Most of the political ads Zambians saw can no longer be read

Of the 1,138 domestic political ads that reached Zambian users in the year before the vote, 68% have their creative content withheld in Meta's own transparency archive, overwhelmingly because the ads ran without the political disclaimer Meta requires (576 ads). The majority of paid political messaging that reached voters ran outside the platform's compliance regime, and the public record shows a removal notice where the message used to be. Where the text is missing for any other reason, the table says so.

The share of domestic political ads carrying a disclosed funding entity has climbed steeply through the campaign.

Whether this reflects stricter platform enforcement or better campaign compliance cannot be determined from the archive alone. Either way, the accountability record for the early campaign barely exists.

Comparing the 12 months before each of Zambia's last two elections shows paid social media moving to the centre of campaigning: domestic ad volume 3.0× higher than the last cycle, and funder disclosure up from 11% to 32%, though from a very low base. Who leads has not changed: the same campaign tops attributed advertising in both cycles.

Meta no longer serves creative content for pre-2023 ads, so the 2021 column covers metadata only: volumes, funders and spend, but no message text.

Among the 286 ads whose text survives in the archive from the 12 months before the vote (the same window as every chart on this page), the dominant theme is not an issue at all: it is the election itself (turnout, registration), whose reach dwarfs every other theme. Candidate personality is the second most common subject by ad count, but its reach is modest: personality ads are many and small. Among policy topics, education leads by reach and ad count alike. Health barely registers. Corruption is the one topic where attack ads outnumber promotional ones. In Zambian paid media it is used as a weapon more than a platform. Themes are ranked by estimated impressions: what people were actually shown, not how many ads were bought.

Estimated impressions per theme, range midpoints. Hover a bar for ad and attack counts. Full per-topic breakdown in the data section. Ads can carry up to three topics and contribute their full estimated impressions to each, so topic totals overlap and should not be added together.

How the campaign speaks and asks people to act

The same 286 analyzable ads also reveal rhetorical posture and what each ad asks the reader to do. Rankings use estimated impressions; ad counts are shown alongside.

Disclosed spending in this cycle is small and concentrated. Estimated USD-denominated spend over the 12 months before the vote (midpoints of Meta's reported ranges; hover a bar for the range and ad count):

Ads billed in CHF, CZK, EUR, GBP, INR, NGN, PHP, PLN, ZAR are excluded here so the axis compares like with like. Full advertiser detail, including all currencies and topic mix per actor, in the data section.

What "unattributed pages" contains

The unattributed row is an aggregate of unresolved advertiser pages, not a single spender and not evidence that the pages share control. These are its largest component pages, ranked by estimated USD spend and then impressions:

Who reached voters: a different ranking

Money is not reach: Meta's pricing varies, so the advertisers who spent the most and the advertisers whose ads were seen the most can differ. Estimated impressions over the same 12 months (midpoints of Meta's reported ranges). Impressions have no currency, so every advertiser is on one axis, including those the spend chart excludes:

Two deterministic checks run over the whole domestic corpus. They produce leads, not verdicts: every number below is reproducible by anyone holding the same public archive, and each has innocent explanations alongside the concerning ones.

Identical messages across distinct pages

One set of pages shares near-identical creative text:

• Salt Sana Movement and UPND Zambia: 2 messages in common across 6 ads, ~1.5M impressions

Funding names reused across pages

A shared funding name can reflect one organisation using several legitimate pages, or a relationship worth examining. It is a traceable link, not proof of coordination:

• UPND Zambia appeared on 129 ads across 2 pages: UPND Zambia, Hakainde Hichilema

Reach priced far from market

Cost per 1,000 impressions, page by page, against the median for pages billing in the same currency. Far-below-market reach can mean broad targeting or genuinely viral creative, but it is also the signature of engagement bait and inauthentic amplification. Far-above-market reach usually means narrow targeting.