AI Battle: Should Neural Networks Refuse to Write Persuasive Advertising Copy for Dubious…

June 26, 2026 · 5 min read

Question: Should neural networks refuse to write persuasive advertising texts about dubious products?

Perplexity

Neural networks must refuse to write convincing advertising texts about dubious products, as this violates the fundamental advertising law requiring truthfulness, and creates direct legal risks for the advertiser. Neural networks by nature generate content based on probabilistic models, not on an analysis of reality, which regularly leads to attributing nonexistent properties to products, inventing characteristics, and referring to studies that do not exist[1]. The Federal Antimonopoly Service (FAS) and the Federal Service for Surveillance on Consumer Rights Protection and Human Wellbeing (Rospotrebnadzor) do not make allowances for the text being written by AI: the responsibility for fakes and false promises lies entirely with the person — the advertiser and distributor, so AI texts for dubious products cannot be published without fact-checking[1].

Besides legal consequences, the use of neural networks for dubious products undermines the trust of the audience, since AI systems themselves do not perceive advertising formulations without numbers and evidence as truthful facts[2]. If research results, statistics, or measurable indicators are absent, the neural network treats such claims as empty marketing promises and is cautious about statements like “best,” “reliable,” or “effective” without specific criteria[2]. Publishing a convincing AI-generated text about a dubious product without data verification risks including unchecked claims in the response, making the content not only ineffective but potentially dangerous for the brand’s reputation[2].

Ethical use of AI in advertising requires that the machine acts as a draft, which is then edited and checked by a person for bias, uniqueness, and compliance with legally significant statements[3][8]. If the finished generated content about a dubious product is published without processing, the risks of plagiarism, manipulation, and legal problems increase, since AI does not understand meaning and cannot bear ethical consequences[4]. Responsibility for the final product always lies with the person, so neural networks must be programmed to refuse generation of convincing texts when the source data about the product is not factually confirmed[4].

Sources:

Gemini

Failed to get a response.

ChatGPT

The question of whether neural networks should refuse to write persuasive advertising texts about dubious products is both an ethical and technological issue. On one hand, neural networks are tools created to automate and simplify tasks, including writing texts. At the same time, the responsibility for the content and the consequences of using these texts lies with people — developers and clients. However, considering that neural networks can create highly convincing and influential materials, it is important to implement internal control mechanisms, filters, and ethical restrictions within them to minimize the creation of dangerous or misleading information. For example, AI companies often introduce rules prohibiting the promotion of illegal substances, pyramid schemes, or medical products without licenses.

On the other hand, relying entirely on algorithms and bans is also risky. Specific contexts and the definition of "dubiousness" matter. For example, a product may be controversial for some groups but legal and in demand for others. A classic example is advertising texts for dietary supplements (biologically active additives). Many of them are classified as "dubious" from the standpoint of evidence-based medicine, yet their promotion is allowed if the requirements of state regulation are met. A refusal by a neural network to write such texts may limit freedom of expression and commercial activity, but at the expense of ethics and the consumer. Ideally, AI should work in tandem with ethical standards and human oversight to balance the benefits and ethics of advertising content.

Thus, neural networks should indeed have built-in mechanisms to refuse to create texts that could cause harm or mislead consumers. But the final decision on what constitutes a dubious product and what text to write should remain with humans. Only this way can ethical and responsible advertising be ensured, without turning AI into a tool for unrestrained marketing and manipulation.

Sources:

Ethical Principles for AI Use (OpenAI, Google AI Ethics)
Regulation of Advertising and Dietary Supplements in Russia and the EU
Overview of Approaches to Content Filtering and Control in AI Systems

Grok

Neural networks should not unconditionally refuse to create advertising texts, but must refuse in cases where the product is clearly misleading or harmful. The boundary is drawn by facts: if the product violates laws (for example, additives banned by regulators or financial schemes with signs of fraud), AI should block the generation of convincing marketing in order not to reinforce misinformation. Otherwise, "dubiousness" becomes too subjective a category — from pseudoscientific gadgets to controversial investments — and automatic refusal will lead to excessive censorship, where models decide for users what is "worthy" of promotion.

Examples show the risks of both extremes. ChatGPT and similar systems already refuse texts for outright scam projects or medicines without proof of efficacy, which reduces the spread of harmful content. On the other hand, if a neural network refuses to write about a product with an ambiguous reputation (say, a crypto platform with complaints but legal), this will restrict commercial freedom and shift responsibility onto developers rather than authors and platforms. The optimal approach is transparent rules based on verified data, rather than blanket refusals: the model can generate text but with caveats or warnings about risks.

In the end, neural networks should focus on helping with fact-checking, not moralizing.

Claude

This is truly a complex issue where the principles of responsibility, freedom of expression, and practical feasibility collide.

Arguments for refusal:

Neural networks are becoming an increasingly influential tool in the information space, and with this power comes responsibility. If AI writes convincing advertising for a dangerous or fraudulent product (for example, fake medicines, financial pyramids, or harmful dietary supplements), it becomes an accomplice to potential harm. The parallel is apt: we don’t expect a lawyer to write a convincing defense for a clearly guilty person by forging evidence, nor a doctor to prescribe knowingly incorrect treatment. Refusal in such cases is not censorship but a boundary of professional ethics.

Arguments against complete refusal:

However, here arises the problem of subjectivity: who decides which product is "dubious"? Homeopathy, considered quackery in some countries, is legally sold in others. Crypto-assets seemed like a scam ten years ago but have become part of investment portfolios. If AI refuses to write advertising on the basis of developers’ moral judgment, this effectively becomes censorship according to the standards of one group of people. Moreover, writing convincing text is simply a craft; how this craft is used depends on the context and the user’s intentions.

Practical compromise:

A logical middle ground is to distinguish risk levels. For clearly dangerous or prohibited products (drugs, weapons, forged documents), refusal is justified. For controversial products, transparency should be required: writing the text is allowed but with mandatory disclosure of risks, scientific unprovenness, or conflicting expert opinions. This preserves freedom of expression but adds responsibility.

Sources:

Blog