EU law harmonises training for doctors, yet nothing comparable stands behind anyone selling AI advice. Rising business adoption, a literacy duty that needs no certificate and consumer law that acts after the fact explain the crowd. For European buyers the sounder test is neither title nor platform but a record a stranger can verify, which some YouTubers hold and many consultants lack.
Table of Contents
A doctor qualified in one member state has that qualification recognised automatically in another, because EU law harmonises minimum training for doctors. Nothing comparable stands behind anyone who sells AI advice in the same single market. The European Commission says organisations need no certificate to meet their AI literacy duty, and no source reviewed sets an entry test for AI experts. Meanwhile 20% of EU businesses used AI in 2025, up from 13% a year earlier. The evidence points to a plain cause. AI experts multiply because the title is free, buyers are many and checks arrive only after a complaint. The popular comparison with surgeons is right about the missing gatekeeper and wrong about the medium, because a YouTube channel is not the defect. An unverifiable record is.
One profession with a harmonised gate
EU law builds a gate for doctors that works across borders. Directive 2005/36/EC gives seven professions automatic recognition on the basis of harmonised minimum training: nurses, midwives, doctors, dental practitioners, pharmacists, architects and veterinary surgeons. Other regulated professions, among them teachers, translators and real estate agents, pass through a general recognition system. The Commission keeps a non-exhaustive database of regulated professions, which this analysis did not search role by role, and no source reviewed lists AI expert as a regulated title. The result is an asymmetry inside one market. A surgeon’s training is checked against a common standard before the qualification travels, while a self-described AI strategist can pitch to a firm in any member state with no comparable test behind the title. That is an absence in the evidence, not proof that no member state regulates any AI-related activity, but it frames everything that follows: where checking exists, it happens later and elsewhere.
A literacy duty with no certificate
The AI Act does create demand. Article 4 has required providers and deployers of AI systems to support the AI literacy of staff since 2 February 2025, and national market surveillance authorities have supervised the rule since 2 August 2026. The Digital Omnibus on AI, in force since July 2026, amended the duty so that no specific or sufficient level is mandated. The Commission’s guidance says no strict requirements or mandatory trainings are imposed, no format is prescribed, and an internal record of measures is enough, so no certificate is needed. The Commission and member states are to publish compliance examples, and the AI Board is to adopt recommendations that build on existing competence frameworks. Until then, a company that must act has no official yardstick for choosing a trainer. That is an inference from the guidance, not a finding in it, yet it matches the Commission’s own caution that copying listed practices does not automatically grant a presumption of compliance.
Buyers who cannot judge the seller
Eurostat’s 2026 edition of Digitalisation in Europe supplies the demand side. In 2025, 20% of EU businesses used AI, up from 13% in 2024, but unevenly: 55% of large businesses against 19% of SMEs, and 42% in Denmark against 5% in Romania and 8% in Poland. Across the Union, 35% of recent internet users had used generative AI tools, yet 40% of EU citizens lack basic digital skills. The Union also targets at least 20 million ICT specialists by 2030 and counted just over 10 million in 2025. Buyers new to AI, especially in SMEs, often lack in-house ability to test a seller, which is an inference, since Eurostat measures use and not the purchase of advice. Profile data hint at the supply incentive, though they are American: a 2026 NBER working paper found that 19.7% of established US LinkedIn users retroactively edit past jobs, and Revelio Labs reports AI terms added to old roles peaking above five times baseline. No European equivalent was found.
Consumer law that polices claims afterwards
Europe’s sharpest tool against a false expert is the Unfair Commercial Practices Directive. Article 6(1)(f) treats a practice as misleading when false or deceptive information concerns a trader’s qualifications, status or approval and is likely to change a consumer’s decision. Article 12 lets courts and authorities require a trader to furnish evidence for factual claims and treat the claims as inaccurate when the evidence is missing or insufficient. The Commission’s guidance covers influencer marketing, endorsements and reviews, while a separate directive mainly protects businesses buying from other businesses. Because firms, not consumers, buy most corporate AI advice, the consumer rules reach only part of the market, which is an inference from how the two directives divide their scope. The law polices a claim after a complaint, not a credential before a sale. Recent practice targets providers: on 30 April 2026 Italy’s AGCM closed investigations into DeepSeek, Mistral AI and NOVA AI with commitments on hallucination warnings and no finding of infringement. This analysis found no EU action against an individual adviser’s credentials, an absence in a limited search. In the US the FTC announced five AI-related actions on 25 September 2024, including an allegation that Ascend Ecom took at least $25 million.
Confidence that outruns accuracy
Self-rating is a weak filter even for sincere people. A Finnish-led study with German co-authors, published in Computers in Human Behavior, gave participants 20 logical reasoning problems from the Law School Admission Test and let them use ChatGPT. Their scores rose by 3 points against a norm population, yet they overestimated their performance by 4 points. Higher AI literacy correlated with lower metacognitive accuracy: participants with more technical knowledge of AI were more confident and less precise about their own results. The familiar Dunning-Kruger pattern disappeared, and a second study of 452 participants replicated the findings. Its authors sit at Aalto University in Finland and at universities in Munich, Bayreuth and Berlin, so the result comes from a European research group, though it does not say where participants lived. The research covered ordinary users on one task, reports an association rather than a cause, and says nothing directly about people who sell AI advice. As an inference, it removes the comfort of judging an expert by fluent confidence, because someone who sounds sure and has used the tools heavily may still misjudge their own results.
Credentials do not stabilise forecasts
Verified credentials do not settle the question either. The largest survey of its kind drew on 2,778 researchers who had published at top AI venues such as NeurIPS, ICML and ICLR, and its aggregate answer gave a 50% chance that unaided machines outperform humans at every task by 2047, which is 13 years earlier than the answer to a similar survey run a year before. If real specialists move their central estimate by more than a decade between two surveys, a confident date from a self-titled commentator deserves less weight than the headline suggests. The same group put a 50% chance on all human occupations being fully automatable as late as 2116, and between 38% and 51% gave at least a 10% chance to outcomes as bad as human extinction. A credential shows what someone has built, not that their forecast will prove right. A sensible reader treats a researcher’s account of what exists, what was measured and how it fails as evidence, and a prediction about dates or jobs as one more opinion to test.
The channel is not the defect
The surgeon comparison deserves a partial defence and a partial correction. It holds where it points at gatekeeping. It fails where it blames the medium. Andrej Karpathy, a founding member of OpenAI and former Director of AI at Tesla, designed Stanford’s first deep learning course during his PhD, lists peer-reviewed papers from CVPR, ICML and ICLR on his own page, and publishes lectures on YouTube. A viewer can check each of those claims without trusting him. The test is whether a stranger can verify the record, not which platform hosts the explanation. The Commission’s consumer guidance expressly covers influencer marketing, so a creator who sells a course can be treated as a trader making claims; that is this analysis’s reading, not the guidance’s wording. The stakes argument has force too. A wrong tip about prompts costs little, but the Commission notes that deployers of high-risk AI systems must still train staff for human oversight, and there a bad adviser can cost far more.
Signals a stranger can check
| Signal | Checkable by a stranger? | Basis in the sources |
|---|---|---|
| Licence or certificate | Not for AI advice | The Commission says no certificate is needed under Article 4, and AI expert is not among the seven automatically recognised professions |
| Papers and public code | Yes | Karpathy’s page lists conference papers and code projects |
| Testimonials and results claims | No | Article 12 lets authorities demand evidence, but a buyer cannot compel it before paying |
| Forecasts of dates and jobs | Only after the date passes | The researcher survey moved its human-level estimate 13 years in a year |
Only the second row lets a stranger verify a claim without trusting the speaker, and a YouTube channel, a consulting site or a conference talk can display that row equally well.
Three decisions for European buyers
A manager at an SME hiring an AI consultant should ask for one named system the consultant built or one evaluation they ran, failure cases included, call a former client, and have any promised results written into the contract. An organisation designing its Article 4 programme should buy training for the systems it actually deploys and keep an internal record, because the Commission says no certificate is needed and no format is prescribed. A vendor claiming the Act mandates a particular course contradicts that guidance, and deployers of high-risk systems carry the extra oversight-training duty. A consumer weighing a paid AI course should treat each results claim as a hypothesis, save the advertisement, and know that national consumer authorities can ask the seller for evidence under national rules implementing Article 12. Verification costs time, so scale it to the stakes: a cheap test for a prompt, a reference call for a contract.
Developments that would change the verdict
The evidence supports one firm conclusion and leaves one question open. The firm conclusion is that the title costs nothing in the EU, demand is legally and commercially sustained, and enforcement works through complaints rather than entry checks. The open question is whether self-titled advisers actually give worse advice than credentialed ones. No source reviewed measures that, so the verdict rests on incentives and calibration evidence, not on a direct comparison. Three developments would change it. If the AI Board’s recommendations under Article 4(3), or the EU AI Skills Academy that began operating on 1 May 2026, produce assessable programmes, a public benchmark for literacy training could form, though not for individual experts. If a national authority ties an incident to missing training, which the Commission says would make sanctions more likely, buyers will start asking for proof. If a consumer authority acts against an individual adviser’s credential claims, the cost of an unsupported claim will rise. The Commission’s ongoing review of EU consumer law on digital fairness, of which the directive is part, is a venue to watch, although no source reviewed proposes a rule for advisers. Until one of these happens, checking the record is the buyer’s job.
Author: Jan Bielik CEO & Founder of Webiano Digital & Marketing Agency
This article is an original analysis supported by the sources cited below
AI Literacy – Questions & Answers
European Commission guidance on Article 4 of the AI Act, establishing that no certificate or mandatory training is required, how the Digital Omnibus changed the duty, who enforces it from 2 August 2026, that high-risk oversight training remains and the AI Skills Academy start.
Digitalisation in Europe – 2026 edition
Eurostat interactive publication supplying the 20% and 13% business AI-use figures, the country and size splits, 35% generative AI use, 40% without basic digital skills and the ICT specialist counts and 2030 target.
Recognition of professional qualifications in practice
European Commission page establishing the seven professions with automatic recognition on harmonised minimum training, the general system for other regulated professions and the non-exhaustive regulated professions database.
Directive 2005/29/EC on unfair business-to-consumer commercial practices
EUR-Lex text of the Unfair Commercial Practices Directive, supplying Article 6(1)(f) on misleading claims about a trader’s qualifications and Article 12 on evidence for factual claims.
Unfair commercial practices directive
European Commission overview establishing that the guidance covers influencer marketing, endorsements and reviews and that Directive 2006/114/EC mainly protects businesses in business-to-business advertising.
The Italian Competition Authority secures transparent information on hallucination risks from AI companies DeepSeek, Mistral and NOVA AI
AGCM press release of 30 April 2026 documenting three investigations closed with commitments and no finding of infringement.
AI makes you smarter but none the wiser: The disconnect between performance and metacognition
Aalto University research record for the peer-reviewed Computers in Human Behavior article, supplying the 246-person and 452-person studies, the 3-point gain, the 4-point overestimate and the AI literacy correlation.
Thousands of AI Authors on the Future of AI
Survey of 2,778 published AI researchers supplying the 2047 and 2116 forecasts, the 13-year shift from the prior survey and the extinction-risk response range.
Andrej Karpathy
Karpathy’s own page, supplying his OpenAI and Tesla roles, the Stanford deep learning course, his listed conference papers and his YouTube lectures as a checkable record.
Time Travel on Professional Profiles
NBER working paper 35546 establishing that 19.7% of established US LinkedIn users retroactively edit past jobs and that AI-related language in those edits rose sharply after 2022.
Revisionist (Employment) History
Revelio Labs summary of the same US research, supplying the peak of more than five times baseline for AI terms in edits to old jobs.
FTC Announces Crackdown on Deceptive AI Claims and Schemes
Federal Trade Commission press release of 25 September 2024 describing the five Operation AI Comply actions and the allegation that Ascend Ecom took at least $25 million, used here as the US contrast.
| Citing this article? Brief excerpts are welcome. Please credit Webiano.digital, name the author where stated, and include a link to https://webiano.digital and to this original article. Full or substantial republication requires prior written permission. Read our Copyright and Content Use Policy. |
This article was prepared with the assistance of artificial intelligence tools. The content underwent expert human review, and Webiano Digital & Marketing Agency assumes editorial responsibility for its final version and publication.















