How to Read an Opinion Poll Before You Share It

Before you share a poll, find six things: who conducted it, who paid for it, when the fieldwork was done, how many people were asked and how, the margin of error, and the exact question. If any of them is missing, treat the graphic as an advert. Kenya has already had a fabricated Ipsos poll in 2026, and fake graphics carrying the Mizani Africa name.

Check one: does the pollster say it is theirs?

On 22 September 2026 PesaCheck rated as fake a graphic that claimed to be an Ipsos Kenya poll showing James Gakuya (Embakasi North) leading the Nairobi governor race at 52.56 per cent. It had been going round since 26 August. Ipsos said it had run no such poll.

Mizani Africa warned in April 2026 about fake graphics using its name, and Africa Check has published fact-checks on Mizani-branded fakes.

The check takes two minutes. Go to the pollster's own website or verified social accounts, and if the release is not there, do not forward it. The two decimal places in 52.56 were part of what made the fake look real.

Check two: when was the fieldwork?

A poll describes the week it was taken. Publication can come much later. TIFA's "wrong direction" figures, reported by The Star on 9 September 2026, came from fieldwork between 13 and 22 June 2026 with 2,048 adults in 47 counties. That release told readers about opinion in June. The Ol Kalou by-election came after it, as did the confirmation of Sifuna's removal and the Nyanza tours.

Write the fieldwork dates next to any figure you quote. Missing dates are a warning in themselves.

Check three: how many people, and who were they?

Three releases from 2026 show how much the basics vary:

"Adults" and "registered voters" are different groups of people. Mizani's Nakuru governor poll (fieldwork 6 to 18 July 2026) sampled 1,640 people from the 2022 register by mobile phone. A phone poll only reaches people who own phones and pick up, and a good pollster says as much in the release.

Bobea gave no sample size at all. You can still quote it, as long as you say what is missing.

Check four: is the lead bigger than the margin?

Mizani's August Nairobi poll put Babu Owino (Embakasi East) at 28.4 per cent and Agnes Kagure at 27.2 per cent, with a margin of plus or minus 2.3 points. The gap is 1.2 points. That margin applies to each figure separately, and the uncertainty on the gap between two candidates is larger again, so the poll has the two of them level.

Polls also disagree for ordinary reasons. Johnson Sakaja (Governor, Nairobi) was at 10 per cent in the Centre for African Progress poll of April 2026, 17.0 per cent in Mizani's August poll and 23.2 per cent in Bobea's. None of the three used the same method or dates. Look at where several polls point before trusting any single figure. Our note on reading a margin covers the same trap in election results.

Check five: the question and the undecideds

Bobea had 17.7 per cent undecided in its presidential question, and Mizani's Nakuru poll had 11.5 per cent. When the undecided share is bigger than the lead, nobody is ahead in any sense that matters.

Then read the wording. "Who would you vote for if the election were held today?" is a voting question. "Who is best placed to fix the economy?" is something else, and a campaign that quotes it as though it were a vote count misleads its own supporters.

What the law says about publishing polls

The Publication of Electoral Opinion Polls Act, 2012 governs polls published in Kenya. It requires a published electoral poll to carry details of who conducted it and who paid for it, the sample and the margin of error. It also bars publication of electoral polls in the five days before an election. A made-up poll graphic that carries none of those details is therefore open to a complaint from any rival who spots it.

A rule for your campaign

Put it in writing for the team. No poll goes into a campaign group, a speech or a poster until someone has found the pollster's own release and noted the fieldwork dates, sample and margin. Our explainer on the Infotrak figures walks through one national release in full, and the guide to deepfakes on the 2027 ballot deals with fabricated material more widely.

Internal polling is a separate matter. Commission it properly, keep it inside the campaign and use it to decide which wards get the next visit and the next shilling.