There is a clear line running through the latest wave of AI enforcement: regulators are becoming less interested in whether a product sounds futuristic and far more interested in whether the company behind it can prove what it says. The Federal Trade Commission’s action against Air AI and its final order against Workado matter not because they are isolated disputes, but because they go to the center of modern software and service marketing. One matter focuses on broad promises about business growth, earnings, and risk-free refunds. The other focuses on a polished performance metric that sounded objective but allegedly was not supported by evidence that matched real-world use. Together, they send a direct message. AI marketing is no longer being treated as a special zone where exaggeration is tolerated. It is being treated as advertising, and that means the old rules against deception, overstatement, and unsupported claims apply in full. ¹ ³
For Michigan SaaS companies, agencies, consultants, managed service providers, lead-generation businesses, and local service companies that have added AI into their operations or marketing, that shift matters now. It matters if a business is selling AI-assisted intake. It matters if it is promoting automated outbound calling, AI-generated content, AI analytics, AI scheduling, AI proposal drafting, AI review handling, AI voice agents, or AI assistants for internal teams. It matters if a website says a tool “replaces” staff, “guarantees” outcomes, “never misses” leads, or is “98% accurate” without clearly defining what that means, how it was tested, what data was used, and under what conditions the claim holds true. The lesson is not that businesses should stop using AI. The lesson is that businesses need to stop using AI language as a substitute for proof.
The Air AI matter is the more dramatic warning because it shows what happens when AI marketing moves from ambitious positioning into promises about money, labor replacement, and certainty. According to FTC’s case materials, the agency asked a federal court to stop Air AI from making allegedly deceptive claims about business growth, earnings potential, and refund guarantees directed at small businesses and entrepreneurs. That alone should get the attention of any company that sells AI-enabled revenue acceleration, business automation, or “scale without hiring” services. The regulatory concern was not simply that the product was branded with AI language. It was that the AI language was allegedly connected to promises about concrete financial outcomes and business performance. ¹
The complaint provides the details that make the case so relevant to ordinary SaaS and service-company marketing. According to the FTC’s allegations, Air AI represented that its conversational AI could conduct full ten-to-forty-minute phone calls that sounded like a real human, with “infinite memory” and “perfect recall,” and that it could do the work of a full-time agent. The complaint also describes claims that the system operated twenty-four hours a day, could be scaled instantly, and would outperform the usual human limitations associated with hiring, burnout, ramp time, and turnover. At the same time, the FTC alleged that some promised features were never delivered or were not available as advertised for certain purchasers. Those are still allegations rather than final judicial findings, but they illustrate the key point that Michigan companies should take seriously: once a seller moves from saying “this tool helps” to saying “this system replaces people and delivers predictable business results,” it is making claims that need evidence, not enthusiasm. ²
The earnings and refund allegations sharpen the warning even further. The complaint alleges that buyers were told they could make double or more than they spent in as little as three months, could potentially earn millions, and were protected by what the company described as an unusually bold refund guarantee. The FTC also alleged that many buyers did not make the promised money, were left in debt, and rarely received refunds despite repeated assurances that the purchase was effectively risk free. Whether the court ultimately agrees with every allegation will be determined through litigation, but the business lesson does not depend on the final outcome. The lesson is that AI branding does not reduce the legal risk associated with classic earnings claims, performance claims, or refund claims. If anything, AI language can intensify that risk because it can make ordinary sales promises sound scientific, inevitable, and systematized when they are not. ¹ ²
That matters because many companies do not think of themselves as making earnings claims when they say an AI product will “10x productivity,” “replace two reps,” “book meetings while your team sleeps,” or “turn your website into a 24/7 closer.” But regulators do not need to accept the marketer’s preferred wording. They look at substance rather than style. If a sales deck, webinar, demo, founder video, email campaign, or contract communicates that the AI will produce a certain level of revenue, staffing reduction, efficiency, or return on investment, the company is not operating in a free-form branding environment. It is making performance claims that may require substantiation, limitation, or both. Air AI is ultimately a warning about outcome inflation. It shows how quickly a pitch that sounds modern can look, from a regulator’s perspective, like a traditional deception case dressed in the language of machine intelligence.
Workado is the quieter but perhaps more broadly important matter for mainstream SaaS marketers because it concerns measurement discipline. According to the FTC’s final order and related case materials, Workado promoted its AI Content Detector as highly accurate, and the FTC required the company to stop advertising the accuracy or efficacy of its AI content detection products unless it possessed competent and reliable evidence showing the products were as accurate as claimed. The final order also imposed evidence-retention obligations, notice requirements, and compliance reporting. In other words, the agency did not merely object to tone or presentation. It imposed a framework centered on substantiation and accountability. ³
The specific allegations explain why the Workado matter matters so much. The FTC stated that Workado claimed its detector could classify text as human-created or AI-generated with more than 98 percent accuracy across many forms of text from systems such as ChatGPT, Claude, and GPT-4. The FTC also alleged that the company lacked competent and reliable evidence for that claim, that its model had been trained or fine-tuned to classify academic content rather than the marketing-oriented content users commonly submitted, and that the detector performed far worse on non-academic AI-generated text. That gap is the entire lesson. The problem was not simply that the number was high. The problem was that the benchmark allegedly did not match the product’s actual use case. ³
That lesson reaches far beyond AI detectors. It applies to any Michigan company marketing AI summarization, call scoring, lead qualification, routing, drafting, search, prediction, classification, fraud detection, sentiment analysis, or automation. If a claim is built on a dataset that does not resemble customer reality, then the supporting evidence may technically exist while still being commercially misleading. If a tool performs well in a controlled environment but not with messy customer inputs, domain-specific jargon, low-quality audio, regional accents, sparse CRM data, or mixed human-AI workflows, then a clean headline statistic may create legal risk rather than credibility. Workado shows what happens when language like “98% accurate” creates the impression of scientific certainty while the underlying support allegedly rests on the wrong testing environment.
For Michigan businesses, the temptation is to view these as federal outliers that affect only venture-backed startups or nationally known AI brands. That would be a serious mistake. The broader message is that ordinary businesses selling ordinary software and services can fall into the same pattern with far smaller budgets and far less dramatic branding. A regional agency can oversell its AI lead qualification service. A local home-services platform can oversell its AI voice receptionist. A legal-tech vendor can oversell AI document review. A healthcare-adjacent software company can oversell AI note generation or triage. A B2B SaaS firm in Detroit, Grand Rapids, Ann Arbor, Lansing, or elsewhere does not need to call itself an “AI company” to create AI advertising risk. It only needs to make an unsupported AI claim.
Michigan law makes that point more immediate. The Michigan Consumer Protection Act provides that unfair, unconscionable, or deceptive methods, acts, or practices in trade or commerce are unlawful, and the statute includes conduct that causes a probability of confusion or misunderstanding as to the source, sponsorship, approval, or certification of goods or services. Consumer protection materials from the Michigan Attorney General likewise emphasize that consumers should not be misled or unfairly treated when buying goods or services. That does not mean every B2B software dispute becomes a state enforcement action. It does mean that Michigan companies should not assume that AI branding creates unusual freedom to exaggerate. The same state-law principles that apply to misleading refunds, misleading performance claims, or misleading representations about what a service does can apply to AI-enabled offerings as well. ⁵
This is why the most important phrase in the Workado materials may be “competent and reliable evidence.” That phrase is not exciting, and that is precisely why it matters. Hype thrives in vagueness. Enforcement thrives in documentation. A business that can show how a claim was tested, when it was tested, what data was used, what limitations appeared, how results varied by use case, and how the claim was reviewed before publication is in a much stronger position than a business that arrived at its headline through a founder anecdote, a vendor’s marketing sheet, or a one-time internal demo. The FTC is not requiring magic. It is requiring substantiation. It is not asking whether an AI product is impressive. It is asking whether the advertising is true. ³
This is also where the National Institute of Standards and Technology AI Risk Management Framework becomes useful for companies that want a mature way to talk about responsible AI use. NIST describes the framework as a voluntary, rights-preserving, non-sector-specific resource intended to help organizations that design, develop, deploy, or use AI systems manage risk and support trustworthy use. NIST also emphasizes testing, evaluation, verification, and validation as important parts of operationalizing AI governance. That is not a legal safe harbor, and businesses should not market it as one. But it does offer a serious model for how companies can replace loose claims with disciplined process. ⁴
In practical terms, that means a business should know exactly what it is claiming before it tries to defend the claim. Saying “our AI helps intake teams respond faster” is very different from saying “our AI replaces human intake staff.” Saying “in a 60-day pilot with one customer, average first-response time decreased by 37 percent” is very different from saying “our AI increases conversion rates.” Saying “this model flags likely compliance issues for human review” is very different from saying “this system ensures compliance.” One claim describes assistance inside a bounded workflow. The other suggests guaranteed outcomes, legal certainty, or universal performance. The more absolute the language becomes, the heavier the proof burden becomes with it.
The trap for many service businesses is that AI hype often enters through sales culture rather than product design. The product team may understand the limits. The operations team may know where human review remains necessary. The founder may understand that the model struggles with edge cases. But the website may say the tool “never misses a lead,” the outbound emails may call it “human-indistinguishable,” the webinar may promise “predictable booked meetings,” and the contract may include a refund guarantee that nobody has operationally funded or consistently administered. At that point, the legal risk is no longer living in the software. It is living in the story being sold. Air AI shows how dangerous that becomes when aggressive claims about performance and earnings are combined with “risk-free” rhetoric. Workado shows how dangerous it becomes when a polished metric gives false comfort. ² ³
None of this means businesses need to strip all ambition from their marketing. It means they need to make their marketing more mature. The strongest AI marketing in the next few years will not come from companies that promise the impossible. It will come from companies that can describe, with precision, what a system does, for whom, under what conditions, and with what measured results. A company that says its AI voice agent handles after-hours call intake, captures standard fields, escalates certain issues to human staff, and reduced missed-call follow-up times during a documented pilot may sound less flashy than a company promising a fully autonomous closer. But the first company sounds credible. In an environment shaped by matters like Air AI and Workado, credibility is not weaker than hype. Credibility is stronger because it survives scrutiny.
The same principle applies to testimonials and case studies. If a client had an exceptional outcome after a custom implementation, substantial human oversight, clean data, and an unusually disciplined internal process, that story may still be useful, but only if the surrounding context does not imply that every customer will get the same result automatically. Likewise, if a customer doubled output after adopting a platform but also hired staff, improved pricing, rebuilt its offer, and tightened internal processes, it is misleading to let AI absorb all the credit. Good case studies do not merely display success. They explain the conditions that produced it. That is not timid marketing. It is trustworthy marketing.
Refund language deserves special attention as well. One of the overlooked lessons from the Air AI allegations is that refund promises can shift from a conversion device into an enforcement theme when they are broad in presentation but narrow, delayed, or impractical in actual operation. A company does not get extra room to overpromise because it says “results guarantee” in the same sentence as “AI-powered.” If anything, the AI framing may increase the buyer’s expectation that the seller has tested, validated, and systematized the promised result. If a business offers a refund, the conditions should be clear, objective, operationally workable, and consistent across the website, sales calls, contracts, onboarding documents, and support procedures. Otherwise, the guarantee may function less as reassurance and more as evidence that the company knew customers needed unusual comfort to accept the underlying claim. ²
What should a Michigan SaaS or service company actually do in response? The answer is not a slogan. It is a governance practice. A company should know which claims appear in which channels, who approved them, what evidence supports them, and when that evidence was last reviewed. It should distinguish between assisted-use claims and autonomous-use claims. It should distinguish between pilot data and production-wide performance. It should stop publishing accurate figures that are not tightly defined. It should stop using language like “human-like,” “human replacement,” “guaranteed,” “error-free,” or “risk-free” unless those terms are both necessary and provable. It should make sure that founder interviews and sales scripts do not casually outrun the product itself. It should treat benchmarking as an ongoing process rather than a one-time event, especially when models, prompts, customer inputs, and integrations are changing from month to month.
The NIST framework is especially useful here because it encourages organizations to think about AI as a lifecycle issue rather than a launch issue. That is exactly right. A claim that was fair six months ago can become misleading after a model update, a vendor change, a dataset shift, or a new use case. A claim that was accurate for academic content can become misleading when repurposed for marketing copy. A claim that was reasonable during a tightly supervised pilot can become inflated when presented as ordinary production performance. AI marketing cannot be governed as static copy because AI systems are not static products. If the underlying system changes, claim review must change with it. ³ ⁴
There is also a competitive advantage in taking this seriously. The AI market is entering a trust phase. Buyers increasingly understand that many systems are brittle, vendor-dependent, prompt-sensitive, and deeply context specific. That does not make AI weak. It makes honest proof more valuable. Companies that can demonstrate disciplined testing, sensible limitations, human-review boundaries, and measured outcomes will increasingly appear more sophisticated than companies that still sound like science-fiction trailers. The firms that win will not be the ones that say “AI” the loudest. They will be the ones that can answer the question that comes after the demo: how do you know?
That is the deeper meaning of Air AI and Workado for Michigan companies. Air AI is a warning against selling destiny, especially when AI is used to imply inevitable revenue, labor replacement, or no-risk growth. Workado is a warning against selling certainty through numbers that do not match the real use case. Michigan law adds a local reminder that deceptive trade practices are not excused simply because the vocabulary has changed. And NIST offers a practical framework for moving from hype to discipline. The common thread is straightforward. AI is not a legal category that floats above ordinary advertising rules. It is a technological layer inside ordinary commercial claims.
The companies that adjust early will be in the strongest position. They can still market boldly. They can still sell efficiency, automation, responsiveness, and measurable improvement. They can still build AI-first products and AI-enabled service lines. They simply need to do what serious businesses have always had to do when making consequential claims: define the promise, test the promise, document the proof, disclose the limits, and deliver what was promised. That is not anti-innovation. It is what real innovation looks like once the magic-show phase is over.
Contact Tishkoff
Tishkoff PLC specializes in business law and litigation. For inquiries, contact us at www.tish.law/contact/. & check out Tishkoff PLC’s Website (www.Tish.Law/), eBooks (www.Tish.Law/e-books), Blogs (www.Tish.Law/blog) and References (www.Tish.Law/resources).
Sources:
1- Federal Trade Commission, Air.ai, FTC Legal Library case page, updated August 26, 2025. https://www.ftc.gov/legal-library/browse/cases-proceedings/airai
2- Federal Trade Commission, Complaint for Permanent Injunction, Monetary Judgments, and Other Relief, FTC v. Air Ai Technologies, Inc., et al., filed August 25, 2025. www.ftc.gov/system/files/ftc_gov/pdf/AiraiComplaint.pdf
3- Federal Trade Commission, Content at Scale AI (Workado) case materials, including FTC final order and Analysis of Proposed Consent Order to Aid Public Comment, 2025. https://www.ftc.gov/legal-library/browse/cases-proceedings/2323092-content-scale-ai
4- National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, January 26, 2023. https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10
5- Michigan Legislature, Michigan Consumer Protection Act, MCL 445.903; Michigan Department of Attorney General, consumer protection materials. https://www.legislature.mi.gov/Laws/MCL?objectName=MCL-445-903