Do voters get influenced by manifestos?

The 2026 local elections in short #

What Ukweli did #

Ukweli analysed 15 manifestos for the 2026 local elections and found 2,686 promises, each quoted in the party’s own words with its page. This folio adds four steps on top of that work. It uses published counts and surveys, a Ukweli count of news headlines, Google search interest, 1,089 posts on X about South African politics sorted by two AI readers, and the results the IEC published.

The 15 manifestos

Promise counts are the numbers on Ukweli’s pages, re-derived from its ledger on 8 Nov 2026 (total 2,686). They are the sizes of the documents, not a measure of anything else. A manifesto is in Ukweli’s set when Ukweli can obtain it from the party’s own publication; that is a rule about documents, never a judgement about a party. Party names: ACDP is the African Christian Democratic Party; AMM the Afrika Mayibuye Movement; ANC the African National Congress; DA the Democratic Alliance; EFF the Economic Freedom Fighters; FF Plus the Freedom Front Plus; ICM the Independent Citizens Movement; IFP the Inkatha Freedom Party; NCC the National Coloured Congress; SACP the South African Communist Party; UDM the United Democratic Movement. ActionSA and GOOD are the full names. BOSA and Al Jama-ah are written as each party writes them.

The SACP did not publish a standalone manifesto document that Ukweli found; its manifesto text is printed in its newsletter, Umsebenzi Online (August 2026 issue), on pp. 7 to 11, and that is the text Ukweli read. Party logos belong to the parties and are shown for identification only. Ukweli is not affiliated with any party.

What Ukweli does not do #

• It does not score, grade or endorse any party, and does not say who should govern.
• It does not rank parties. Step 3 shows the results the IEC published, with party names, because they are public record. No party is named on any finding about manifestos.
• It does not say whether a promise is good, honest or will be kept.
• It does not say that a manifesto caused, or did not cause, any vote.
• It makes no statement about what any party knows or understands.

Step 1. What the news and the surveys were talking about #

Window: 8 May to 4 November 2026 (or six months). It is split into two parts. The first runs from 8 May to 6 October. The last runs from 7 October to 4 November. Each topic is compared with itself across the two parts, not with other topics. Some topic words are not plainly about the topic: a word may have other meanings, or be a name, such as “Madlanga” for crime. Where a topic has such words, the figure is also given without them, or the share they account for is stated.

What it shows. Water rose from 31% to 53%. Leaving out “rain”, 17 water headlines fall in the last part. Six of them are about water outages and failing water systems in Gauteng, including the South African Human Rights Commission’s report of 2 November. The rest are other water and sewage stories and opinion pieces. Johannesburg Water said high demand was limiting the recovery of its reservoirs. The Commission said the crisis was caused by systemic failures, not a shortage of water. We do not say that any of this moved a voter. Coalition moved from 47% to 50%. Crime, electricity and roads stayed close to level on the any-word measure. Unemployment, corruption, health, cost of living, housing and education fell. It does not show how many headlines there were on each topic overall, because each search contained the topic’s words.

How far each topic moved. The list below gives each topic’s move in percentage points, from the largest move to the smallest. A point is the gap between the two printed shares. Topics that moved by the same amount are in alphabetical order. Most searches found 31 to 57 headlines in each part, so one headline is worth about two to three points, and a move of two or three points is about one headline.

• Corruption fell 28 points, from 46% (16 of 35 headlines) to 18% (6 of 34).
• Water rose 22 points, from 31% (11 of 36) to 53% (20 of 38).
• Unemployment fell 20 points, from 53% (27 of 51) to 33% (15 of 46).
• Cost of living fell 19 points, from 51% (29 of 57) to 32% (14 of 44).
• Housing and land fell 17 points, from 52% (24 of 46) to 35% (16 of 46).
• Education fell 16 points, from 71% (27 of 38) to 55% (17 of 31).
• Health fell 12 points, from 23% (8 of 35) to 11% (4 of 37).
• Coalition rose 3 points, from 47% (17 of 36) to 50% (16 of 32).
• Electricity fell 3 points, from 48% (26 of 54) to 45% (21 of 47).
• Crime and safety fell 2 points, from 66% (82 of 124) to 64% (71 of 111).
• Roads fell 2 points, from 21% (7 of 33) to 19% (6 of 31).

Corruption, crime and education. Corruption had the largest move on the graph, and it was a fall. Crime and safety stayed level, at 66% and 64%. Its searches found 124 and 111 headlines, more than twice as many as any other topic’s searches. Crime had the longest word list of the eleven topics, and each search returns at most 20 headlines in each part, so a longer list finds more headlines. Crime’s search was split into 8 parts, against 2 or 3 for each of seven other topics. Plain crime words alone gave 32 and 30 hits, against at most 29 and 21 for any other topic. Names of people in the policing scandals, which were on the list, appear in 39 and 32 of the headlines found, about three in ten. The headlines found did not have to mention the election. We read the titles and Google snippets of the three crime searches that held these names. From 8 May to 6 October, 4 of 59 results held an election word. From 7 October to 4 November, 24 of 60 did, and 7 of 40 in the two searches that held only names. Most were Commission and court news, so many of crime’s hits are news about the scandal, not about the election. These counts do not show how much of crime’s lead is search size and how much is coverage of the scandal. Education fell from 71% to 55%, on 38 and 31 headlines. In the Afrobarometer survey of 2025, 18.6% of people named crime as the most important problem, the second most named. Education was named by 2.6%, the lowest of the nine problems in Graph 2. The same is true of every topic: the number of headlines found depends on how many searches we ran for it, so no count or share here puts one topic above another. Headline shares and survey figures measure different things and cannot be added together. These are shares of the headlines the searches found. They do not show how much was written on a topic, and we do not say why any share moved.

What the three show together. Water is in all three: a rise in headlines and in search in 2026, and the third most named problem in the Afrobarometer survey. Unemployment was the most named problem in the survey, while its share of the headlines found fell. The two measure different things and cannot be added together. We do not call one a check on the other.

Limits. Headlines show what the news printed, not what drove anyone’s vote. A search containing “election” returned some headlines that do not mention it (see how we counted). The survey was taken 16 months before polling day and before the campaign. Google Trends gives a relative index, not a count, and includes searches that are not about the election. Google builds its figures from a sample and adds small random noise, most noticeable for searches with little volume, so a repeat read can differ: our two reads of 3 November gave 78 and 77. The news count covers six outlets and the first two pages of Google results for each search; Google shortens long headlines. Some water news did not reach the count: reports from 7 October of an illness linked to contaminated tap water in Bronkhorstspruit, in Tshwane, were not on the first two pages of the water searches. Words in languages other than English were checked in isiZulu headlines (Isolezwe and Ilanga), where election headlines were mostly about the election itself, with few service topics. Words in Sesotho, Setswana and Sepedi were only partly usable. In the six-outlet count, Netwerk24 headlines are matched on English words only. A separate search used Afrikaans words for Netwerk24 alone: “verkiesing” with each topic’s words, the same two windows, and the first two Google pages for each search. It returned 397 headline results, and some appear under more than one topic. Each topic had 0 to 7 hits in a window, too few to read a trend, so no shares are printed. Water had 1 hit in the first window and 3 in the last. Crime had 7 in each. isiXhosa, Xitsonga, Tshivenda and siSwati are not covered by the data.

Step 2. Which manifestos carry those topics #

What it shows. Of the 2,686 promises Ukweli analysed, 123 are filed under water. Every one of the fifteen manifestos has promises filed under jobs, local economy and small business.

What it does not show. Coalition, cost of living and education, three of the eleven topics in Step 1, have no matching theme here and are not shown. A theme count says where a promise is filed, not what a manifesto contains. It does not say which party makes which promise: counts for each party are in Folios 21 and 22. It says nothing about how specific, funded or achievable a promise is.

Step 3. How each party’s share of the vote in the eight metropolitan councils compared with 2021 #

The IEC’s published results for the eight metros, read from its results dashboard on 10 November 2026, with the 2021 results beside them. Parties among the fifteen analysed: ANC, DA, ActionSA, EFF, UDM. Not among the fifteen: MK Party. Shares are the IEC’s own.

What it shows. In every metro the ANC’s share was lower in 2026. Which parties’ shares were higher differed from metro to metro: the MK Party’s was largest in eThekwini (38.29%), ActionSA’s in Tshwane (22.07%), the DA’s was higher in five metros, and the UDM’s rose to 18.39% in Buffalo City. National turnout figures are in Folio 24.

What it does not show. It does not say why any voter voted as they did. Nothing in this step is set beside the topics or the promises in Steps 1 and 2.

Step 4. Did voters know what the manifestos contain? #

The test (set before the data were opened): an average voter does not read a manifesto. The test is whether voters know of, or were exposed to, three to five promises on issues they care about, through media and other platforms, and whether those promises influenced their decision to vote.

Two kinds of evidence are kept apart. Knowing: voters themselves show they know or cite specific 2026 local-election promises. Exposure: specific promises were put in front of voters (news, search, campaign).

Knowing: what voters were asked. We read the summary of results of Afrobarometer round 10 for South Africa (interviews in 2025, summary published in May 2026). We also read the HSRC and IEC election satisfaction survey on polling day (13,198 voters; preliminary, unweighted results). We read three Ipsos South Africa releases and the public pages of the News24 and SABI poll. The poll interviewed 5,001 registered voters. Interviews ran from 25 August to 17 September 2026. None asks voters whether they knew a manifesto or a party’s promises. The HSRC asked what influenced the decision to vote, not which party to choose. Two of its listed reasons name a party or a leader (“give another party a chance” 26%, “like a particular leader or candidate” 22%), but the question was why people voted. It also asked when voters decided to vote, and 60% said more than six months before polling day. The News24 and SABI issue tables were behind a subscriber wall and were not read.

Knowing: what was posted on X. We collected the posts that X’s search returned and kept those about South African politics: 1,089 posts, from 565 accounts, dated 9 May to 4 November 2026. Each post was read and given one label under rules written before any post was read. Two AI readers did this; neither is a person. Posts are paraphrased here. No post is quoted, no account is named and no party is named. Accounts are counted once under each label they appear in, so the account counts below do not add up to 565. Left out: posts that were not about South African politics or elections (other countries, off-topic, spam) and posts dated after polling day.

• 491 posts, from 115 accounts, came from party, campaign, official, media or institution accounts. They show what was put in front of voters. They are not voters.
• 598 posts, from 455 accounts, came from individuals writing about South African politics. Of these:
  • 10 posts, from 6 people, named a specific promise from a 2026 campaign, for example free basic services, insourcing of workers, a minimum wage, a higher grant, or a pledge not to work with another party. Five of the ten came from one account on one day (a manifesto launch day) and read like campaign wording; the account name is not a party, so they are counted as an individual. One of the ten posts, from one person, tied a named promise to their own vote.
  • 11 posts, from 11 people, said they would vote, or judge parties, by whether a party kept its promises, without naming a 2026 promise. 4 of the 11 also said how they would vote.
  • 67 posts, from 63 people, recalled older broken promises, with no statement about how they would vote.
  • 284 posts, from 212 people, were about the 2026 local elections (campaigns, registration, candidates, how they would vote) and named no promise. 95 of them said how the person would vote.
  • 226 posts, from 201 people, were about South African politics with no link to the local elections.

The five groups add up: 10 + 11 + 67 + 284 + 226 = 598 posts from individuals. With the 491 posts from party, media and institution accounts that is 1,089 posts. Another way to read it: 88 posts from individuals named or judged promises. Of these, 67 recalled older broken promises. People on X are not typical voters, and these are the posts X’s search returned, not all posts (see How we counted, and what limits this).

How much attention the posts got. Likes, reposts, replies and views are the counts X returned with each post when we collected it. Views count how often a post was shown, not how many different people saw it. Across the 1,089 posts there were 82,346 likes and 22,251 reposts. There were 7,713 replies and 4.4 million views. Most posts drew little. The middle post (the median) had 9 likes and 563 views. It had 2 reposts and 1 reply. 468 posts (43%) had no replies and 304 (28%) had no reposts. A few posts drew most of the attention. The ten most-liked posts hold 38% of all likes. 144 posts had 100 or more likes, 8 had 1,000 or more, and none had 10,000. Six posts were viewed 100,000 times or more: four from party, media and institution accounts and two from individuals. The middle post from a party, media or institution account had 11 likes and 1,052 views. The middle post from an individual had 7 likes and 382 views. Of the 1,089 posts, 768 were original posts and 321 were replies. These counts show how much attention the posts got. They do not show who saw them, where those people live, or whether anyone voted differently.

Exposure: news. In the six outlets, 31 of 39 headlines in the first window carried a promise word. In the last window, 17 of 43 did. That window ran from 7 October to 4 November. The first window was mostly manifesto launches. The last window was more specific and topic-led, with several on water. Many headlines also framed promises with doubt. This shows that specific promises on voters’ topics were reported. It does not show that voters took them in.

Exposure: search. Google searches for “manifesto” in South Africa were close to zero on most days until August. From 8 May to 31 July there were 85 days. The index read zero on 81 of them and 7 or less on the other four. Google shows very low volumes as zero. On Google’s scale, the index read 26 on 2 November. It read 77 on 3 November. It read 100 on 4 November. We read these on 10 November, for the window 8 May to 4 November. Searches for a named party’s manifesto were small on the same scale, between 1 and 14 on 4 November. Manifesto searches peaked in the last three days. The HSRC found that 60% of the voters it interviewed on polling day said they had decided to vote more than six months before; it asked when they decided to vote, not when they chose a party. The two figures come from different measures and cannot be linked to the same people.

Reading. Exposure: promises on voters’ topics, several on water, were reported in the news, and searches for “manifesto” peaked in the last three days. Knowing: no survey we read asked which promises voters heard, so this folio has no measure of it. On X, 6 people named a specific 2026 promise. Where people on X did write about promises, most (67 of 88 posts) recalled older broken promises. By the reading rules set before the data were opened, this is the “exposure only” reading: influence not established. It does not say manifestos have no influence. It says this evidence cannot show it.

What earlier studies found #

Five published studies, each opened and read at its owner’s own site. Experiment studies were not opened and are not included. This is not a review of the research on voting. The Ghanaian and British studies are context from other countries, not evidence about South African voters. They show what voters elsewhere say when asked for their reasons, or whether they would read a manifesto. The studies ask different questions in different countries and years, so no figure here is set against another.

Asked voters for their reasons.

1. South Africa. University of Johannesburg, Centre for Social Development in Africa, fieldwork by Ipsos in the last quarter of 2023: 3,511 respondents, nationally representative, margin of error 1.8%. Asked why they would choose a party in the 2024 national election; each person could give up to five reasons. 49% gave “promises to improve people’s lives”. Other reasons given: 55% the party will create jobs, 48% improve services, 44% pay grants. The authors tested a related question with a regression, a statistical test of what predicts a choice. It found that naming any of three reasons (the party pays grants, promises to improve people’s lives, will improve services) did not predict whether a person chose the ANC or an opposition party. The test used 1,668 of the 3,511 respondents, because the others had missing answers. Its items were about what a party promises or does in general, not about a manifesto, and the election was national.
2. Ghana. Akeliwira and Owusu-Mensah, PanAfrican Journal of Governance and Development, 2022: a pre-election survey of 11,600 voters in 116 constituencies, chosen on purpose rather than at random. The question asked for the reason for voting for a party, and the paper ties it to Ghana’s 2012 and 2016 elections. 65.8% chose “liked the party program/manifesto”. 17.8% admired the presidential candidate. 8.3% said it is the party of their tribe. 4.3% chose “just voted”. 1.4% received a financial inducement. 2.4% chose “other”. The six figures add to 100%. It reports a stated reason. We found no question in it on whether voters knew what a manifesto contained.
3. Ghana. Afrobarometer round 7 took place in September 2017. Interviews ran from 9 to 25 September. It was a random sample of 2,400 adults, with a margin of error of 2 points. Asked the major reason for the choice of presidential candidate at the 2016 election, 28% of the 2,400 respondents said the campaign promises, policies and programmes of the candidate they voted for were more credible. 28% said the candidate could do a better job managing the country’s affairs. 7% named the same party and 1% the same ethnic group. 1% said the candidate was going to win and 2% chose other. 1% or less did not know or refused. 32% are marked “not applicable”. These are the survey’s weighted shares, not head counts.

Asked whether voters knew or would read manifestos.

5. Great Britain. MORI interviewed 1,960 British adults face to face. Interviews ran from 17 to 22 May 2001, before that year’s general election. Asked how likely they were to read or look at any of the party manifestos in the campaign, 39% said very or fairly likely (16% very, 23% fairly). Another 58% said not very or not at all likely. In April 1997, 53% said likely. 26% chose the statement that manifestos are “a waste of time and I wouldn’t pay any attention to them” (14% in 1997). It records how likely people said they were to read, not whether they did.

What they show together. Each figure answers a different question, so none is set against another. Where voters were asked for a reason, 49% in the South African study (2024 national election) gave “promises to improve people’s lives”. 65.8% in the Ghanaian pre-election survey of 11,600 voters named the party’s programme or manifesto. 28% of all 2,400 respondents in Afrobarometer Ghana (32% were marked not applicable) named the candidate’s campaign promises, policies and programmes. The South African study’s regression found that its group of three reasons (grants, promises to improve lives, better services) did not predict choosing the ANC or an opposition party. No study opened asks South African voters which promises they heard. The studies that ask about reading or knowing report a stated likelihood, an author’s own reasoning, or awareness of a government programme rather than of a manifesto.

Our view #

Ukweli’s view, kept apart from the counts above. Not a finding.

Do voters get influenced by manifestos? On this evidence we cannot say yes, and we cannot say no. Nothing we read measures influence. No survey we read asks South African voters which promises they heard. On X, 10 of the 598 posts from individuals named a specific 2026 promise, and one of those ten tied it to the writer’s own vote.

There is a twist. A party can write its manifesto around what voters already worry about. If it does, a voter who cares about water and reads a water promise has not shown that the promise moved them, because the worry was there first. A match between a promise and a voter’s concern is what we would see if manifestos move voters. It is also what we would see if manifestos follow voters. A match cannot separate the two, and both can be true at once.

The order of events is consistent with the second pattern. The Afrobarometer survey was taken in June and July 2025. Twelve of the fifteen manifestos were out, or announced for a set date, at least 39 days before polling day, and the earliest was launched on 27 June 2026. For the other three we could not establish a launch date. Voters’ worries were measured before the manifestos came out.

Immigration shows the same order. Google searches for “immigration” in South Africa peaked between 4 June and 7 July 2026. The average was 23 in the first window and 10 in the last. The scale is Google’s own, for that one search. Of the eleven manifestos with an exact launch date, ten came after 7 July. Folio 09 counted 16 promises about immigration. They sit in 7 of the 15 manifestos. Eight make none, so this is not a rule.

Cost of living had headlines in the Step 1 searches but no matching theme in Step 2. We searched all 2,686 promises for cost-of-living words. 16 promises in 7 of the 15 manifestos contain one, and the phrase “cost of living” occurs in three promises, all in one manifesto. A word search understates the topic: one manifesto has a section headed cost-of-living relief whose promises use none of the words. So we do not say that any manifesto makes no cost-of-living promise.

What this does not show. It does not show that any party read a survey or watched a search trend, and Ukweli makes no statement about what any party knows or intends. A party can follow its own timetable. The dates give an order of events, not a cause.

What would help is a survey that asks voters which promises they heard, and when they heard them. None of the surveys we read does.

What this folio does not show #

• It does not say manifestos influence voters, or that they do not.
• It does not say any topic, promise or party result caused another. Steps 1 to 3 are set side by side, and no line is drawn between them.
• It does not show what drove any vote. Headlines show what was loud; surveys ask what people say.
• People on X are not typical voters. The 1,089 posts are a slice of what X’s search returned, not all posts. 29 of the 1,089 are tagged by X as not in English, and their meaning was read by an AI reader and not checked by a speaker of those languages. For headlines, the languages other than English that were checked are limited (isiZulu headlines; Sesotho, Setswana and Sepedi in part; isiXhosa, Xitsonga, Tshivenda and siSwati not covered).
• The HSRC asked about reasons to vote and about when people decided to vote, not about which party to choose.
• The Afrobarometer survey was taken before the campaign; the news count covers six outlets.
• It does not judge the parties or say how any party’s manifesto compares with another’s.
• Our view rests on launch dates for twelve of the fifteen manifestos, on one Google Trends search (“immigration”, which includes searches that are not about the election), and on word searches of the promises. Launch dates for three manifestos could not be established.