Artificial intelligence is making fraudulent online survey responses harder to detect, creating a growing challenge for Nigerian businesses, researchers and institutions that rely on digital questionnaires to understand consumers and make decisions.
Research examining 31 different fraud-detection techniques found that no single method was reliable enough on its own to identify sophisticated fraudulent respondents. Instead, combining several signals produced substantially stronger results, highlighting the need for organisations to rethink how they verify online research data.
The study, published in Frontiers in Research Metrics and Analytics, tested 31 indicators and six combinations using 1,944 responses from two agricultural surveys conducted in California. Researchers found that the strongest combination detected 96 per cent of fraudulent responses with a 7 per cent error rate.
Brandspur Brand News reports that while the research was not conducted in Nigeria, its findings have clear relevance for a market where companies, universities, development organisations, pollsters and research agencies increasingly use online surveys to gather information quickly and reach respondents across different locations.
The problem is no longer limited to people rushing through questionnaires or repeatedly selecting the same answer. Artificial intelligence tools can produce convincing written responses, while organised fraudsters can imitate legitimate participants and attempt to bypass conventional screening measures.
Some of the traditional checks researchers have relied upon are consequently becoming less dependable. The study found that techniques such as open-ended questions, attention checks, geolocation and other standalone indicators can struggle when confronted with increasingly sophisticated fraud.
That matters because poor survey data can travel much further than the questionnaire itself.
A consumer goods company may use research findings before launching a new product. A bank or fintech may survey customers before changing a service. Telecommunications companies can depend on customer research to understand network experience, while government agencies, universities and development organisations routinely gather information to shape programmes and policies.
If a substantial portion of those responses comes from bots, duplicate participants or people pretending to belong to a target audience, the organisation could end up making decisions based on customers or citizens who effectively do not exist.
Nigeria’s research market is already showing signs of taking online data quality seriously. Ipsos, on its Nigeria-facing platform, describes a multi-layered approach that includes respondent verification, real-time detection of bot-like behaviour and checks for similarities between accounts. The company says introducing multifactor authentication reduced suspicious panellists by 50 per cent and recruitment through high-risk phone numbers by 20 per cent within its system.
Industry guidance from ESOMAR and the Global Research Business Network similarly recommends measures covering participant validation, prevention of duplicate participation, respondent engagement and transparency around online sample sourcing.
The emerging lesson is that researchers may need to treat fraud detection less like a single checkpoint and more like a collection of evidence.
Completion speed could be considered alongside device information, location, response patterns and previous participation behaviour. A fast response by itself may not prove fraud. Neither should passing one attention test automatically establish that a respondent is genuine.
Historical behaviour can add another layer. Someone who has participated reliably across numerous studies presents a different risk profile from an account repeatedly associated with suspicious responses, even if both appear normal in one questionnaire.
That approach could become particularly important in Nigeria as brands seek faster consumer intelligence and organisations increasingly turn to digital channels to reach respondents across the country’s 36 states and the Federal Capital Territory. Online panels already offer researchers nationwide access without requiring every interview to take place physically.
There is, however, a balance to strike. Aggressive fraud controls can also remove genuine respondents if researchers automatically reject everyone who triggers one unusual signal. Someone completing a questionnaire quickly, for instance, may simply be familiar with the subject.
The challenge is therefore not to build the strictest possible filter, but to gather enough independent evidence to distinguish suspicious behaviour from legitimate differences among respondents.
For Nigerian companies spending money on customer research, the issue ultimately comes down to confidence in the information reaching management.
Artificial intelligence can make surveys faster to create, distribute and analyse. It can also make fraudulent participation more convincing.
As both sides adopt more sophisticated technology, the value of research will increasingly depend not only on how many responses an organisation collects, but on how certain it is that real people are behind them.
