AI-Generated Marketing and the Trust Gap: What the Data Says

The first problem with AI-generated marketing is not whether the model can write a cleaner subject line, resize a visual, or produce ten campaign variants before lunch. The harder problem is that marketers often evaluate the work from inside the production system, while customers evaluate it as another request for attention and trust.

That gap is now measurable. In a Klaviyo/Datalily consumer survey of 8,000 people fielded in December 2025, only 7% of consumers said they trust brands more when they see AI-generated marketing, while 31% said they trust those brands less.[1] In an IAB/Sonata Insights study conducted from October 2025 to January 2026, 82% of ad executives said they believed Gen Z and Millennial consumers felt positive about AI ads. Only 45% of those consumers actually did.[2]

That is not a small messaging problem. It is the kind of asymmetry that shows up later as weaker comments, lower confidence in personalization, more review cycles, and awkward budget conversations. A marketing team may be optimizing for speed and asset volume. The audience may be reading the same assets through a different filter: authenticity, disclosure, emotional fit, and whether the brand is trying to pass off synthetic work as human judgment.

The Confidence Gap Is Wider Than the Adoption Story

AI adoption data can make the marketing case feel settled too early. If teams are using AI, if executives expect efficiency, and if tools are improving, it is tempting to treat consumer acceptance as a lagging indicator that will catch up. The newer trust data does not support that shortcut.

The IAB/Sonata finding is especially useful because it does not simply ask whether consumers like or dislike AI advertising. It compares what ad executives believe younger consumers feel with what those consumers report. The study found a 37-point gap in 2026 between executive perception and Gen Z/Millennial sentiment, up from 32 points in 2024.[2]

There is an important boundary here: the consumer sample covers ages 16 to 43. It does not include older Gen X or Boomer consumers, who may not respond the same way. So the study should not be treated as a full-market estimate. If anything, it warns against one of the more comfortable assumptions in budget meetings: that younger audiences are automatically relaxed about AI-generated ads because they are more digitally native.

The Klaviyo/Datalily numbers widen the lens beyond ad-industry expectations. A 7% trust lift versus a 31% trust decline means the downside is not evenly balanced by upside enthusiasm.[1] For a brand manager, that ratio matters more than a generic claim that consumers are “getting used to AI.” If a visible AI label or obviously synthetic experience makes nearly a third of consumers less trusting, the team needs a plan for the trust cost, not only a forecast for production savings.

That does not mean every AI-assisted campaign is doomed. It means the adoption argument has to include exposure: where AI appears, how obvious it is, whether the use case is customer-facing, and who signs off before the work leaves the organization.

Consumers Are Not One Skeptical Block

The Klaviyo/Datalily survey is also useful because it separates consumers into four AI attitude personas: AI-Averse, Cautiously Curious, Pragmatic, and All-In.[1] That segmentation should keep teams from flattening the market into a simple “people hate AI” story.

The AI-Averse customer is not waiting for better prompting. This person is more likely to see AI-generated marketing as a reason to distrust the brand. A Cautiously Curious customer may accept AI when the benefit is clear and the execution does not feel deceptive. A Pragmatic customer is less interested in the production method than in relevance, usefulness, and whether the experience saves effort. The All-In customer is the easiest to overgeneralize from because they may enjoy novelty that the rest of the audience reads as a warning sign.

This is where segmentation becomes operational. A campaign that is acceptable in a product-led, tech-forward audience may not travel safely into a healthcare, financial, education, or family-oriented context. A lifecycle message using AI to improve timing may create less resistance than a synthetic founder video. A product recommendation that feels useful may be tolerated; a generated apology after a service failure may not.

Some marketers use an informal “30% rule” to keep AI contribution below a perceived threshold of obviousness. That may be a practical internal shorthand, but it should not be treated as research-backed. The better question is not what percentage of the asset came from AI. It is whether the customer-facing experience crosses a trust boundary for the audience and channel involved.

Better Output Does Not Remove the Label Effect

A common response to AI skepticism is to raise the quality bar: better inputs, better models, better editing, better brand voice training. Those improvements matter. They just do not fully solve the trust problem.

Controlled experiments from NIM tested consumer reactions to identical ads in the US, UK, and Germany, with samples of 1,000 people per market. When ads were labeled “AI-generated,” consumers rated the same creative as less natural, less useful, and less emotionally resonant. They also reported lower likelihood to click or purchase.[3]

That finding should be handled carefully. Controlled online experiments are not the same as real-world scrolling, buying, or unsubscribing behavior. People may say they are less likely to click and still behave differently in a live feed. But the mechanism is important: the label itself can change interpretation. The customer is not only judging the asset. They are judging what the asset implies about the brand’s effort, sincerity, and willingness to disclose how the work was made.

That creates a difficult but manageable tension. Disclosure can trigger skepticism, yet nondisclosure can create a larger breach if consumers feel misled. Emplifi found that 91% of consumers expect brands to disclose AI use in marketing, and 52% would stop buying after an inauthentic experience.[4] The practical answer is not to hide AI use by default. It is to decide where disclosure is required, what form it should take, and which uses should not ship even if they can be generated.

Where AI-Generated Marketing Becomes Reputationally Fragile

The risk is not evenly distributed across every marketing use case. AI-assisted resizing, variant generation, research synthesis, tagging, and internal drafts usually carry a different customer trust profile than synthetic testimonials, generated spokespeople, AI-written crisis communications, or highly personal messages that pretend to know the customer intimately.

A Gartner 2026 survey found that 50% of US consumers would prefer to give business to brands that avoid generative AI in customer-facing content.[5] Because some commissioned-study methodologies are not always fully visible publicly, that number should not be used as a universal boycott forecast. It is still a useful warning for teams that treat customer-facing generation as a low-friction efficiency play.

The more an asset asks for belief, the higher the review standard should be. A generated product background in a display ad is one thing. A synthetic customer story is another. A personalized email that recommends products based on behavior may be welcome when it is accurate and useful. The same email can feel invasive or lazy if it carries obvious generated phrasing, incorrect assumptions, or emotional language the brand has not earned.

This is why the ROI conversation cannot stop at production hours saved. If a team wants leadership to fund AI-generated marketing, it needs to show how the program protects revenue quality, not only how it increases output. The same logic applies to measurement: teams that cannot connect AI work to commercial outcomes will struggle to defend the program when customer feedback turns negative. Signal & Convert’s Jasper AI ROI measurement framework is useful here because it frames AI investment as a proof problem, not just a tooling decision.

The Budget Line Most Teams Underfund

The governance gap is visible in budget allocation. Improvado reported that teams allocate 22% of AI marketing budget to generation and 3% to governance.[6] That split explains a lot of the operational anxiety around AI. The money goes toward making more assets. Much less goes toward deciding which assets are safe, who reviews them, how disclosure works, and what happens when performance data and customer sentiment disagree.

Governance does not need to mean a slow committee for every social post. It does need to mean that riskier uses have a path before the campaign is live. A practical operating model usually separates low-risk production assistance from customer-facing generation that can affect trust.

The point of a table like this is not to create bureaucracy for its own sake. It gives the person accountable for the campaign a defensible answer when someone asks, “Who approved this?” It also prevents the common pattern where AI use expands quietly until a public-facing asset creates a trust problem no one budgeted for.

Teams looking for the operating model, not just the policy language, can start with documented AI content marketing workflow patterns. The useful workflows are the ones that make review visible: intake, generation, editing, claim checking, brand approval, performance monitoring, and a clear escalation path.

What Disclosure Has to Do in Practice

Disclosure is often treated as a legal or ethical checkbox. In AI-generated marketing, it also functions as expectation management. A vague footnote that says “AI may have been used” does not do the same job as a clear label near an AI avatar, generated image, or synthetic voice. The disclosure should be close enough to the experience that the customer understands what they are seeing before they rely on it.

There is no single disclosure format that fits every channel. A paid social ad using a synthetic avatar may need an on-screen label. A landing page using AI-assisted product copy may need stronger internal substantiation rather than a prominent consumer label, depending on how materially AI shaped the claim. A chatbot, recommendation module, or AI-personalized experience may need both upfront disclosure and clear handoff to a human support path.

The safest disclosure decisions start with the customer’s likely misunderstanding. If a viewer might think a synthetic person is a real employee, creator, or customer, the disclosure needs to remove that ambiguity. If a message sounds personally written by a human but was assembled at scale, the brand should be careful about emotional language it cannot stand behind.

Concrete examples help because disclosure gets abstract quickly. AI-labeled avatar ads, such as those built through TikTok Symphony, show how the label becomes part of the creative environment rather than a separate policy document. Signal & Convert’s guide to TikTok Symphony AI avatar ads is a useful reference point for teams deciding how visible AI involvement should be in the asset itself.

Quality Gates Should Measure More Than Fluency

Generated marketing often sounds acceptable before it is strategically safe. Fluency is a weak quality gate because a polished sentence can still exaggerate a claim, flatten brand voice, misuse customer data, or adopt a tone that feels emotionally false.

A stronger review asks four questions before customer-facing AI work ships:

• Is every factual claim traceable to an approved source?
• Would the message still feel appropriate if the customer knew how it was produced?
• Does the asset imply human experience, customer testimony, or personal attention that is not actually present?
• Is the expected efficiency gain large enough to justify the added review and monitoring burden?

That last question is where many AI programs get uncomfortable. A generated campaign can be faster and still be a poor investment if it requires cleanup, damages trust, or produces assets the team cannot confidently attribute to revenue. For context on where AI marketing ROI is real rather than assumed, the 2026 AI marketing ROI evidence is a better companion to a budget case than a generic productivity claim.

The same discipline applies when AI is working well. Personalization, for example, can be commercially useful when it reduces friction and improves relevance. But measured personalization wins do not automatically validate every generated message. The useful distinction is between AI that improves the customer’s task and AI that mainly lowers the brand’s production cost. The former has a clearer trust argument. The latter needs stricter review.

Teams comparing successful use cases can look at measured AI landing page personalization examples, but the lesson should be narrow: performance lift is strongest when the AI use makes the experience more relevant, not merely more automated.

Human Oversight Needs a Real Job Description

“Human in the loop” is too vague to defend a campaign after launch. The useful version names the person, the decision, and the rejection criteria. A content lead may own voice and substantiation. A brand manager may own authenticity and fit. Legal may review regulated claims. A channel owner may decide whether the disclosure format works in the placement. Someone has to be able to stop the asset.

That role becomes more important when leadership expects AI to reduce cost. If the governance budget is missing, the review burden moves into unpaid time: editors doing extra fact checks, lifecycle managers rewriting generated copy, social teams absorbing comment risk, and brand owners explaining why a “quick AI test” created a bigger reputation issue than the saved hours were worth.

HubSpot/McKinsey adoption-tier research is useful here as supporting context: teams are not all at the same maturity level, and higher adoption does not automatically mean stronger controls.[7] The more AI moves from experimentation into repeatable marketing operations, the more the operating model matters. Early experiments can survive on individual judgment. Scaled programs cannot.

Failure-pattern analysis from Pragmatic Digital across seven brands reinforces the same point: weak governance tends to show up as recurring patterns, not isolated bad luck.[8] The specific lessons should not be overextended beyond the cases studied, but the pattern is familiar enough to take seriously. AI marketing fails less because no one knew the tool could hallucinate and more because no one had authority, time, or criteria to catch the problem before launch.

This is also why AI ROI failures are rarely just analytics failures. When governance is thin, bad assets still consume media budget, customer attention, support time, and executive confidence. Signal & Convert’s AI marketing ROI failure patterns connect that operational weakness to the broader problem of proving value after the pilot phase.

What to Take Into the Budget Meeting

The defensible case for AI-generated marketing is not “we can make more content faster.” That may be true, but it is incomplete. A stronger case separates production value from trust exposure and funds both.

The most useful numbers in the current research are not the ones that make AI look inevitable. They are the ones that show where assumptions break: 82% of ad executives expecting positivity from younger consumers while only 45% report it; 7% of consumers trusting brands more when they see AI-generated marketing while 31% trust them less; 91% expecting disclosure while many teams still treat governance as a small afterthought.[1][2][4]

Those numbers do not argue for avoiding AI. They argue for treating customer trust as a budgeted operating requirement. Better models may reduce awkward phrasing and obvious synthetic artifacts. They will not, by themselves, decide when disclosure is necessary, which claims need substantiation, which use cases are too sensitive, or who is accountable when an asset goes live.

Teams that invest heavily in generation while underfunding disclosure standards, human oversight, quality gates, and governance are not closing the trust gap. They are making it easier to scale directly into it.

References

1. Klaviyo/Datalily consumer AI marketing survey, Klaviyo/Datalily, December 2025
2. IAB/Sonata Insights AI ads perception study, IAB/Sonata Insights, October 2025-January 2026
3. NIM controlled experiments on AI-generated ad labels, NIM
4. Emplifi consumer authenticity and AI disclosure data, Emplifi
5. Gartner 2026 survey on generative AI in customer-facing content, Gartner, 2026
6. Improvado AI marketing budget allocation data, Improvado
7. HubSpot/McKinsey AI adoption tiers, HubSpot/McKinsey
8. Pragmatic Digital 7-brand AI marketing case study pattern analysis, Pragmatic Digital