The compliance problem nobody priced in when AI content went 10x

The pitch for generative AI in marketing was always about volume. Produce more variations, more localizations, more channels, faster. The numbers suggest brands took it seriously: in Salesforce’s State of Marketing research, the share of marketers using generative AI in at least one workflow rose from 51% in 2024 to 87% in 2026, with content creation among the most common uses (as summarized in industry reporting). When a team that once shipped 50 assets a quarter starts shipping 500, something has to absorb that increase. Usually it is the part of the process nobody costed: review.

This is the line item that did not make it into the business case. Drafting got cheaper. Checking did not. And checking is where legal exposure lives.

The volume problem is a review problem

Compliance, legal, and brand-safety review were designed for a slower pace. A human reads the claim, checks it against what the company can prove, flags the disclosure, signs off. That model works at the throughput of human writers. It does not work when the writing step is automated and the reviewing step is not. The bottleneck simply relocates from production to approval, and the queue gets longer.

The tempting response is to review a sample and wave the rest through. The problem is that liability does not scale down with your sampling rate. Each unchecked asset is a separate published statement, and regulators evaluate published statements one at a time, not as a percentage of your output.

What the rules actually say

In the United States, the Federal Trade Commission finalized its Rule on the Use of Consumer Reviews and Testimonials, announced on 14 August 2024 and effective 21 October 2024. It expressly prohibits creating, buying, or disseminating fake or AI-generated consumer reviews and testimonials that misrepresent the reviewer’s identity or experience, and it carries civil penalties (FTC announcement). The Commission has also acted against an AI tool itself: in its 2024 order against Rytr, the FTC targeted a service whose AI-generated review feature could produce false and deceptive testimonials (FTC press release).

Separately, the FTC has warned brands to substantiate AI-related claims. Its guidance is direct: marketers must be able to back up what they say their AI does, and the agency does not treat “artificial intelligence” as harmless puffery (FTC business guidance, “Keep your AI claims in check”). Through its “Operation AI Comply” initiative, the FTC has continued to pursue deceptive AI marketing across industries.

In the European Union, the AI Act entered into force on 1 August 2024 and applies in phases. Its transparency rules, set out in Article 50, become applicable on 2 August 2026; they require, among other things, that AI-generated synthetic content be marked in a machine-readable format and that deployers disclose deepfakes and certain AI-generated text published to inform the public (Article 50, EU Artificial Intelligence Act; European Commission overview). The penalties are not trivial: under Article 99, breaches of the prohibited-practices provisions can reach 35 million euros or 7% of total worldwide annual turnover, whichever is higher, with lower tiers for other violations (Article 99, EU Artificial Intelligence Act).

India is moving in the same direction through self-regulation. The Advertising Standards Council of India has issued draft guidelines for responsible labelling of AI-generated content in advertising, built on a risk-based framework: content that fabricates endorsements or uses deepfakes and unauthorized likenesses is treated as non-compliant even if labelled, while medium-risk uses such as synthetic influencers require clear disclosure (analysis of the ASCI draft guidelines). These build on ASCI’s existing influencer-disclosure code.

On top of all this sit the sector rules that already governed claims before AI arrived. Finance, health, pharmaceuticals, and food carry their own substantiation and disclosure regimes, and AI does not create an exemption. A misleading therapeutic claim is misleading whether a copywriter or a model produced it.

The substantiation gap

The deeper issue is that a language model generates fluent, confident text without any knowledge of what your company can actually prove. It will happily write “the most trusted,” “clinically proven,” or “guaranteed results,” because those phrases are statistically common in marketing copy. It has no access to your evidence file, your regulatory approvals, or the specific wording your legal team negotiated.

That is the substantiation gap. The model produces claims; the obligation to back them with evidence stays with the brand. At 10x volume, the gap widens proportionally, because every fluent unsubstantiated claim is a candidate for an enforcement action or a competitor complaint.

What a pre-publish approach looks like

The structural answer is to move compliance from a manual gate at the end to a check that runs before anything is published, at the same speed content is produced. In practice that means a few things working together: claims are checked against an approved evidence base rather than a reviewer’s memory; superlatives and regulated terms are flagged automatically; required disclosures (AI involvement, material connections, sector-specific notices) are applied consistently; and the relevant market’s rules, which differ across the FTC, the EU, and ASCI, are resolved per asset rather than assumed to be uniform.

This is the category some tools now address directly. kbie.ai, built by Kapis AI Tech Private Limited, is one example of a pre-publish compliance approach: it checks AI-generated content against a brand’s own knowledge and against applicable advertising rules before publication, so that scaling output does not mean scaling unreviewed risk. The point is not the tool but the principle: review has to become a property of the production pipeline, not a meeting that happens afterward.

Scaling content production is a reasonable goal. The mistake is treating review as a fixed cost that the old process can still absorb. It cannot, and the regulations above are not waiting for marketing operations to catch up.

This article is informational and not legal advice. Confirm your obligations with qualified counsel for your markets and sector.

FAQ

Does the FTC’s fake reviews rule apply to AI-generated content?

Yes. The FTC’s final rule, effective 21 October 2024, expressly prohibits creating, buying, or disseminating fake or AI-generated consumer reviews and testimonials that misrepresent the reviewer’s identity or experience, and it carries civil penalties (FTC).

When do the EU AI Act’s transparency rules for AI content take effect?

The transparency obligations in Article 50, including marking AI-generated synthetic content and disclosing deepfakes, become applicable on 2 August 2026 (EU Artificial Intelligence Act, Article 50).

Do I need to disclose when an ad uses AI in India?

Under ASCI’s draft labelling guidelines, medium-risk uses such as synthetically generated influencers or settings require clear disclosure, while certain high-risk uses such as fabricated endorsements and deepfakes are treated as non-compliant even if labelled (ASCI draft guidelines).

Who is responsible if an AI tool generates an unsubstantiated claim?

The brand publishing the claim is responsible for substantiating it. The FTC has made clear that marketers must be able to back up claims about their products and about AI itself, regardless of how the copy was produced (FTC, “Keep your AI claims in check”).