By Nco Dube | 03 September 2026

South Africa is heading into another election season in which opinion polls will be asked to do more than measure public sentiment. They will persuade, reassure, alarm and, inevitably, be weaponised. 

A poll will place the ANC lower than expected and its supporters will interrogate the pollster before they interrogate the result. Another will show the DA rising in Johannesburg and the party will circulate the number long before most people have read the methodology. MK will seize on a favourable KwaZulu-Natal finding as proof that the old order is still coming apart. The EFF will dismiss an unfavourable survey as establishment propaganda. Newspapers will turn movements within the margin of error into stories of surges, collapses and political earthquakes.

The lesson of the 2024 election is not that polling failed. It is that polling requires reading. The better pollsters were, in broad terms, remarkably close to the final result: ANC 40.18%, DA 21.81%, MK 14.58% and EFF 9.52%. There were misses. Some surveys were conducted too early to capture late movement, turnout assumptions distorted others, and MK rose so quickly that numbers which had looked reasonable a few weeks earlier aged badly. There were also legitimate questions about independence.

None of this makes polling useless. A poll is a measurement taken at a particular moment, from a particular sample, through a particular method, with assumptions built into it. It is evidence. The politics begins with what follows.

That becomes even more important in the 2026 local government elections because local elections are harder to poll than national ones. South Africans will not vote in one giant municipal contest. They will vote in Johannesburg and Jozini, Tshwane and uMhlathuze, eThekwini and Emalahleni. They will vote in wards where personalities count and local conditions can outweigh national loyalties. A voter may split the ward and PR ballots, while someone who voted enthusiastically in 2024 may simply stay home in 2026.

This is why one of the most seductive figures of the coming campaign may also be one of the easiest to misuse: the national local-election poll. It tells us something, but it does not tell us where power will fall. If a survey places the ANC at 34% nationally and the DA at 27%, the political temptation is immediate: the gap is closing, the race is tightening, history is turning. Perhaps. But before poetry becomes headline, there should be arithmetic. Who was interviewed? Registered voters or likely voters? How was likelihood to vote determined? Was the survey face-to-face, telephonic or online? 

How were undecided voters treated? What turnout assumptions were built in? And what does a national percentage actually tell us about Johannesburg, Tshwane or eThekwini?

A poll is not the electorate. It is a photograph of the electorate taken through a particular lens. Much of the political mischief begins after that. Margins of error, for instance, are routinely pushed to the footnotes when they should sometimes sit closer to the headline. A party moving from 25% to 27% between surveys with margins of error of two or three points may not have moved at all. Yet the verbs attached to those figures can manufacture movement where the statistics do not: one party is “surging”, another is “bleeding support”, one is “closing the gap”, another is “in freefall”. The numbers may have barely shifted. The politics often lies in the sentence wrapped around them.

This is how a poll becomes a political instrument without the poll itself being corrupted. A favourable survey is published. A party extracts its strongest number. A graphic appears before the methodology has been read. Supporters circulate it, commentators pronounce momentum, donors see viability, and rivals are forced to respond to a narrative they did not create. Research intended to measure political reality begins to alter the political environment it was measuring. No fraudulent spreadsheet is required. Interpretation can do the work.

South Africa’s polling debate too often collapses two separate questions into one: was the research competently and honestly conducted, and how is that research being politically used? Those questions can overlap, but they are not interchangeable. The controversy around the Social Research Foundation and Victory Research is a useful example. SRF’s final polling before the 2024 election was good. Its last tracker placed the DA almost exactly where it eventually finished and came reasonably close on the ANC, MK and EFF. It would be intellectually lazy to discard the work simply because critics dislike the institutions associated with it.

But it would be equally lazy to pretend those associations are irrelevant. Victory Research, which conducted fieldwork for SRF, had a founding managing partner, Johan van der Berg, who simultaneously served as the DA’s head of research during the 2024 election period. Gareth van Onselen had previously worked for the DA. Those facts do not prove that a single result was manipulated, but they do create a legitimate independence question. Accuracy does not erase it. Research can be statistically sound while the institutional arrangements around it remain uncomfortable.

The answer is transparency. If a media organisation or politically connected foundation funds a poll, say so clearly. If a polling company works privately for political parties while publishing public polling, disclose it. “Internal polling” should not acquire the aura of independence merely because a newspaper republishes it. Disclosure does not discredit research; it gives the public enough information to weigh it properly.

There is an equal danger on the other side. Concerns about independence can become a refuge for political parties trying to escape inconvenient evidence. The ANC would be foolish to dismiss alarming Johannesburg polling simply because it distrusts the institution producing it. The rational response is replication against by-elections, previous voting patterns and other reputable polls. Likewise, the DA would be foolish to treat favourable polling as a deed of transfer. A survey showing the party above 40% in Johannesburg or Tshwane is evidence of opportunity, not ownership. Voters remain inconveniently sovereign.

Turnout complicates the picture further. Municipal participation is lower and more uneven than national turnout. A party may enjoy broader passive sympathy and still lose to an opponent whose supporters are simply more likely to vote. Two reputable pollsters can therefore survey broadly similar electorates and produce different estimates because they make different judgments about who will actually vote. Neither has to be dishonest, but the assumptions need to be visible. “Likely voter” should always invite another question: likely according to what?

Every serious public election poll should prominently state who commissioned and paid for it, who conducted the fieldwork, how respondents were selected, how undecideds and turnout were treated, the margin of error and any material political relationships.

The media has obligations too. A statistically insignificant shift should not be promoted into a political trend merely because “nothing happened” makes a poor headline. A national poll should not casually become a prediction for Johannesburg, and a small metro sample should not be discussed with the certainty of a large national survey. No single poll should carry more authority than it deserves. The most persuasive thing in polling is rarely one spectacular result. It is convergence. If several credible organisations independently begin finding broadly similar movement in Johannesburg, Tshwane or eThekwini, confidence in the trend should rise. If one organisation discovers an electoral earthquake while everyone else records only tremors, caution is simply good statistical hygiene.

South Africa does not need fewer polls going into 2026. It needs more credible pollsters, more competition, more disclosure and more independent replication. Polls can expose both governing parties and oppositions to realities they would rather not confront. But they should never become substitutes for the electorate.

A poll is not a vote already cast, nor a promise of a vote still to come. It is a measurement of a moving public before the public finally speaks for itself. So when a party celebrates a favourable survey or denounces an unfavourable one, the rest of us should resist being recruited into either reaction. Read the sample, the dates and the turnout assumptions. Ask who paid and who conducted it. Compare the result with other polls, then watch what political actors choose to emphasise, omit or exaggerate.

The poll may only be holding up the mirror. The politics begins with who points at the reflection, and what they tell us we are seeing.