A pervasive surge in generative artificial intelligence content is creating widespread deception and eroding public trust across digital platforms.
A landmark study published by digital rights organisation Paradigm Initiative, titled ‘Trusting AI: Unpacking the deception on social media platforms’ and authored by research lead Sani Suleiman, has revealed that synthetic content is no longer an emerging novelty — it has become a daily reality.
“Generative artificial intelligence has introduced a new kind of pollution into the information environment. Like environmental pollution, it does not announce itself. It seeps in quietly, invisibly, mixing with legitimate content until the entire information ecosystem becomes difficult to breathe,” it states.
The report documents an information ecosystem saturated with synthetic media, where 83 per cent of surveyed internet users encounter AI-generated material frequently or very frequently.
Social media serves as the primary gateway for 67 per cent of users, and the deluge is predominantly visual: synthetic images at 84 per cent, videos at 67 per cent, text articles at 55 per cent, and audio or voice recordings at 34 per cent.
But only 16 per cent believe social media platforms are doing enough to address the problem, while 60 per cent say they are not.
Trust in technology companies’ transparency is the lowest-scoring area in the survey.
“The government is not doing what is right in protecting people’s data, and the service provider is using our own data that we entrusted [to] them,” one respondent said.
The report warns that any governance framework focused only on text-based AI generation “will miss most of what people are actually experiencing”.
Yet the most striking revelation — one that upends conventional digital policy — is the breakdown of personal vigilance as a protective mechanism.
In policy circles, public education, digital literacy and self-assurance are routinely promoted as the primary line of defence against online disinformation.
The study’s data exposes this assumption as a dangerous myth: confidence is simply no shield.
When cross-referencing self-reported confidence against actual deception rates, the research uncovered a startling paradox.
Respondents who rated themselves highly confident in identifying AI-generated content reported being fooled at almost the exact same rate as those expressing low confidence — 62 per cent compared with 59 per cent.
Active verification provided almost no additional protection: among participants who routinely attempt to authenticate content, 58 per cent still admitted believing AI-generated media was real before discovering it was fake.
Even digital natives proved deeply vulnerable.
Among 25 to 35-year-olds, the largest and most digitally fluent group, 63 per cent report having been deceived. Among those above 55, the figure rises to 67 per cent. Even 18 to 24-year-olds, who grew up with the internet, report a 57 per cent deception rate. Women report slightly higher deception rates at 64 per cent than men at 58 per cent.
Overall, 60 per cent of all respondents — a clear majority of highly educated, frequent internet users across 20 African nations — have been directly deceived by synthetic content.
This finding demonstrates that generative algorithms have crossed a critical threshold of realism, rendering individual human judgement insufficient.
The report observes that asking users to protect themselves through vigilance alone is equivalent to asking ordinary citizens to “fact-check a flood with a teaspoon”.
“Synthetic content is designed to be indistinguishable from authentic content. The most sophisticated tools leave no obvious fingerprints.
And even when they do, the sheer volume of AI-generated material circulating online overwhelms any realistic capacity for individual scrutiny.”
Among people who actively try to verify content, the deception rate is 58 per cent, barely lower than among those who never check.
Ninety-one per cent actively or sometimes attempt to determine whether content is AI-generated.
The most common methods are checking the source or author at 64 per cent, looking for unnatural wording at 53 per cent, using fact-checking websites at 41 per cent, relying on intuition at 39 per cent, and comparing with trusted news sources at 38 per cent. Only 31 per cent use reverse image search or dedicated AI detection tools.
Deprived of reliable platform-level signals, nearly four in ten active verifiers, 39 per cent, admit relying on intuition or gut feelings to spot synthetic media — outmatched by modern generative tools.
“AI-generated content [is being] weaponised to manufacture legitimacy for authoritarian narratives, at a scale and speed that outpaces the verification capacity of newsrooms like the one I work for,” one media professional said.
The systemic failure of individual vigilance has created a widespread crisis of epistemic confusion.
Seventy-five per cent of respondents report that AI content frequently or very frequently creates uncertainty about what to trust online, while 81 per cent regularly experience doubt regarding content authenticity.
This environment fosters a harmful “liar’s dividend”, where public scepticism spreads so widely that authentic journalism, public health announcements and official communications are routinely dismissed as fabrications.
As one participant noted, “The biggest risk? We will no longer know who to trust.”
Beyond the erosion of trust, the document highlights immediate human rights harms. Synthetic media threatens democratic integrity during critical election cycles, fuels targeted fraud and disproportionately targets women and children through non-consensual deepfake imagery.
In public health, AI-generated false medical advice risks preventable illness and loss of life. “False health information can cause people to avoid medication, take unsafe remedies, or reject proven health advice,” one respondent warned.
Moreover, an equity gap leaves rural populations, older users and those with lower digital literacy exceptionally vulnerable — a concern respondents raised themselves.
“Increasing the gap between the educated and non-educated,” one noted, because “the uneducated are at the disadvantage of accepting just anything they receive as true or real”.
While gains have been made, public awareness and demand for regulation are high. But much remains to be achieved.
Regulatory frameworks across much of Africa have yet to keep pace with the scale of the problem. Voluntary commitments have gone unenforced. Labelling frameworks have remained optional.
In response, public trust in technology companies has plummeted. Sixty per cent of respondents state that social media platforms are not doing enough, while a mere 16 per cent believe platform efforts are adequate.
Tech company transparency received the lowest trust rating in the entire survey.
Rather than viewing governance through a single lens, 51 per cent of respondents hold governments responsible, 51 per cent blame social media platforms and 49 per cent hold AI developers accountable, demonstrating a strong public demand for a shared accountability model across the tech ecosystem.
Independent regulators were named by the fewest, at 31 per cent — a figure reflecting scepticism about regulatory capacity rather than a rejection of regulation, since respondents repeatedly called for enforceable laws.
To address this crisis, the report urges policymakers to abandon reliance on individual user vigilance and instead enforce structural safeguards.
Key recommendations include establishing mandatory, persistent labelling and watermarking across all AI-generated imagery, video and audio; requiring tech companies to publish region-specific transparency reports; and establishing statutory regulatory frameworks to enforce full-chain accountability across AI developers and distribution platforms.
It also urges investment in local-language detection systems for African languages such as Swahili, Hausa, Amharic, Yoruba and Wolof, and the implementation of rapid-response escalation channels during elections and health emergencies.
The report also calls for treating elections and public-interest information as priority areas, targeting AI literacy at vulnerable populations and strengthening systematic monitoring of AI-generated harm across African digital spaces.
The objective, it concludes, cannot simply be to reduce false AI-generated content.
It must also be to preserve people’s ability to identify, trust and use credible information — an information-integrity challenge that no single actor can solve alone.
