What the Research Says About Consumer Trust in AI-Generated Content

In six experiments with 4,600 participants, people attempted to distinguish AI-generated self-presentations from human-written ones. They performed at near-chance accuracy. More specifically: the heuristics they relied on to make that judgment — associating first-person pronouns and references to family and relationships with human writing — were systematically wrong. Those features appeared in AI-generated text at least as often as in human-written text.

That finding, from Jakesch, Hancock, and Naaman (PNAS, 2023), is the baseline for understanding consumer trust in AI-generated marketing content. Three published studies together establish what this means for disclosure, brand voice, and the strategic requirements of using AI to produce content that represents your brand.

The full citations are in the AI-Generated Content and Consumer Trust research index.

Heuristic detection fails systematically

Jakesch, Hancock, and Naaman (PNAS, 2023) ran six experiments with 4,600 participants who tried to distinguish AI-generated self-presentations — LinkedIn-style professional profiles — from human-written ones. Detection accuracy was near-chance across all conditions.

The heuristics participants used: they associated first-person pronouns ("I," "my"), references to family and relationships, and colloquial language with human writing. They were wrong. These features appeared in AI-generated text at the same frequency as in human-written text — because AI-generated text is trained on human-written text and produces the same distributional patterns.

Two direct implications for brand strategy:

Transparency cannot be delegated to reader intuition. Because heuristic detection fails, explicit disclosure is the only available transparency mechanism. A brand that relies on readers to recognize AI-assisted content is relying on a mechanism the research shows does not work.

Brand distinctiveness requires active construction. Readers cannot intuitively distinguish generic AI output from branded content on surface signals alone. Without deliberate brand voice discipline — specific directives, enforced consistently — AI-generated content converges toward the generic patterns that detection fails to identify.

AI writing tools without brand constraint shift writer views

Jakesch et al. (CHI, 2023) ran an experiment with 1,506 participants who co-wrote essays on controversial topics using AI writing assistants that had embedded suggestions favoring one political side. The assistants influenced not just the written output — which was expected — but participants' self-reported opinions in post-task surveys.

Users shifted their stated views in the direction the AI was nudging. Not just their text. Their stated opinions.

This finding is the most consequential one in this literature for marketing teams. AI writing assistance without brand constraint does not produce neutral assistance. It produces output shaped by whatever patterns dominate the model's training distribution — and those patterns influence both the content produced and the thinking of the people using the tool.

For brand voice, the risk is not dramatic inconsistency in any single piece. It is gradual drift across many pieces where each reads as close enough individually but the aggregate loses the brand's distinctive position — and where the writers using AI tools are themselves being shaped by the AI's unconstrained patterns. Explicit brand context loaded into the AI session is not just a formatting preference; it is the intervention that addresses a measured, documented drift mechanism.

LLM persuasion achieves human-level or better effectiveness

Noels et al. (arXiv, 2024) surveyed the published empirical literature on LLM persuasion effectiveness — the evidence base for how persuasive AI-generated content actually is.

The finding: LLM-based persuasion systems frequently achieved human-level or even superhuman persuasiveness across the domains studied. Personalization and model scale were identified as key factors. And AI source disclosure was identified as a moderating variable: in some conditions, disclosing that content was AI-generated reduced its persuasive effect.

Two implications follow:

Disclosure is a trust calibration signal, not a formality. If disclosure moderates persuasive effect, consumers are using that information to calibrate their trust response. A disclosure that reads "some of our content is produced with AI assistance" is not just a legal or ethical statement — it is information that consumers actively use when evaluating what they read. Brands that build disclosure practices proactively are serving a genuine consumer need, not performing a compliance ritual.

AI persuasion capacity is real and growing. This makes brand voice discipline more strategically important, not less. Content that demonstrably represents a brand's authentic expertise — produced with AI assistance but governed by explicit brand context — is categorically different from AI persuasion operating without disclosed brand authorship. The former is a brand capability; the latter is a reputation risk.

The strategic position that follows

Three findings together define the strategic requirements for AI-assisted marketing content:

Consumer heuristic detection fails (Jakesch et al. PNAS 2023) → explicit disclosure is the only available transparency mechanism, and brand distinctiveness requires deliberate construction, not assumption.

Unconstrained AI writing assistance shifts both content and writer views (Jakesch et al. CHI 2023) → brand context must be actively loaded into AI sessions, not assumed or left to model defaults.

AI persuasion achieves human-level effectiveness and disclosure moderates it (Noels et al. 2024) → disclosure is strategically relevant, not just ethically appropriate, because consumers use it to calibrate trust.

Taken together, these are not concerns about AI's limitations — they are findings about what AI-assisted marketing requires. The brands that get this right are the ones that build explicit brand context systems, practice consistent disclosure, and understand that AI without constraint is not a neutral tool. It produces output shaped by patterns that benefit the model's training distribution, not any particular brand's position.

For disclosure practices, see disclosing AI marketing content. For the human oversight side, see human in the loop and reported, not generated.

Frequently Asked Questions

If consumers can't detect AI content, does disclosure matter?

Yes — and the research specifies why. Because heuristic detection fails (Jakesch et al. 2023), explicit disclosure is the only available transparency mechanism. The alternative — relying on reader intuition — relies on a mechanism the research shows does not work. The Noels et al. 2024 survey found that disclosure moderates persuasive effect, meaning consumers use it as information to calibrate their response. Brands that build disclosure into their AI content practices are responding to how the research says consumers actually process this information, not performing a formality.

How concerned should marketing teams be about brand voice drift from AI tools?

Significantly. The Jakesch et al. CHI 2023 study demonstrated that AI writing tools actively shape both content and writer views in the direction of the model's embedded patterns — not a hypothetical risk but a measured outcome across 1,506 participants. For brand voice, the risk accumulates gradually: each piece reads as close enough individually while the aggregate drifts from the brand's distinctive position. AI assistance without explicit brand context loading does not hold a neutral position. It pulls toward training distribution patterns.

What makes AI-assisted content trustworthy?

The research points toward two factors: transparency and quality signals that function as trust proxies. Because heuristic detection fails (Jakesch et al. 2023), consumers rely on brand authority, source reputation, and content quality as proxies for trustworthiness. AI-generated content that carries a known brand's distinctive voice and substantive expertise does not require consumers to perform detection; the brand's reputation is the trust vehicle. This is why brand voice consistency and content substance matter independently of the disclosure question.

Should all AI-assisted marketing content be disclosed?

The research supports disclosure as both ethically appropriate and strategically sound. Ethically: because consumers cannot detect AI-generated content reliably (Jakesch et al. 2023), disclosure is the only available transparency mechanism. Strategically: the Noels et al. 2024 survey shows disclosure is a variable that affects how consumers process AI-generated content, meaning it is information they actively use. Copper Sun's position on this is in disclosing AI marketing content.