Why You Should Be Skeptical About Popular Opinion

I have been attending some Bangalore civic forums recently. In person meetings, neighbourhood discussions, planning debates. Bangalore is probably the most civically engaged city in India. More educated residents, more organised communities, more people who genuinely care about how the city functions than almost anywhere else in the country.

As someone who works in analytics, my instinct in any of these discussions was to look for data. What does the evidence say? Where is the research? What are the numbers? And what struck me, consistently, was how little of that existed. Decisions were being shaped by anecdotes, viral incidents, and whoever happened to be most organised and most persistent in the room.

But the more I sat in these discussions, the more a different pattern started to bother me. The loudest voices were not always the most informed ones. And they were almost certainly not representative of the city. Yet they were setting the agenda. I started to wonder whether civic energy, without the right infrastructure behind it, might actually be making things worse. And by infrastructure I do not mean roads or metro lines. I mean the analytical infrastructure: a rigorous framework for deciding which problems matter most, which interventions produce the most impact for the most people, and how to weight the signal from a vocal minority against the needs of a quiet majority. Without that, civic energy does not produce better decisions. It just produces louder ones.

Then I realised this was not a civic problem. It was everywhere.

Popular narratives gain dominance not because they are the most evidence backed but because they are the most repeated. A vocal group, an influential figure, a well timed piece of media, and suddenly a view achieves critical mass. At that point it stops being evaluated on its merits. It just becomes the default.

Research on public participation consistently shows that the people who dominate forums, fill out surveys, and shape policy discussions are systematically different from the broader population. More educated, more economically comfortable, more likely to have strong preexisting views. A 2019 study in Urban Affairs Review found that participatory planning processes in major cities consistently overrepresented higher income residents while underrepresenting renters, migrants, and lower income communities, even when those processes were explicitly designed to be inclusive.

The same dynamic plays out in almost every domain. The people loudest about a supplement, a fitness trend, or a business strategy are rarely representative of the people affected by it. But their volume creates the impression of consensus. And consensus, even manufactured consensus, is persuasive.

How the effect of this spans across personal lives to firm level decision making

NMN became one of the most discussed longevity supplements of the last five years largely because one prominent researcher talked about it enough times on enough podcasts. A 2024 systematic review and meta analysis found that while NMN does raise NAD+ levels in the blood, most clinically relevant outcomes showed no significant difference between NMN and placebo groups. The evidence base is preliminary, small, and in several cases at high risk of bias. The gap between what the loudest advocates claim and what the science currently supports is considerable. Most people taking it are not reading the meta analyses. They are listening to the podcasts.

Hyrox is a more interesting case. As a format for getting people moving it has genuine merit. But the cardiac evidence on chronic high intensity endurance training is worth knowing about. Research published in JACC Clinical Electrophysiology points to a J-shaped curve: moderate exercise is strongly protective, but sustained high intensity endurance load is associated with increased risk of atrial fibrillation through atrial stretching, fibrosis, and autonomic imbalance. A meta analysis of six case control studies found athletes had roughly five times higher odds of developing AF compared to non-athletes. None of this is in the Hyrox community conversation, because the community looks fit and the vibe is good and nobody wants to be the person asking uncomfortable questions.

The Institutional Level

It was around the same time that I started noticing something similar playing out at a much larger scale. Firm after firm was restructuring operations, reallocating headcount, and rewriting strategy documents around AI. Some of it was clearly warranted. But a lot of it had the same texture as everything else I had been observing: decisions being driven more by the volume of the conversation than by a rigorous evaluation of what actually made sense for that specific business.

In pharma, the GLP-1 gold rush is the cleanest current example of an industry converging on a dominant narrative faster than the evidence warrants. After the success of Ozempic and Wegovy, virtually every major pharmaceutical company pivoted significant R&D resources toward GLP-1 or adjacent metabolic mechanisms. Deloitte's 2025 analysis of the top 20 pharma companies by R&D spend found that obesity assets now account for 25% of total forecast sales in the late-stage pipeline, overtaking oncology for the first time in 16 years. GLP-1 related assets are so dominant that when they are stripped out of the analysis, the industry's underlying rate of return drops from 7% to just 2.9%. The headline looks like a boom. The subtext looks like a bubble.

The risk is not that GLP-1 drugs do not work. They clearly do. The risk is that an entire industry is crowding into one validated mechanism while the rest of the therapeutic landscape gets underinvested. History in pharma is not kind to monoculture thinking. The companies that win the next decade will probably be the ones that resisted the narrative long enough to find the next one.

The most dramatic version of this story happened in commercial real estate. Between 2016 and 2019, WeWork's $47 billion valuation sent a signal across the corporate world: the era of long-term leases and owned office space was over. Companies restructured their real estate strategies around it. Shorter leases, abandoned headquarters, entire facilities teams reorganised around the assumption that flexible workspace was the future. The signal was not a business. It was a story built on metrics that obscured a company losing $219,000 every hour. When the IPO prospectus arrived and analysts finally read the numbers underneath the narrative, the valuation collapsed by over 90% within months. The industry had reorganised itself around the loudest number in the room without asking whether the number reflected anything real.

The Bias Does Not Stop at the Collection Stage

You might assume that once data enters the picture, things get more objective. They do not. The bias toward the vocal starts at data collection itself. Surveys get higher response rates from people who are already engaged and already motivated to push a particular view. And even when analysts receive the data, confirmation bias compounds the problem. A 2021 paper in Nature Human Behaviour found that analysts given identical datasets but different motivating hypotheses produced significantly different conclusions. Not because the data was ambiguous. Because their framing shaped what they looked for. The numbers change. The distortion persists.

What to Do About It

The remedy is not to distrust popular narratives by default. Some popular things are popular because they work. The remedy is to hold them to the same evidentiary standard you would apply to anything else.

Ask what the evidence actually shows, not what the loudest advocates claim it shows. For NMN that means the meta analyses, not the podcasts. For any trend it means going back to the primary sources and asking who funded the research and what their prior was. Advocacy organisations, industry bodies, and influential individuals all have institutional interests. That does not make them wrong. It means their output should not be treated as neutral data.

Weight hard outcomes over reported experiences. Mortality rates, economic survival statistics, measurable clinical endpoints: these are harder to distort than surveys, sentiment scores, and engagement metrics. WeWork's valuation looked compelling right up until the IPO filing. The GLP-1 pipeline looks like a boom right up until you strip out two mechanisms and watch the return fall to 2.9%. Always ask what the hard number underneath the narrative actually is.

Apply the falsifiability test to any claim you are about to act on. What evidence would change your mind? If the answer is nothing, that is not a conviction based on evidence. That is a belief dressed up in the language of one. Narratives that can absorb any counter evidence without updating are no longer empirical claims. They are ideology, and should be discounted accordingly regardless of how frequently they appear or how many credible names are attached to them.

Disaggregate always. The people loudest about a trend are rarely the median case. A supplement validated in middle aged Japanese men may not transfer to a 28 year old in Bangalore. A fitness format designed for competitive athletes may not be appropriate for someone managing chronic work stress and limited recovery time. Aggregate enthusiasm almost always masks enormous variation underneath.

Treat silence as data. If a counterargument has almost no presence in the conversation, ask why. Research gaps are not evidence that the problem does not exist. They are evidence of where attention and funding have not gone. The communities not showing up to the forum, the therapeutic mechanisms not getting funded, the business risks not being discussed in the board presentation: their absence from the conversation is itself a signal worth investigating.

Remember that elite is not median. The people setting the narrative are almost never representative of the population it affects. A prominent researcher taking NMN is not the median ageing adult. A pharma board chasing GLP-1 deals is not representative of the patient populations who are actually searching for treatments. Always ask whose experience is generating the signal and whether that experience generalises.

The goal is not scepticism for its own sake. It is to ensure that when you do act on a narrative, you have actually evaluated it rather than simply absorbed it. Popular is not the same as true. Frequently cited is not the same as validated. And the loudest voice in the room is very rarely the most informed one.