Will AI misinformation and deepfakes reshape what we trust?

Short answer

Concern about AI-generated misinformation and deepfakes is real, durable, and growing — but it is reshaping everyday trust more than it is dismantling journalism overnight. Across 2024–2026, multiple survey houses converge on a 60–77% level of public worry, with one cross-country research team finding 70% of respondents globally can no longer tell whether online content is trustworthy because it might be AI-generated. The clear pattern is a steady erosion: voters less sure they can identify a forged video, workers hesitant about AI tools that get facts wrong, and Gen Z becoming more skeptical even while using AI daily.

Why people are concerned

The worry has two anchors: incident recall and information overload. Voters in the U.S. and U.K. remember concrete examples. In February 2024, an automated robocall using an audio deepfake of President Joe Biden urged Democratic primary voters in New Hampshire not to vote. Around the same period, a synthetic video of then-UK Prime Minister Rishi Sunak appeared to show him endorsing policy positions he had never taken. These cases moved election-integrity worry from a hypothetical into a recent memory for a large share of the public.

The second anchor is the daily experience of not knowing what is real. A 2025 study by KPMG with the University of Melbourne, covering roughly 48,000 people across 47 countries, found 70% of respondents globally struggle to trust online content because it may be AI-generated, and 87% want laws against AI misinformation (KPMG 2025; KPMG report PDF, Figure 28). This is not panic; it is cautious handling of a medium where forgery is cheap.

What is true

A handful of cross-survey findings are now well established.

Inaccurate information from AI is the rare concern on which the public and AI experts broadly agree. Pew Research (April 2025, n=5,410 + n=1,013) found 66% of U.S. adults and 70% of AI experts highly worried about inaccurate AI information — a 4-point gap, compared with a 50-point gap on jobs (Pew 2025-04-03).

Election-era deepfakes worry voters most, but few feel confident they could spot one. Jumio’s 2024 Online Identity Study (n=8,077, Censuswide Mar–Apr 2024) found 72% of Americans worried about AI and deepfakes influencing upcoming elections, 70% more skeptical than in the prior election, and only 37% confident they could spot a political deepfake (22% among those 55 and older) (Jumio 2024).

Regulator trust is unevenly distributed across democratic systems. In a 25-country Pew survey (October 2025), the median share who trust their national government to regulate AI was 53% in the EU, 37% in the U.S., 27% in China. Within the U.S., 44% trust their own country, with a 54%/36% Republican/Democratic split (Pew 2025-10-15). Stanford HAI’s AI Index 2026 reports the U.S. at 31% own-government trust — the lowest of any country surveyed — and that number restates the Pew data (Stanford HAI 2026).

Information quality is a top trust-shaping event of the last five years. The 2026 Edelman Trust Barometer (n≈34,000, 28 countries, released January 18, 2026) names “increasing prevalence of misinformation” as cited by 50% as one of the events most affecting trust over the previous five years, alongside growing use of generative AI platforms at 37%. Edelman reports 65% worry that foreign actors are injecting falsehoods into national media (2026 Edelman Trust Barometer).

Gen Z’s information skepticism is intensifying year over year. Gallup’s Voices of Gen Z 2026 wave (n=1,572, Feb–Mar 2026) found only 25% of Gen Z believe AI will help them think carefully about information, against 42% who believe it will hurt. Among daily AI users, enthusiasm dropped 18 points year over year (Gallup Voices of Gen Z 2026).

Workplace AI hallucinations are a documented practical risk. AP News, reporting on a Gallup Q4 2025 workforce poll (n=23,617), describes lawyers and other professionals encountering AI tools that “hallucinate” — confidently generate false legal citations or fake case law (AP News / Gallup 2026). Among workers with AI tools available who do not use them, about four in ten cite ethical opposition, data-privacy concerns, or disbelief that AI can help.

What is exaggerated, misleading, or unsupported

• “Nobody can tell what is real anymore.” Strong framing, weak evidence. Survey data measures confidence in spotting a deepfake, not accuracy. Jumio’s 37% U.S. figure is a self-rating, not an error rate. People are more cautious, not necessarily worse detectors.
• Polls show Americans have stopped trusting all news. Misleading as stated. Edelman’s 39% figure is a partisan-exposure finding, not a blanket collapse. Trust in specific institutions, not “news” as a category, is what has fallen.
• Deepfakes have already swung a major election. Unsupported at this time. Documented cases exist (the Biden robocall, the Sunak video), and worry is real, but proving a single deepfake changed a vote outcome requires multi-cycle forensic evidence the public record does not establish. Treat this as a hypothesis, not a finding.
• One survey house’s findings contradict the consensus. A January 2026 Blue Rose Research poll placed AI 29th of 39 issues, but the same firm’s trend shows AI concern rising fastest among the issues tested. A low rank and a steep climb are not contradictions — they describe an issue at the start of public salience, not the end.

What remains uncertain

Several open questions matter but cannot be answered from public-opinion data alone:

• Whether deepfake incidents causally change vote shares, turnout, or trust in specific elections (multi-cycle forensics are not yet published).
• The full per-country cut of misinformation worry from the KPMG Figure 28 series (the global aggregate is clear; individual-country variation has not been pulled in detail).
• Whether parent-reported child AI chatbot use, which Pew measured at 8% among parents of 5–12-year-olds and 15% among parents of 11–12-year-olds as of March 2026 (Pew 2026-03-12 short read), translates into measurable information-accuracy harms for children.
• The political effect of 2026 U.S. midterm-cycle deepfake-ad spending, much of which is documented in 2026 reporting but lacks a primary URL accessible to this brief.

Where we are likely headed

Editorial judgment, not research result. Looking two to five years out, the most likely trajectory is not collapse of trust but a slow re-architecture of how trust signals are produced and verified. Platform-side provenance tooling, including cryptographic content credentials and on-platform AI-generated labels, is moving from optional to expected in jurisdictions with active rulemaking, but adoption outside Western democracies will be uneven. Voters are likely to rely more on procedural cues — identity verification, watermarking, attestation chains — and less on the visual realism of any single video. Workplaces are likely to treat AI “hallucination” risk as a standard compliance item, similar to phishing or data-handling training, with named-responsibility review for AI-assisted outputs in regulated fields. Public concern, judging from Gen Z trajectories, will probably keep rising among heavy users before it plateaus, particularly if a high-profile deepfake incident in the 2026 U.S. cycle lands. None of this is a forecast a survey can prove; it is the most plausible path given how the 2024–2026 evidence is moving.

What this means for people and small businesses

For individuals, three habits are already worth forming. First, treat any viral video of a politician, executive, or public figure as unconfirmed until corroborated by two independent mainstream outlets — not because all viral video is fake, but because the cost of forgery is now low enough that one share can outrun the correction. Second, when an AI assistant produces a fact in a high-stakes context (medical, legal, financial, employment), verify it at the primary source before acting on it. Third, teach children and older relatives the single most useful heuristic: a polished video is no longer evidence of anything.

For small businesses, the practical implications are concrete. Customer-facing AI copy, images, and audio should carry provenance where possible, since platforms and regulators are beginning to require it. Hallucination risk is already a documented liability in regulated professions and is becoming one in ordinary commerce; processes that require a human reviewer to sign off on an AI-assisted output before it reaches a customer are no longer optional best practice. And in any internal communications that mention regulated topics — health, finance, employment, voting — assume the recipient will double-check, and structure the message to make that check easy.

Bottom line

Take seriously: people losing confidence in their own ability to identify a synthetic political video; the worry spanning ordinary consumers, voters, workers, and AI experts; and misinformation and generative-AI growth now being named in the same sentence as inflation and COVID-19 as five-year trust-shapers. Do not panic about: a sudden collapse of all trust, “no one can tell what is real,” or the claim that deepfakes have already swung specific elections — the evidence supports the worry, not those conclusions. Watch for: the next high-profile 2026-cycle deepfake incident; provenance and labeling rollouts from major platforms; and updates to Gallup’s Gen Z and Edelman series, the clearest longitudinal read on whether the concern is still rising or beginning to plateau.