AI Tools in Marketing: Where They Help, Where Quality and Trust Suffer

Why has AI in marketing become a question of quality?

Only a few years ago, writing was a bottleneck in small-business marketing. An expert article, a newsletter, the copy for a landing page cost days, and it often remained a good intention. Today a language model delivers a draft in seconds. According to the German digital association Bitkom, a majority of companies in Germany use AI for the first time in 2026, as the article on the EU AI Act describes. The bottleneck has shifted: from quantity to quality and from writing to reviewing.

This assessment answers the question: For which marketing tasks do AI tools help a small business, and where do quality and trust suffer? We first clarify the terms and present the model of the jagged frontier. Then we look at what experiments show about speed, quality and uniformity and how customers and search engines react. We weigh two views and apply the findings to typical marketing tasks.

AI makes writing cheap. That makes what cannot be written more valuable: experience, judgment and accountability.

What can AI tools do, and what is the jagged frontier?

Generative AI refers to software that creates new content: text, images, audio or video. For text, such tools work with language models that have learned from very large amounts of text which word is likely to follow another. They therefore produce fluent, plausible language. But they do not know whether a statement is true. Invented facts, sources or figures that sound convincing are called hallucinations.

How well AI actually helps a person is described by a model from Dell’Acqua and colleagues (2026), developed in one of the most thorough experiments on the subject. The authors speak of a jagged technological frontier. What they mean is that AI’s capabilities do not run along a straight line from easy to hard. Tasks that seem equally difficult to a human can lie on different sides of the frontier. Inside the frontier, AI improves the result; outside, it worsens it, often without anyone noticing.

The everyday analogy is a navigation device in an unfamiliar city. On the main roads it gets you to your destination faster and more safely than any map. At a construction site it does not know, it sends you into a dead end in the same calm voice. The device sounds equally confident in both cases. You have to recognize the difference yourself.

What do experiments show about speed and quality?

Dell’Acqua and colleagues (2026) worked with the consulting firm Boston Consulting Group. 758 consultants, about seven percent of the firm’s individual-contributor consultants, first solved tasks without AI to establish a baseline. They were then randomly assigned to three groups: no AI, GPT-4, and GPT-4 with an introduction to prompting. On 18 realistic tasks inside the frontier, from creative to analytical, consultants with AI completed on average 12.2 percent more tasks, 25.1 percent faster and at significantly higher quality.

On a task the researchers had deliberately chosen to lie outside the frontier, this reversed. It was a complex management task on which, in the researchers’ assessment, the model tended toward a plausible but wrong answer. With AI, according to the authors, they were 19 percent less likely to reach the correct solution than without. A second finding: consultants in the bottom half of the baseline test benefited most. The top half also improved, but less.

For writing tasks in the narrower sense there is a second experiment. Noy and Zhang (2023) gave 453 college-educated professionals occupation-specific writing tasks, such as press releases, short reports or emails. Half were randomly allowed to use a language model. In this group, the time needed fell by 40 percent, and the quality of the texts rose by 18 percent. Differences between weaker and stronger writers narrowed.

Both studies have limits that matter to owners. The tasks were self-contained and without real customers, and quality was judged by graders, not by the market. Precisely what counts in a business’s marketing, namely its own experience, real customer cases and detailed knowledge of its own offering, hardly featured in the tasks.

Does AI make texts more uniform?

A third finding concerns not the individual text but all texts together. In an online experiment, Doshi and Hauser (2024) had people write short stories. Some received ideas from a language model. These stories were rated as more creative, better written and more enjoyable, especially among writers who were less creative to begin with. At the same time, the AI-assisted stories were more similar to each other than the stories written without AI.

The authors describe this as a social dilemma: the individual is better off with AI, but overall less diversity emerges. For a business, this is more than an academic question. If all competitors write their websites, expert articles and ads with the same tools, they sound the same. Yet distinctiveness is at the heart of positioning and brand, as the article on branding for small businesses explains.

How do customers and search engines react?

For trust, what matters is how readers react to AI content when they know about it. The article on the EU AI Act summarizes the evidence. Its common finding: people often rate the same performance worse when they know a machine is behind it, partly because they assume less effort behind the message. The results are not uniform, and among people with AI experience the effect is sometimes weaker.

Google has clarified its position in its search guidelines. What is assessed is the quality of content, not how it was produced. Since the March 2024 update, however, a separate rule applies against mass-produced content: creating many pages primarily to manipulate rankings rather than to help users violates the spam policies, whether people or machines wrote them (Google Search Central 2024). Something similar applies to visibility in AI answers. Sources with clear information and evidence get mentioned, as the article on visibility in AI answers describes.

Two views: efficiency gain or trust risk?

The first view sees AI mainly as an efficiency gain. Small businesses never had enough time for marketing. If AI delivers drafts, the owner can finally publish regularly. The experiments by Noy and Zhang (2023) and Dell’Acqua and colleagues (2026) support this view, especially for people who do not like writing or are not good at it.

The second view sees mainly a trust risk. Texts become uniform, errors creep in unchecked, and readers turn away when they recognize mass-produced material. The findings of Doshi and Hauser (2024) and the studies on disclosure support this view.

I consider both findings correct and the opposition wrongly framed. Whether AI helps or harms is decided not by the tool but by the task and by accountability. Inside the frontier, for outlines, rewording, summaries and variants, it saves time without loss. Outside the frontier, for facts, figures, customer cases and judgments, it replaces nothing and creates risks. The business’s own experience is exactly what no AI can provide. It is therefore the most important ingredient of every text.

If you let AI write what only you can know, you end up publishing what everyone knows.

Which marketing tasks lie on which side of the frontier?

For sales, the article on AI in sales introduced the AI traffic light: green for drafts with full review, amber for analyses with spot checks, red for tasks that stay with people. The same classification can be applied to marketing. The jagged frontier explains why it is needed: you cannot tell by looking at a task which side it lies on, so a fixed rule is required.

AI is particularly well suited to reusing existing content. Once the core of an expert article is fixed, it can produce drafts for newsletters, network posts and checklists from it, as described in the article on repurposing content. The thinking has already been done; AI takes care of the form.

What you can put into practice right away

• List every point in marketing where AI tools are already used, including by service providers.
• Assign each task a color of the AI traffic light and write down the rule for it.
• Decide who reads and approves every text in full before publication.
• For the next expert article, first write down the core and two of your own examples, and only then ask the AI for a draft.
• Check every figure and source in AI drafts individually against the original source.
• Compare three sentences from your own website with three from a competitor’s: would it be clear who wrote what?

Conclusion: where do AI tools help in marketing, and where do quality and trust suffer?

The guiding question was for which marketing tasks AI tools help a small business and where quality and trust suffer. The experiments show clear gains in speed and quality on tasks inside AI’s capabilities (Noy and Zhang 2023; Dell’Acqua and colleagues 2026) and noticeable losses on tasks outside them. Doshi and Hauser (2024) show that AI-assisted texts become more similar to each other, and research on disclosure shows that readers often judge AI content more critically.

This results in a clear division of labor: AI for form, variants and drafts, people for core, facts, experience and accountability. The jagged frontier requires fixed rules, because you cannot tell by looking at a task which side it lies on. If you use AI this way, you gain time for what distinguishes you from others. If you let it write what distinguishes you, you save time and lose exactly what customers choose you for.