How to keep human judgement at the heart of research

Insight gathering is always evolving, but it’s hard to recall a year so radical as 2025 in terms of the number of emerging trends embedding themselves so quickly.

Tech advances meant that regardless of how data is collected and intended to be used, there has been a shift in the way of working compared to a year or so ago.

Research has never been faster or more accessible, but AI and self-serve platforms create concealed traps along the way that, without the experience of an insight specialist to defuse, can easily result in conclusions and recommendations built upon unreliable data sets.

Insight gathering has got quicker, but at the cost of context.

Trends shaping insight gathering

The most impactful trends usually burn low in the background for a time, proving their usefulness and quietly embedding themselves as permanent fixtures.

Some of the changes that crept in recently include:

• The rise of short, punchy mobile surveys and the increasing need to develop question types that work best for mobile-first experiences.
• The increased understanding of consumer psychology and the implementation of frameworks such as behavioural economics, alongside a shift towards integrating system 1 and 2 thinking into research design.
• The move away from in-person qualitative research due to the pandemic, driven by a lingering desire for speed and convenience. Stories are increasingly gathered and told through screens rather than experienced in the field.

Faster ways to collect valuable data, combined with questions written to a framework that appreciates how the human mind processes the world, have emerged. These changes have resulted in quicker insight, but there is a potential loss of depth, context, and experience if not used carefully.

The rise of self-serve platforms

One trend that really pushed through recently is the explosion of self-serve platforms.

A decade ago, an ad hoc insight project followed a slow, familiar plod:

In the past couple of years, sophisticated self-serve platforms have torn up the old playbook. The ability to quickly get feedback unlocks the opportunity to design, test, and refine concepts rapidly, measuring key metrics in a few hours.

This shift has empowered non-researchers to gather their own data, meaning everyone on a project can get involved in understanding the audience's perspective. However, it also introduced a risk: speed without scrutiny. Launching a survey easily does not guarantee unbiased questions, the right sampling framework, or the ability to interpret results fully.

There’s a new level of risk here, which makes the role of an insight professional more important than ever, preventing bad research from happening quickly by educating those performing insight gathering on the risks that have appeared. It’s a good thing we have an experienced team aware of these pitfalls and capable of performing insight gathering with these considerations in mind.

Question design and non-biased tests are just one element to consider. Opening up the research opportunity across an organisation also opens the door for tricksters to seep in. Ghosts have become digital, and they have endless opportunities to haunt.

Spotting ghosts in the data

There was a time when fraudulent data was easy to spot. It used to be relatively simple to ensure data was of good quality for analysis. I began in research when the decade began with a lower level of fraud than now, and recall an occasion where a panel partner assured me a dataset had been quality-checked and cleaned. Not knowing much better, I jumped straight into the open-ended questions. I started scanning down rows, sense-checking for words that jumped out, getting a feel for the sentiment. All seemed fine initially… until I noticed oddities in the openings every few dozen rows:

HAMMERSMITH, GOLDHAWK ROAD, SHEPHERD’S BUSH, WOOD LANE, LATIMER ROAD, LADBROKE GROVE

“Huh, that’s strange. I wonder what else they say in later questions…”

EDGEWARE ROAD, BAKER STREET, GREAT PORTLAND STREET, EUSTON SQUARE, KINGS CROSS

Upon closer inspection, patterns appeared across dozens of respondents, hidden in a dataset of thousands. Lists of London Underground stations, horse racing line-ups, WW2 battleship names, and supermarket products were copied and pasted, sometimes with a token phrase like “I like it because…” at the start to hide the deception.

Grouping these respondents revealed shared characteristics: all clicking the top or third option at each question, artificial 25–34-year-olds, “neither/nor” on scales, and “I couldn’t say” on yes/no questions. Strange ghosts reciting strange rhymes, but easy to spot and remove once you know what to look for.

As time went on, analysis software evolved. Grouping those speeding through a 10-minute survey in 40 seconds or segmenting by top-and-third responses quickly found fraud for review and removal.

The AI revolution touches everything, including survey fraud

Fraud was always present, but the rise of generative AI has caused an explosion in incidents.

Modern bots are sophisticated:

• They read and interpret nuance across a series of questions
• They generate open-ended responses that make sense
• They vary response times and answer patterns to avoid detection

Old tricksters now generate new masks on the fly, blending into whichever audience we engage. All for 20p a go, paid to their survey account, but on a scale of hundreds or thousands of surveys a day.

Synthetic users and the future of insight

This knife-edge of AI ally vs AI agent continues to expand in 2026.

On one front, insight agencies spend millions battling AI entities entering surveys. On another, we’re welcoming AI to help us learn:

• Synthetic users: AI agents trained on verified data to simulate consumer reactions to new products and concepts.
• Digital twins: allow “what-if” scenarios at scale, with users who don’t suffer survey fatigue and without expending resources collecting first-step insight.

These tools are exciting, but the danger arises when human oversight is lost entirely. Bots trained on bot-generated data would degrade trust in synthetic methodologies. For now, we experiment, analyse, and wait to see the true impact.

Human judgment always matters more than technology

The past decade has shown that technology can give us more data, but it cannot automatically give us more truth or trust in what has been collected.

Looking forward, the most successful researchers and agencies will balance:

High-speed data collection with the best AI advances, for gathering, analysing, and preserving quality insights.

Final tips for reliable insight gathering

1. Perform ruthless data scrubbing – Move beyond speed traps and eyeballing responses. Use fraud detection that captures subtle behavioural signatures of modern AI bots.

2. Speed with caution – Self-serve platforms are agile, but ensure quality control prevents the fast delivery of poor insights.

3. Verify the source – Whether using human panels or synthetic data, always ask: What is the “origin”? Ensure data is anchored in verified human behaviour and check for AI tells.

Insight in 2026 is fast, sophisticated, and AI-driven, but human judgment is indispensable. Collaboration with experienced insight specialists (like us) is essential to navigate this evolving landscape safely and effectively.

Get in touch with our team to see how we can safeguard your insights, embrace innovation responsibly, and ensure your research delivers both speed and depth.