaching approximately $1.5 trillion⁽¹⁾ and organizations rapidly integrating it into core workflows. In research specifically, AI is reshaping how data is collected, analyzed, and interpreted. Yet, as with any transformative force, its impact is not uniformly positive. The reality is more nuanced: AI is simultaneously a force multiplier, a source of distortion, and a potential threat to trust itself.

The Good: A Force Multiplier for Insight

At its best, AI enhances the efficiency, scale, and capability of research. It “improves the plumbing” of the insights process—automating repetitive tasks, accelerating analysis, and enabling researchers to focus on higher-value thinking. Tasks such as survey design, coding, thematic analysis, and reporting can now be completed in a fraction of the time. Qualitative research, traditionally limited by cost and scale, can be conducted simultaneously across markets, generating richer datasets more quickly.

Adoption figures reinforce this shift: around 78% of organizations now report using AI, with 71% deploying generative AI in at least one business function⁽²⁾. In research, this translates into faster iteration cycles, more agile experimentation, and the ability to simulate scenarios using synthetic data.

Synthetic data, in particular, offers significant advantages. It enables privacy-safe analysis, supports compliance with regulations such as GDPR, and allows us to generate rare or sensitive scenarios that would be difficult to capture with real-world data. It also accelerates innovation by enabling rapid testing and iteration without the constraints of traditional fieldwork.

Ultimately, the promise of AI lies in augmentation, not replacement. When used correctly, it enhances human capability—freeing researchers to focus on asking better questions, interpreting meaning, and delivering strategic value.

The Bad: Scaling Without Transformation

Despite widespread adoption, the benefits of AI remain uneven. Many organizations are experiencing only modest gains—often less than 10% cost reduction and under 5% revenue growth⁽³⁾ —because adoption has outpaced true transformation . In other words, companies are adding AI tools without rethinking the processes they are meant to improve.

This creates a dangerous illusion of progress, as Martin Sjoorda has written: “The Efficiency Fallacy”. Pilot projects are abundant, but end-to-end redesign is rare. Integration challenges, fragmented data ecosystems, and organisational resistance continue to limit impact . Worse still, there is often confusion between using AI tools and fundamentally changing business models.

The research sector also faces a deeper conceptual risk: the conflation of efficiency with effectiveness. Faster outputs do not necessarily lead to better decisions. AI can produce more data, more quickly—but without critical thinking, this can simply result in more noise. True digital transformation enables a company to shift focus from speed to value, from outputs to outcomes, and from automation to augmentation.

Synthetic data illustrates this tension clearly. While powerful, it comes with significant limitations. It can amplify biases present in original datasets, struggle to replicate real-world complexity, and create validation challenges—the so-called “sim-to-real gap”. If synthetic data systems are asked to explore beyond observed (or more precisely, ‘learned’ data),  they become unreliable, as they cannot represent genuinely unseen causal mechanisms, and cannot anticipate discontinuities.

Overreliance on synthetic inputs risks distancing research from reality rather than enhancing it.

One dimension that is often overlooked in discussions of AI is its environmental footprint. The computational intensity of modern AI systems comes with significant energy and resource demands. Data centres powering AI are now consuming 1.5% of global electricity, equivalent to the annual electricity demand of the Netherlands, with usage growing at an estimated 20–30% annually.

Training a single large model can require vast amounts of water for cooling—reportedly equivalent to a substantial portion of an Olympic-sized swimming pool—and contributes to rising carbon emissions, resource extraction, and electronic waste.

In a sector increasingly focused on sustainability and responsible business practices, this presents a paradox: the same tools enabling faster, smarter insights are also adding pressure to already strained environmental systems. As AI becomes embedded infrastructure, its environmental cost must be treated not as an externality, but as a core consideration in how research is designed and delivered.

The Ugly: Trust Under Threat

The most serious implications of AI in research lie in its potential to undermine trust. As AI-generated content becomes more sophisticated, distinguishing between authentic and synthetic data is increasingly difficult. This is not a theoretical concern—AI-related incidents are rising rapidly, with reported cases exceeding 233 in 2024 and already surpassing that figure by the third quarter of 2025⁽⁴⁾.

In the research context, this manifests in several ways. First, impersonation technologies such as voice cloning enable fraud at scale. Second, misinformation ecosystems pollute the data environment, making it harder to identify reliable signals. Third, synthetic data—if not properly managed—can contaminate datasets, creating feedback loops of artificial insight .

The consequence is profound: authenticity becomes the central risk. If stakeholders cannot trust the origin, integrity, or validity of data, the entire research process is compromised. This challenge is compounded by broader societal risks. AI is already implicated in deepfake abuse, wrongful arrests due to flawed algorithms, and election interference. These are not isolated issues—they represent systemic vulnerabilities that extend into the research ecosystem.

Navigating the Future: Integrity as Advantage

In this complex landscape, the research industry’s competitive advantage will not be technology alone—it will be trust. The principles required to maintain that trust are clear: provenance, transparency, and accountability.

Researchers must ensure they know where data comes from, obtain proper consent, clearly disclose synthetic content, and maintain audit trails that preserve methodological integrity .

Human oversight remains essential; AI outputs must be interpreted, challenged, and validated by skilled professionals… we should not talk about a “human in the loop”, which brings a connotation of the human being secondary; rather we should reference “A.I. in the loop” !

The emerging model for research reflects this balance: Human + Machine + Meaning. In this framework, AI handles scale and speed, while humans provide judgement, context, and ethical oversight. As great minds such as Simon Chadwick and David Smith have been saying for some time, the role of the researcher must evolve from data collector to insight curator—someone who synthesises information, ensures quality, and designs decisions rather than merely describing reality.

Practical steps are already emerging as best practice. Organisations should firstly audit their processes, then automate the relevant routine ones; protect the evidence chain, disclose synthetic data transparently, invest in AI literacy, and publish clear policies on AI use. These are not optional safeguards—they are foundational requirements for operating in an AI-driven environment.

Nor is this rocket science, or having to create something new….as an example, an international group of practitioners monitor and curate the ISO 20252 quality standard (designed specifically for the research & insights sector), which now includes clear policies on the use and communication of A.I. technologies; these are supported by guidelines issued by many industry bodies. The work has been done – the challenge now is to use it!

Conclusion

AI in research is neither wholly good nor bad—it is amplifying both strengths and weaknesses. It offers unprecedented opportunities to enhance insight generation, but also introduces new risks that challenge the very foundations of trust and validity.

The key distinction will not be who adopts AI fastest, or who promotes loudest, it will be who can prove they use it most responsibly. As the technology becomes ubiquitous, differentiation will come from integrity, transparency, and human judgement.

AI may amplify insight—but it is integrity that will make it irreplaceable

References

⁽¹⁾ Gartner 2025 AI spend

⁽²⁾ McKinsey State of AI 2024

^{(3) &(5)} Stanford AI Index 2025

⁽⁴⁾ https://www.linkedin.com/posts/martijnsjoorda_the-efficiency-fallacy-why-executive-teams-activity