Limitations and side effects: What is the "dark side" of AI marketing?

From generating false information and triggering a brand trust crisis, to content homogeneity and disrupting SEO order, to the black box nature of smart bidding, traffic hijacking of low-quality content, and legal risks of intellectual property and data privacy, the side effects of AI marketing are being seriously underestimated. This article will deeply analyze the limitations and risks behind AI marketing, explore how to avoid falling into new difficulties while embracing technology, and provide thoughts and warnings for the healthy development of the marketing industry.

Google AI has been in a very embarrassing situation recently.

During the Super Bowl, Google wanted to use 50 small and micro business cases in 50 states to show how GeminiAI can help businesses grow their operations and win more potential customers' interest and attempts in AI marketing. However, it never occurred to Google that one of the cheese merchants used AI-generated product descriptions, "Gouda accounts for 50 to 60 percent of the world's consumption," which sparked widespread doubts. Sharp-eyed cheese lovers pointed out that this data was clearly inconsistent with the facts.

Advertisers are therefore worried that using AI for marketing may not be as efficient and labor-saving as they imagined.

Taken from a video ad

This is just one of the problems of AI marketing that is easily seen by the public. Under the surface, the dark side of AI marketing is growing: the proliferation of homogeneous AIGC content; invalid content disrupting SEO order, which in turn drags down search marketing; and even more and more AI-generated content "plagiarized" information, which not only pollutes the credibility of Internet information, but also connects to alliances to obtain a large amount of advertising.

Apparently, the side effects of AI marketing have been seriously underestimated. A marketing professor at the University of Rhode Island said that of the 290 articles published in 15 advanced marketing journals in the past 10 years, only 33 mentioned the potential "dark side" of AI marketing.

From ChatGPT to DeepSeek, AI has been advancing by leaps and bounds. According to the prospectus of Eastcom Marketing, the market size of the AI ​​marketing industry will increase from 20.9 billion yuan in 2020 to 53 billion yuan in 2024, with a compound annual growth rate of 26.2%. Perhaps now is also a good time to "prevent problems before they happen" before AI marketing is fully rolled out, and talk about the "dark side" of AI marketing.

Generates noise materials that are inconsistent with facts/values, which negatively affects brand favorability

The above example from Google reminds us of the problem of accuracy of AI content. Although AIGC has greatly reduced the cost of material production, the manual review and traceability of AIGC content should not be relaxed. After all, once there is erroneous content or a "failure" that is inconsistent with user values, the reputation and goodwill that the brand has worked so hard to accumulate will be destroyed.

Last year, Coca-Cola’s first fully AI-generated Christmas ad, “Holiday Magic Is Here,” was badly criticized on social media, with users commenting that the video “lacks vitality,” “lacks soul,” “has no real creativity at all,” and “is creepy, like a scene from a horror movie.”
Another tragic example is the launch of Microsoft's chatbot Tay on Twitter (X) in 2016. But within a few hours, Tay made various racial and sexist remarks and was immediately taken offline. Imagine that in a marketing scenario, the brand owner originally intended to use AI to save customer service costs and bring a better interactive experience, but the AI ​​ended up talking nonsense, offending customers, and causing huge losses. The reason behind this is that the controllability of AI marketing has not yet reached a level that people can be completely assured of.

Homogeneous AI content erodes brand uniqueness

Even if it is the right content, AIGC should not be the centerpiece of marketing content at this time.

“As marketers rely more and more on AI to generate content, homogeneity is starting to set in. This is the exact opposite of what brands are looking for, which is differentiation and authenticity, ” said Jen Iliff, founder of 3X Marketing.

Marketing relies on creativity, innovation, and originality. Relying too much on generative AI risks losing these elements and may lead to content that is cookie-cutter. Imagine if every campaign relied on the same process, every message was based on similar rhetoric, and there were exquisite-looking brand digital people that consumers could not tell apart. Your brand uniqueness would disappear.

When management is complacent about the company's achievements in AI content, they can easily ask front-line sales staff to "mass-send" these noisy materials, excessively disturbing users and undermining brand favorability.

Digital people of Jindian, Huaxizi, Fotile, and Meiji masks

Who do you remember?

Invalid content disrupts SEO order and causes search engine ranking demotion

Of course, if the AI ​​content is not done well, it will not only directly offend users, but will also indirectly offend the "platform" and damage brand interests.

For example, some AI-generated marketing materials lack information value, and the content does not meet the search intent of platform users. After users click in, they will quickly leave the page, resulting in a very high bounce rate, which affects SEO performance. Some brands, when creating AI content, over-stack keywords (black hat SEO) in order to improve SEO rankings , or publish a large amount of homogeneous content, which causes search engine penalties.
In short, low-quality, repetitive or pieced-together AI marketing content may be regarded as spam by the platform, resulting in a drop in rankings.

Smart bidding is a black box, and advertisers have difficulty understanding the logic

In addition to the creative content that users can see with their naked eyes on the surface, AI actually has a greater impact on the advertising and marketing industry in the delivery stage - automated advertising delivery.

We have seen that all major platforms and service providers have become "intelligent". For example, when targeting the audience, they can automatically match the target group, accurately estimate user interests, and scale the group size as needed; when bidding, they can realize self-service delivery and adjust the bidding strategy. For example, Bytedance has launched the automated advertising delivery product UBMax.

In 2023, Alimama also proposed AIGB (AI Generated Bidding), which models the automatic bidding problem as a generative sequence decision problem. The recent hot stock Applovin is using AI-driven predictive modeling to help advertisers place ads more efficiently.

The benefits of this are self-evident. Pitchers don’t even need to create advertising plans or watch the market all day. They can just let the system automatically run and optimize. They just need to check the system’s results every day.

However, a big pain point for advertisers is that they may find it difficult to understand the logic of AI bidding. The data is black-boxed within the platform, reducing transparency. If advertisers cannot fully understand the decision-making logic of AI, it will be difficult for them to adjust their strategies in time when the effect of their advertising drops. If they let AI make adjustments, they may misjudge the data and continue to bid high for irrelevant users, resulting in low conversion rates and wasted budgets.

AI's low-quality content quietly "steals traffic", "raises small accounts", and "eats up budgets"

Let me digress a little - in the Internet advertising industry, "the dog pays for the wool and the pig pays for it" has long been a common business model - you produce content at a cost of n yuan to attract traffic or directly attract traffic, and sell this traffic (user clicks) to advertisers at a price of m yuan. As long as m is greater than n, you are profitable. The previous Qutoutiao used gold coin incentives to attract traffic and then monetize it through advertising based on this logic.

Now AI can be said to have hit the floor directly. According to the calculation of IT technology blogger Ruan Yifeng, it only costs 0.00138 yuan to generate an article using a domestic AI model, and the advertising revenue for a single view is about 0.00145 yuan. According to a report by Silicon Star, an account agency that publishes fake news about "sudden explosion in a certain city" can generate 4,000 to 7,000 fake news articles a day at its peak, with a daily income of more than 10,000 yuan. The actual controller of the company, Wang Moumou, runs 5 such agencies and operates 842 accounts.

Some black and gray production groups are using AIGC large models to batch generate homogeneous materials. In the past three months, Xiaohongshu has dealt with more than 3.2 million black and gray production accounts, clearly pointing out the chaos of AIGC large model account raising. According to Toutiao's "2024 Annual Governance Report", the platform intercepted more than 930,000 low-quality AI content throughout the year and punished more than 7.81 million homogeneous posts.

The above is the perspective of traffic owners, and let’s shift the focus back to the perspective of advertisers who pay the money. Yes, the budget that everyone has worked so hard to squeeze out of their teeth may just be spent on low-quality AI content.

Triggering legal disputes in areas such as intellectual property rights and data privacy

Of course, the biggest risk of using AIGC in marketing is the unclear intellectual property and copyright situation.

"Shen Xiang" once had in-depth conversations with lawyers from relevant fields. Existing precedents cannot be directly applied to all situations. The basis for determining copyright ownership is the proportion of human intelligence and originality reflected in the entire process of AI-generated content. The more precise the user's training and the more options there are, the greater the possibility of ultimately obtaining the copyright of the generated content.

Therefore, brands should (at least for now) use generative AI for research, inspiration, or fine-tuning, but not blindly use AI-generated text, images, or audio content directly.

At the same time, some financial institutions may use customer data beyond the scope of customer authorization when using AI technology for marketing, or share data with third parties without customer consent, thereby causing legal issues.

Challenges for organizations: How to hold AI accountable when it makes mistakes?

The above are all business-level discussions, but we cannot ignore the challenges that AI brings to advertisers in terms of organizational management after it deeply penetrates into marketing work. You may be facing or will face the following confusions:

• With consumer data in hand, is it necessary to build a big model yourself? If you buy a ready-made big model service, how do you accumulate “AI assets”?
• How to establish a reasonable human-machine collaborative workflow? Which businesses can be handed over to AI, and should they be supervised by humans? How to set performance evaluation goals for AI and its collaborators? How to hold AI accountable when it makes mistakes?
• With AI as a helper, is there still value for the “Party B”? If so, what is it?
• AI-driven marketing requires close cooperation across departments, such as collaboration between data teams, technical teams, and marketing teams. However, departmental barriers and poor communication may hinder the effective application of AI technology. Are brands determined to "refresh the organization", break vested interests, and adapt the organization to the model of AI collaboration?
• How to ensure that the team thinks in unison on "AI issues" to reduce conflicts and bridge disputes?
• …

Having said that, I am not trying to pour cold water on AI marketing. In fact, AI has already penetrated into every aspect of the marketing industry. We expect AI to bring more highlights to creative work in the future, to free optimizers from tedious infrastructure, and to truly improve efficiency and quality for brands, saving effort and money.

But before the good things come, it is equally important to remain "cautious".

Author: Yalan