A supplier presents you with an "intelligent" tool: automatic request sorting, prospect scoring, a writing assistant, sales forecasts. The price is high and the pitch is appealing. But is it really AI, and does it actually do what you are being promised?
The phenomenon has a name: AI washing, that is, dressing up a product that contains no AI, or hardly any, in "artificial intelligence" branding. It is costly for buyers: overpriced subscriptions, disappointing results, and data handed to third parties without understanding it.
This guide gives you the framework we use in our training to assess an offer, in eight questions, along with the contract clauses to check. For the basics (what is AI and what is not), start with our guide to recognising AI.
questions to assess an AI supplier
trial on your own data before any signature
The eight questions to ask the supplier
- 1. What exactly does the AI do, and what does the product do without it? — A precise answer ("it sorts your incoming emails into five categories") is a good sign. A vague answer ("it optimises your processes") is a warning sign.
- 2. Which model runs behind the scenes? — A model developed by the supplier, or access to a service from a major player (OpenAI, Google, Anthropic, Mistral…)? Both are legitimate, but you need to know which one, because it changes the price, the dependency and how your data is handled.
- 3. What data was it trained on? — Public data, the supplier's data, its customers' data? Will yours be used to train it?
- 4. Where and how is my data processed? — Hosting country, subcontractors, retention period. For personal data, this is a GDPR requirement.
- 5. What is the error rate, and how is it measured? — A genuine system has a quantified error rate, measured on difficult cases. "It never makes mistakes" is not a credible answer.
- 6. Who checks, and how do we correct an error? — Can a human easily take back control? Is there a record of what the AI decided and why?
- 7. Can I try it on my own cases? — With your documents, your customers, your difficult cases, not on a prepared demonstration.
- 8. What happens if I leave? — Data recovery, export, deletion, end of contract. A supplier confident in its product is not afraid of reversibility.
Warning signs during the demo
- The demo only works on its own examples — Suggest an unexpected case: if the supplier refuses or dodges, be wary.
- Humans behind the screen — Some "automatic" services are actually handled by hand by operators. Ask how requests are processed at volume.
- Marketing vocabulary with no mechanism — "Proprietary intelligence", "exclusive algorithm", "cutting-edge AI" with no explanation of how it works.
- A price that doubles with the "AI" option — Ask what it concretely brings.
- Testimonials without figures — "Our customers love it" proves nothing. Ask for a before/after measured indicator, and for the possibility of contacting a customer.
A good automation without AI is better than poorly designed AI. The goal is not to buy AI at any price: it is to solve a specific problem, at the right price, with a measurable result.
Contract clauses to check
- Data ownership and use — Your data remains yours and is not used to train models for other customers without your written agreement.
- Subcontractors and hosting — List of subcontractors, data location, GDPR-compliant data processing agreement.
- Performance or service commitments — Availability, service level, what happens in the event of a serious AI error. Without a measurable commitment, the salesperson's promises are worth nothing.
- Liability — Who bears the consequences of an AI error? Is the liability cap reasonable given the risk?
- Reversibility — Export of your data in a usable format, certified deletion at the end of the contract.
- Compliance — Is the supplier able to document its system against the European AI regulation, in particular if the tool is used for recruitment, for scoring individuals or for access to a service? See our article on the AI Act for Belgian SMEs.
For an important contract, have it reviewed by a specialist lawyer: this guide does not replace legal advice.
What the law says about false promises
A false statement about the characteristics of a product or service can amount to misleading commercial practices under Belgian consumer and competition law. Abroad, regulators have already penalised companies that claimed to use AI without genuinely doing so. A buyer is therefore well advised to keep in writing the promises made before signing (offer, emails, presentation), as they serve as evidence in the event of a dispute.
And what if you are the one selling AI?
The rule applies the other way round too: only present as AI what really is AI, and describe its role honestly. It is an increasingly sought-after trust argument among informed buyers. A company that simply says "this module uses a language model to summarise your documents, it can make mistakes, and you stay in control" inspires more confidence than a promise of magic.
This framework is part of our "recognising AI" training: applied to your own tools and to the offers you receive. See the programme →
Your questions about AI washing
How can you tell whether software really uses AI?
Ask what exactly the AI does, which model is used, what data it learns from and what its error rate is. A supplier who cannot answer these questions is probably using simple automatic rules.
Is AI washing illegal?
Falsely presenting a product as using AI can amount to a misleading commercial practice. Regulators in several countries have already penalised such cases. If in doubt, have a lawyer review the situation.
Is it better to buy AI or automation?
It depends on the problem. Rule-based automation is simpler, more predictable and cheaper for repetitive, well-defined tasks. AI is useful when cases are varied, unpredictable or involve language and images.
What should you do if you have already signed up for fake AI?
Gather the written promises, compare them with what has been delivered, ask the supplier for written explanations and have a lawyer review the contract. A well-argued complaint sometimes leads to a goodwill gesture or to termination.
Assess your tools with an independent eye.
We train your teams to ask the right questions and spot false promises.