Why an online poll is worthless — and how we count voices here

Everyone has seen an online poll announcing that 87% of people believe something. Everyone vaguely senses the number is worthless. Few people know exactly why — which is a shame, because the reasons are precise, long established, and they determine what a counter can honestly be made to say.

This guide sets out the four main flaws of internet voting, then what we do about each. Including where the answer is "not much".

1. Self-selection, the flaw no technique fixes

A professional survey questions a constructed sample: people are chosen, then results are weighted by age, sex, region and occupation, producing an approximate picture of the population. An online vote does the exact opposite: people choose themselves. Those the subject bores do not vote, so what gets measured is the population of people who felt like voting, which has no reason whatsoever to resemble the general population.

This bias is structural: no amount of volume or engineering corrects it. A million self-selected votes are worth less than a thousand sampled ones, and a counter reaching enormous figures is not more representative — merely bigger. That is why we never present a result as "what the country thinks": we present what the people who voted here said.

That does not make the figure useless, provided you know what it measures. On a debate where both sides are equally motivated, the split is interesting. On one where a side is militant and the other indifferent — the case with most regional rivalries — the gap measures fervour, not demographics. We say so on the pages concerned, because it is the only honest way to publish that kind of number.

2. Multiple voting, and why blocking it entirely is impossible

The second flaw is more mundane: one person can vote several times. The standard defences are all circumventable. IP address? Thousands of legitimate mobile subscribers share one because of carrier-grade address sharing, so blocking it punishes innocents in bulk — and flight mode changes it anyway. Cookies? They get cleared, and a private window produces a fresh one. Mandatory accounts? They drive away the great majority of visitors and stop nobody from creating three e-mail addresses.

What we apply is a layered defence, no layer of which is perfect. Every vote consumes a single-use token, issued in the page, valid for fifteen minutes and consumed at the moment the vote is recorded: replaying a request or looping a thousand votes on one token is impossible. The free voice is then capped at one per device per day, guaranteed not by a check but by a database constraint that physically refuses the second row. Finally, inhuman cadences are spotted afterwards and the voices involved are removed from the scores.

That leaves the determined visitor who clears their cookies and starts again. They exist, we know it, and we made an explicit choice: not to turn the site into an identity checkpoint to stop them. The reason is set out below.

3. Brigading, or an organised community arriving

This is the spectacular one: a forum, a chat server or an account with a hundred thousand followers decides to make one side win, and the ranking flips overnight. Anyone who has run an online vote knows it, fan communities above all, capable of coordination that would embarrass a political party.

Technically these votes are genuine: real people, voting once each, for what they want. There is nothing to cancel. What brigading distorts is not the counting but the interpretation: a result produced by a sudden mobilisation says nothing about majority opinion, it says a community organised itself. That is why we record lead changes with their timestamps, and keep each match's daily history: an abrupt swing shows up in the curve, and the curve stays visible.

4. Bots

The last flaw is the simplest to describe and the most technical to handle. A script can cast thousands of votes in seconds. Single-use tokens remove most of that, since obtaining one requires loading a full page. Beyond that, a rate limit kicks in on bursts, and a verification challenge can be triggered for suspicious requests only — never for everyone, because putting a test in front of every vote loses more sincere voters than it stops bots.

What we refuse to do, and why

There are far more effective ways to guarantee one vote per person: demand a phone number, require identity documents, or fingerprint browsers across sites. They work. We do not use them, and that is not decorative scruple.

The reason lies in what is at stake. On this site, winning pays nothing: no money, no prize, no draw. The expected gain from cheating is therefore close to zero, while the cost of heavier surveillance — collecting identity data to settle an argument about bread — would be wholly disproportionate. The level of control should match the stakes, and here the stakes are pleasantly trivial.

What we will never do

One final point, the most important, and the one that separates this site from a magazine poll: no counter is ever seeded. No score has been inflated, rounded up or warmed up, including at launch, when every match showed zero and the home page looked empty for it. A debate showing few voices has received few.

That is a real constraint for a site starting out — a counter at zero is far less inviting than one at ten thousand — and it is exactly why so many sites cheat. We would rather start slowly than publish a false figure: a counter known to have been seeded once means nothing ever after.

Further reading: the literature on bias in self-administered online surveys (self-selection, non-response) is long-standing and accessible; carrier-grade address sharing by mobile operators is described in the public technical documents of internet standards bodies.