When Public Opinion Is Manufactured

Kenya has spent years learning to distrust what politicians say.

It may now have to learn to distrust what apparently ordinary citizens say online.

A newly disclosed influence operation used artificial intelligence (AI) to generate batches of political posts explicitly designed to look like spontaneous grassroots opinion. The messages praised Energy Cabinet Secretary Opiyo Wandayi and promoted narratives suggesting that the United Opposition was fracturing ahead of 2027.

There is no evidence that the Kenyan government organised the operation. Anthropic, the company behind the Claude AI system, did not identify the organisation that commissioned it and explicitly found no evidence of government involvement.

That distinction is essential.

But the significance goes beyond attribution. Artificial intelligence dramatically reduces the cost of manufacturing the appearance of public enthusiasm, anger or consensus.

The documented network itself was not successful. Anthropic classified it as Category One on the Breakout Scale: its activity remained inside its own inauthentic network on a single platform and did not measurably reach real audiences.

That makes the episode less frightening as an accomplished campaign — and more useful as a warning.

In an election already likely to revolve around economic frustration and political trust, Kenya now faces another question: How can citizens judge public opinion when the appearance of public opinion can be industrially produced?

Fifty voices. One operator.

The mechanics were surprisingly simple.

Anthropic says one actor supplied topics, political talking points and hashtags, then used Claude across several sessions to turn them into batches of exactly 50 social-media posts. The prompts explicitly asked that the posts look like organic grassroots commentary rather than a coordinated campaign.

One part of the operation praised Energy Cabinet Secretary Opiyo Wandayi over an electricity-tariff issue, using hashtags including #PowerReliefKE and #PoweringTheNewKenya. Another promoted claims that the United Opposition was disintegrating, targeting former deputy president Rigathi Gachagua and former president Uhuru Kenyatta.

The same workflow was also used for ordinary retail marketing under a commercial persona. That detail matters because it shows that the political method did not require a specialised propaganda apparatus. A content-marketing workflow could be repurposed for politics.

Anthropic assessed the network as plausibly aligned with supporters of the ruling coalition because of its tone and hashtags. But it did not identify the real-world sponsor and found no evidence of government involvement.

Just as important, Anthropic says the campaign failed to break out into authentic communities. It did not prove that AI propaganda can already swing Kenyan elections.

It demonstrated how cheaply the machinery for simulated public opinion can now be assembled.

The scarce resource is no longer content. It is authenticity.

Political propaganda is not new to Kenya. Paid bloggers, coordinated hashtags, influencers, partisan pages and organised digital networks have been part of election politics for years.

Generative AI changes the cost structure.

A political operator no longer needs 50 different people to produce 50 slightly different reactions. One person can generate 50 apparent voices, then another 50, then another — each varied enough to look less obviously coordinated.

Anthropic’s investigation is especially revealing because the human operator retained control over the politics. The actor chose the ideological message, themes, talking points and hashtags. Claude supplied volume, stylistic variation and what Anthropic described as the appearance of authenticity.

This distinction matters. The immediate democratic threat is not an autonomous machine deciding whom Kenyans should support. It is human political persuasion becoming cheaper to disguise as independent human speech.

That changes the economics of influence. Content becomes abundant. Authenticity becomes scarce.

AI does not need to change your mind first

Political behaviour is shaped not only by what people believe, but by what they believe other people believe.

A citizen may reconsider a candidate if a position appears overwhelmingly popular. A journalist may treat a hashtag as evidence of public mood. A television producer may invite discussion because a narrative appears to be trending. A politician may respond to what looks like spontaneous anger. Donors and activists may decide that momentum is moving.

That is why artificial political participation can matter even before it persuades anyone of the underlying argument.

The first target can be the social perception surrounding the argument: everybody is saying this; people are turning against them; the public has already decided.

This is the core risk ahead of 2027. The challenge will not only be verifying whether a claim is true. It will be verifying whether the apparent crowd repeating it exists.

The macro numbers and the household story are diverging

Kenya’s macroeconomic picture contains real signs of stabilisation. The latest working-day exchange rate from the Central Bank of Kenya (CBK) was KSh 150.22 per euro on 11 September. The Central Bank Rate was 8.75%, while August inflation stood at 6.6%.

Households tell a less comfortable story.

TIFA Research interviewed 2,048 adults across all 47 counties in face-to-face household interviews. Sixty-six percent said their personal economic situation had worsened since the 2022 election. Only 12% reported improvement, while 23% said their situation was broadly unchanged.

TIFA also found that economic concerns dominate the national agenda. Unemployment and poverty were the most frequently cited concern, followed by inflation, high prices and taxation; corruption ranked behind them.

This is perception data, not an independent measurement of household income or consumption. But perceptions are politically consequential because voters experience an economy through food, rent, transport, jobs, school fees and disposable income — not through foreign-exchange stability alone.

Kenya’s macroeconomic repair is therefore confronting its harder stage: stability must become something households can feel.

Otherwise technically improving indicators and politically deteriorating economic confidence can coexist for a surprisingly long time.

The campaign is starting before the institutions are ready for it

Kenya’s 2027 election is already visibly under way.

On Saturday, rival groups clashed as United Opposition leaders entered Garissa. Citizen Digital reported that youths threw stones at opposition vehicles and supporters, damaging at least one vehicle before the confrontation was brought under control.

Opposition leaders, including Kalonzo Musyoka, accused Health Cabinet Secretary Aden Duale of responsibility for the disruption. The reporting cited by AFRICA3000 presents no evidence establishing that allegation, and we do not treat it as fact.

The incident is small compared with the severe electoral violence Kenya has experienced historically. It should not be inflated into evidence of a national pattern.

But it arrives at an important moment. Parties are already negotiating alliances, leaders are holding regional rallies and the physical contest for political space has begun.

Kenya therefore faces two different integrity problems at once: protecting authentic political speech online and protecting legitimate political assembly offline.

Kenya is discovering a new form of civic pollution

Public debate depends on citizens being able to distinguish advocacy from manipulation.

That boundary was already blurred by paid influencers. Generative artificial intelligence makes the scale problem much larger.

Anthropic found that the Kenyan operator used essentially the same content-production workflow for political messaging and ordinary commercial marketing. The technology did not invent the political positions; people supplied those. AI supplied variation, speed and volume.

For journalists, that should change newsroom habits. Trending hashtags, large volumes of similar comments and apparently spontaneous online reactions increasingly cannot be treated automatically as evidence of genuine public mood.

A useful distinction for 2027 will be between content verification and participation verification. The first asks: is this statement, image or video authentic? The second asks: are these really many independent people speaking?

The second question is becoming almost as important as the first.

The employment argument keeps returning to the wrong denominator

President William Ruto used Saturday to clarify his position on foreign investors and workers. He said Kenya remained open to both, provided they comply with Kenyan employment, immigration and investment rules.

He also reaffirmed Kenya’s commitment to the East African Community (EAC) and other regional trade frameworks.

The clarification matters because EAC rules go much further than visa-free travel. The Common Market provides for free movement of persons, labour, services and capital, together with rights of residence and establishment.

These rights operate through national law and procedures, but the framework also rests on non-discrimination between citizens of partner states.

For young Kenyans, however, the central employment arithmetic remains unchanged.

Restricting competition for existing informal businesses may redistribute some opportunities. It does not increase the total number of productive jobs.

TIFA’s finding that unemployment and poverty are the leading public concern makes that distinction especially important. Kenya’s employment challenge is ultimately about expanding productive economic opportunity, not merely policing who competes for what already exists.

Today’s stories reveal two different struggles over perception.

The first is economic. Government can point legitimately to a stable currency, a lower policy rate and an economy no longer experiencing the same foreign-exchange stress seen earlier in the administration. Yet two-thirds of respondents in the TIFA poll say they personally feel worse off than before the 2022 election.

The second is political. Artificial intelligence now allows operators to manufacture the appearance that dozens or hundreds of citizens independently share the same enthusiasm or hostility.

Both matter because democratic politics runs through perception.

But governments cannot repair economic confidence through messaging if households continue to feel squeezed. And citizens cannot judge political popularity reliably if coordinated messaging successfully disguises itself as public sentiment.

The Kenyan AI network documented by Anthropic did not succeed in reaching real audiences. That should reduce alarm about this specific operation, not complacency about the method. It is easier to improve a failed workflow than to reinvent one from scratch.

Kenya’s 2027 election may therefore depend unusually heavily on an old democratic resource that technology cannot manufacture: credibility.

Who has it, who loses it — and whom citizens still believe when everybody claims to speak for “the people”.

Kenya offers an early case study of a problem every European democracy is also confronting.

Generative artificial intelligence makes political influence operations cheaper, faster and easier to personalise. A highly networked political culture such as Kenya’s makes the country particularly useful for understanding how these techniques evolve before they become more effective.

The economic story matters too. European institutions often evaluate partner economies through inflation, currencies, debt and growth. Kenya shows why political economy requires another layer: whether citizens experience those improvements in everyday life.

And the foreign-trader debate touches regional integration in ways familiar to Europe. Free movement remains politically durable only when citizens believe economic openness expands opportunity rather than merely redistributes scarcity.

Kenya is currently testing all three propositions at once: whether democratic speech remains recognisably authentic, whether macroeconomic repair reaches households, and whether regional openness survives domestic economic anxiety.

Those are Kenyan questions. They are also very European ones.

AI election rules: Whether the Independent Electoral and Boundaries Commission, media organisations and technology companies develop clearer standards for AI-assisted political content and coordinated inauthentic behaviour before campaigning peaks.

Political violence: Whether Garissa proves isolated or whether organised disruption begins recurring at government and opposition events.

Economic sentiment: Whether the gap between stable macro indicators and TIFA’s household perceptions narrows as 2027 approaches.

Newsroom verification: Whether Kenyan media begin distinguishing more systematically between what is trending and whether a trend is authentically human.

Foreign-trader framework: Whether legislation and enforcement reconcile domestic employment politics with Kenya’s obligations under the East African Community Common Market.