A few days ago, The Guardian published a report confirming what many of us suspected: PR executives say UK companies are pressuring them to present ordinary automations as artificial intelligence. It’s called AI washing, and it’s a growing problem that’s costing businesses of all kinds money, time, and trust.
If you own a business or lead operations, you’ve probably already noticed it: overnight, every tool you use is “powered by AI,” every vendor is “AI-first,” and everyone is promising you an intelligent transformation. The uncomfortable question is: how much of that is real?
In this article, we explain what AI washing is, why it’s exploding now, how to spot it, and what questions to ask before you sign a contract. The goal is simple: to make sure you don’t pay AI prices for something that’s really just a if/else with good presentation.
What AI washing actually is
AI washing is a deceptive marketing tactic in which companies exaggerate or fabricate the development or implementation of AI solutions to give the impression that technology alone can transform the business. The term is not new — it was first defined by the AI Now Institute, a research institute based at New York University, in 2019, and it derives from “greenwashing,” another deceptive marketing technique that similarly misrepresents a product’s environmental impact.
In practice, AI washing takes several forms:
- Renaming the old: an Excel macro, a Python script, or a rule-based automation suddenly sold as a “proprietary AI engine.”
- Using the word as decoration: using buzzwords like “smart” or “AI-powered” without the product actually delivering or using that capability.
- Attributing results to AI: presenting process improvements that come from good operational design as though they were the result of advanced algorithms.
- AI where it adds nothing: forcing an LLM into workflows that would be better solved with a spreadsheet.
Why it’s happening now (and why it affects you)
The reason is purely economic. The hype around generative AI has reached a fever pitch in recent years — to such an extent that it is virtually a death sentence for any company to say it is not investing heavily in generative AI to transform its business. The problem is that many companies are either deceiving or lying about the extent to which they are using that technology.
And it’s not a marginal phenomenon. A 2019 study by investment firm MMC Ventures found that 40% of new European tech companies describing themselves as “AI startups” were actually using almost no AI at all: it was purely a marketing pitch to help them attract investment capital. And that was before the ChatGPT boom. Today the proportion is almost certainly worse.
The cost to you as a customer is twofold:
- You pay more for the same thing: the “AI” label inflates the price of products that, without it, would cost half as much.
- You get locked into weak solutions: you buy something expecting real AI capabilities and end up with a rigid system that doesn’t learn, doesn’t adapt, and doesn’t scale.
As one industry expert puts it, AI washing “thrives in a climate where technological optimism is high, understanding is low, and oversight lags far behind innovation”. In plain English: the fewer people understand how AI really works, the easier it is to sell it with a fresh coat of paint.
The real risks to your business
Beyond the money spent, there are serious operational consequences:
| Risk | What happens in practice |
|---|---|
| False functionality | You buy a promise and receive a product that doesn’t do what it promised. |
| Vendor lock-in | You get tied to a closed platform you can’t migrate or adapt. |
| Regulatory risk | If you resell the service, you inherit the provider’s transparency problems. |
| Loss of internal credibility | The team stops trusting “AI projects” after a bad experience. |
| Opportunity cost | While you waste 6 months on a fake solution, your competition moves ahead with a real one. |
In fact, this not only puts companies at risk of litigation and regulatory scrutiny, but also unnecessarily increases their compliance costs by providing assurances about risks that neither the organization nor its technology actually pose. In other words: you end up paying to protect an AI that technically isn’t even there.
How to spot AI washing before you sign
Here’s the practical part. When a vendor tells you their solution uses AI, ask these questions. The answers will tell you far more than the brochure.
1. “What exactly is the AI doing here?”
The answer should be specific. Something like: “A language model classifies incoming emails by intent and routes them to the right department, and an embeddings model finds similar answers in our knowledge base.”
If the answer is vague (“our smart engine optimizes the process”), that’s a red flag.
2. “What happens if I remove the AI from the system?”
If the answer is “pretty much the same” or “it would work just as well with rules,” then you’re not buying AI. You’re buying automation with a new label. Which may be fine — but you should pay automation prices, not AI prices.
3. “Can I see the system working with my data?”
A generic demo doesn’t count. Ask for a test using one of your real cases. Real AI systems have observable behavior: they handle ambiguity, generate responses that weren’t prewritten, and adapt to unexpected inputs.
4. “What happens when it fails?”
A serious vendor will explain the failure modes: hallucinations, edge cases, latency. If they tell you it “never fails,” they either have never deployed it or they’re lying.
5. “What tools is it built with?”
You don’t need to understand every technical detail, but you should hear concrete names: specific APIs (OpenAI, Anthropic, open-source models), orchestration tools (n8n, LangChain), vector databases. Technical transparency is a sign of honesty. “Proprietary technology” with no further explanation is usually a sign of smoke and mirrors.
6. “Do I own the solution, or am I renting a black box?”
This is key. Companies are expected to be transparent and clearer when communicating the use of AI in their products or services. Consumers can mitigate this by asking companies for concrete proof of their use of AI tools. If you can’t see what’s inside, you can’t judge whether it’s worth the price.
The difference between real AI and decorative AI
To make it clear, here’s an honest comparison:
Traditional automation (not AI, and that’s fine)
- Fixed rules: if A comes in, do B.
- Works perfectly for predictable processes.
- Cheap, fast, reliable.
- Example: when an email with “invoice” in the subject arrives, move it to a folder.
Applied AI (this is real AI)
- Interprets language, context, and ambiguity.
- Handles cases that weren’t pre-programmed.
- Learns from examples or documentation.
- Example: read an email, understand that it’s an urgent complaint disguised as a polite inquiry, escalate it, and draft an initial response.
Both are useful. Both have their place. The problem is selling the first as the second.
How we do it at Studio SmartWork (no smoke and mirrors)
At Studio SmartWork, we build applied AI solutions for SMEs and mid-sized companies, and our way of working is designed specifically to avoid AI washing:
- Full technical transparency: we show you exactly what’s being built, with what tools, and how it works. No black boxes.
- Open source whenever possible: we use n8n and standard AI APIs. If one day you want to move the solution elsewhere, you can. We don’t lock you in.
- A demo with your data before you sign anything: if a solution doesn’t work with your real case, it’s not worth your money.
- Measurable results: we talk about hours saved, response times, and conversion rates. Not “intelligent transformation.”
- Implementation in 4–8 days: if something can be prototyped in a week, you already know it’s real. Smoke and mirrors don’t get delivered in 7 days with metrics.
Not every problem needs AI. Sometimes the right answer is a well-built traditional automation. Sometimes it’s a redesigned process. And sometimes, yes, you do need a language model. The difference is knowing which is which — and charging for what is actually delivered.
Conclusion: healthy skepticism, not paralysis
The existence of AI washing doesn’t mean AI isn’t useful. It means the market is saturated with offers, and you have to do the filtering. The good news is that filtering is easy when you know what to ask.
Three takeaways:
Don’t pay for the label, pay for the outcome. What matters is how many hours you get back, how many leads you handle, how much your response time drops. If those numbers don’t move, it doesn’t matter how much AI is inside.
Ask to see the engine. Any serious vendor can explain what their system does without falling back on buzzwords. If they can’t, look elsewhere.
Start small and measure. A well-built applied AI project shows results in days, not quarters. If someone asks for 6 months and half a million before showing you anything that works, there’s a 90% chance it’s smoke.
Real applied AI exists, works, and is within reach for any SME that knows exactly what it wants to automate. The trick is separating the real from the marketing — and fortunately, that’s getting easier if you know where to look.