To disclose or not disclose, is no longer the question – effect of AI-disclosed brand voice on brand authenticity and attitude

Alexandra Kirkby, Carsten Baumgarth, Jörg Henseler•View original

HighlightsBalancedharry & harper hosts

Harry: Every brand manager has been quietly avoiding a question. If artificial intelligence wrote your marketing copy, do you tell people? The working assumption, which is baked into industry practice and a lot of academic literature, is that you don't. Because the moment consumers know a machine generated those words, the thinking goes, the brand feels less real. You've broken the spell. Harper: And that assumption is now directly testable. Alexandra Kirkby, Carsten Baumgarth, and Jörg Henseler conducted a controlled experiment to determine whether it's actually true — and it is not. Not in the way the field expected. Harry: I want to hold onto that assumption for a second because it's not irrational. Brand voice and brand authenticity are two of the most important concepts in contemporary marketing. Brand voice is what consumers perceive from the text itself — the attitude, the tone, and the choice of language. Brand authenticity is whether the brand comes across as real, continuous, and honest. The research that Kirkby and colleagues cite shows that authenticity has, in some markets, surpassed quality as a purchase driver. So if AI disclosure damages either of those, it's a real problem. Harper: That's the framework — three constructs in a chain. Brand voice authenticity feeds into brand authenticity, which drives brand attitude, meaning the consumer's overall evaluation of the brand. Structural equation modeling, which tests whether relationships between variables hold after accounting for measurement error, confirmed that both links are strong. The path coefficient from voice authenticity to brand authenticity is 0.52. The path coefficient from brand authenticity to brand attitude is 0.84. The chain is solid. The question is, what disrupts it? Harry: And the hypothesis was that disclosure would lower the perceived authenticity of the voice, which then drags down the rest of the chain. Harper: So they tested it. They conducted a three by three factorial experiment, with three disclosure conditions crossed with three levels of emotionality. Participants saw one of three real Adidas texts from Adidas dot co dot UK — a product specification, a product description, or a chatbot conversation — labelled either "generated by artificial intelligence," "written by a human," or not labelled at all. Six hundred and twenty-four English-speaking participants, with a mean age of twenty-seven, were randomly assigned to one condition. Harry: The disclosure effect on brand voice authenticity was not significant, with a path coefficient of 0.07 and a p-value of 0.16. The impact on brand attitude was also not significant. The mean scores were nearly identical: brand voice authenticity was 3.15 for human-labelled text, 3.22 for AI-labelled text, and 3.24 for unlabelled text. On a five-point scale, that's noise. Harper: Disclosure does not hurt. Transparency is safe — at least in this context. Harry: Here's where I push back. The brand is Adidas, one of the most recognized sportswear brands on the planet. The sample consists of students who are comfortable with AI-generated content. The texts include product specifications and a chatbot exchange, which are low-stakes and low-involvement. Of course, disclosure doesn't damage authenticity in that context. The real test is whether this holds for a brand that consumers don't already love, or for higher-involvement decisions. Harper: The paper states exactly that. Kirkby and colleagues highlight the student sample and single-brand design as explicit limitations. They are not overclaiming. What the study establishes is a proof of concept: in a well-controlled experiment, the feared negative effect of AI disclosure simply did not materialize. Additionally, regulatory pressure — including the European Commission's AI Act — is pushing toward mandatory disclosure. Brands that have been staying quiet because they assumed transparency would cost them something now have evidence that it may not. Harry: That's fair. The finding shifts the default. Before this, silence felt like the safe option. Now silence is a choice you are making against evidence, not in the absence of it. Whether that impacts your brand, your consumers, or your product category — that is still open. Harper: And that's the work this paper opens up, rather than closes down. Harry: This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

Every brand manager has been quietly avoiding a question. If artificial intelligence wrote your marketing copy, do you tell people? The working assumption, which is baked into industry practice and a lot of academic literature, is that you don't. Because the moment consumers know a machine generated those words, the thinking goes, the brand feels less real. You've broken the spell.

And that assumption is now directly testable. Alexandra Kirkby, Carsten Baumgarth, and Jörg Henseler conducted a controlled experiment to determine whether it's actually true — and it is not. Not in the way the field expected.

I want to hold onto that assumption for a second because it's not irrational. Brand voice and brand authenticity are two of the most important concepts in contemporary marketing. Brand voice is what consumers perceive from the text itself — the attitude, the tone, and the choice of language. Brand authenticity is whether the brand comes across as real, continuous, and honest. The research that Kirkby and colleagues cite shows that authenticity has, in some markets, surpassed quality as a purchase driver. So if AI disclosure damages either of those, it's a real problem.

That's the framework — three constructs in a chain. Brand voice authenticity feeds into brand authenticity, which drives brand attitude, meaning the consumer's overall evaluation of the brand. Structural equation modeling, which tests whether relationships between variables hold after accounting for measurement error, confirmed that both links are strong. The path coefficient from voice authenticity to brand authenticity is 0.52. The path coefficient from brand authenticity to brand attitude is 0.84. The chain is solid. The question is, what disrupts it?

And the hypothesis was that disclosure would lower the perceived authenticity of the voice, which then drags down the rest of the chain.

So they tested it. They conducted a three by three factorial experiment, with three disclosure conditions crossed with three levels of emotionality. Participants saw one of three real Adidas texts from Adidas dot co dot UK — a product specification, a product description, or a chatbot conversation — labelled either "generated by artificial intelligence," "written by a human," or not labelled at all. Six hundred and twenty-four English-speaking participants, with a mean age of twenty-seven, were randomly assigned to one condition.

The disclosure effect on brand voice authenticity was not significant, with a path coefficient of 0.07 and a p-value of 0.16. The impact on brand attitude was also not significant. The mean scores were nearly identical: brand voice authenticity was 3.15 for human-labelled text, 3.22 for AI-labelled text, and 3.24 for unlabelled text. On a five-point scale, that's noise.

Disclosure does not hurt. Transparency is safe — at least in this context.

Here's where I push back. The brand is Adidas, one of the most recognized sportswear brands on the planet. The sample consists of students who are comfortable with AI-generated content. The texts include product specifications and a chatbot exchange, which are low-stakes and low-involvement. Of course, disclosure doesn't damage authenticity in that context. The real test is whether this holds for a brand that consumers don't already love, or for higher-involvement decisions.

The paper states exactly that. Kirkby and colleagues highlight the student sample and single-brand design as explicit limitations. They are not overclaiming. What the study establishes is a proof of concept: in a well-controlled experiment, the feared negative effect of AI disclosure simply did not materialize. Additionally, regulatory pressure — including the European Commission's AI Act — is pushing toward mandatory disclosure. Brands that have been staying quiet because they assumed transparency would cost them something now have evidence that it may not.

That's fair. The finding shifts the default. Before this, silence felt like the safe option. Now silence is a choice you are making against evidence, not in the absence of it. Whether that impacts your brand, your consumers, or your product category — that is still open.

And that's the work this paper opens up, rather than closes down.

This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

Related lectures