Can you really make €10,000 a month with AI?

Introduction — a round number, turned into a sales pitch

€10,000. The figure keeps coming back in the adverts touting AI as a shortcut to financial independence: AI freelancing, an automation agency for small businesses, an app coded in a weekend with an AI assistant and sold on, a “faceless” YouTube channel or TikTok account run by AI, automated trading, or a course promising to hand over the method. The number is round, it sounds like a senior executive’s salary, and it is repeated by dozens of accounts followed by tens of thousands of people.

Le Recul has already tested three concrete versions of this promise: AI courses at €1,997, AI-automated sales prospecting, and AI-automated social media accounts. In all three cases, the same verdict: partly true, heavily oversold. This new article steps back. It cross-references those three tests with official data on the other methods sold under the same banner — AI freelancing and agencies, SaaS coded with AI, automated trading, monetisable content (ebooks, stock, blogs), and selling courses/prompts/derivative products — to answer the central question the adverts carefully avoid: concretely, how many people genuinely reach €10,000 a month, and at what price?

Test summary

What we checked

This test does not consist of launching our own AI freelance business, agency or SaaS for 90 days: that field work has already been done for three specific methods (courses, prospecting, social media), with published figures. Here, the question is different: across the whole set of methods sold under this banner, what do the official figures — not the testimonials, not the screenshots — say about the real probability of reaching €10,000 a month?

We cross-referenced:

• the French legal and tax thresholds (URSSAF, Légifrance) applicable to income of this size;
• the real incomes of the self-employed in France (INSEE, URSSAF);
• the rates actually charged on freelance platforms (Malt, Upwork, Fiverr);
• independent studies on the viability of SaaS products and apps built with AI’s help;
• the warnings and court rulings of financial regulators (AMF, CFTC, SEC) on automated trading;
• the policies of content platforms (Amazon KDP, Shutterstock, Adobe Stock, Google Search);
• the regulatory and judicial proceedings already brought against this precise type of promise (the FTC in the United States, the DGCCRF and French courts);
• economic studies published in peer-reviewed journals on AI’s real effect on self-employed incomes.

Where a marketing figure could not be cross-checked against a reliable primary source, we say so explicitly rather than repeating it as it stands.

The wall the adverts never show: the legal ceiling

This is the most solid point, the most verifiable, and the least mentioned by the people selling methods. Almost every “AI freelance”, “AI agency” or “AI side hustle” offer presents the micro-entreprise status (France’s simplified self-employed regime) as the natural starting point: no heavy accounting, set-up in a few clicks, simplified tax. But that status has a legal turnover ceiling — and €10,000 a month, that is €120,000 a year, goes well beyond it.

The calculation, in the most favourable case (invoicing up to the exact legal ceiling, with the flat-rate income tax option):

In other words: it is mathematically impossible to reach €10,000 a month of net income while staying under the micro-entreprise status for a service activity, whatever the talent or the day rate charged. The legal ceiling caps maximum net income at around €5,300 to €5,400 a month — a little over half the promise. A freelancer charging the top day rate observed on the AI market (€900 to €1,500 a day, see below) in fact exceeds that ceiling in fewer than 100 invoiced days, that is before half a full year has passed.

To exceed that threshold sustainably, you have to change status — sole trader on the standard regime, EURL or SASU (France’s single-member limited company forms) — with an entirely different social and tax mechanism. And that is where the promise hits a second wall, even less well known than the first: changing status does not guarantee reaching €10,000 a month net either.

EURL/SASU lines: simulations cross-checked on two independent professional calculators (tjmetre.fr and archipel-lyon.fr), at an invoiced turnover close to €120,000 a year — a level the micro-entreprise cannot legally reach, hence the comparison on two different bases. Corporation tax at 15/25% and the 31.4% flat tax on dividends (rates in force since 1 January 2026) are already included. The exact result depends heavily on the salary/dividend split chosen.

The finding is clear: at that level of turnover, the EURL can even turn out less advantageous than the micro-entreprise’s legal ceiling, and the SASU — often presented as the status that “protects you best” — also tops out below €5,200 a month net in these simulations. In other words, whatever status is chosen, invoicing €10,000 a month never translates into €10,000 a month net in your pocket. None of the adverts selling “€10,000 a month in AI freelancing” mentions this step — nor the previous one.

What the self-employed in France really earn

Behind the promises of five-figure incomes, the reality of self-employment remains subject to tax ceilings, charges and running costs.

To measure the gap between promise and reality, you need a point of comparison. One exists, and it is published every year by INSEE (the French national statistics institute) and URSSAF (the social contributions collection body).

Source: INSEE, Insee Première no. 2060 (June 2025, 2023 data); URSSAF (Q2 2025 data).

Two further points reinforce the table. First, only 49.8% of micro-entrepreneurs declare an actually positive turnover in a given quarter — half of them literally invoice nothing. Second, the average income of classic self-employed workers fell by -4.4% in constant euros in 2023, after -5.2% in 2022. The trend is not going in the promise’s direction.

Put together, the “€10,000 a month” promise represents around 15 times the average income of a French micro-entrepreneur, and around 2.5 times the average income of “classic” self-employed workers who already make a full living from their activity. That is not proof that it is impossible — it is proof that it is not the typical result of the status these adverts present as an accessible starting point.

A useful point of comparison, outside AI freelancing: on ride-hailing and delivery platforms, another sector long sold as a simple way to top up your income thanks to technology, the dedicated French regulator (ARPE) measures a trend opposite to the promise. The average hourly income of ride-hailing drivers fell to €15.9 an hour in 2025, its lowest recorded level (-3.2% year on year), while delivery riders’ hourly income fell by -25.9% at Uber Eats and -12.9% at Deliveroo since 2019. This is not AI, but it is the same initial marketing argument — “the platform/the technology makes you more efficient, therefore better paid” — applied to a sector where several years of hindsight are, this time, available.

What economic research says: who gains, who loses

Beyond testimonials and platform barometers, three economic studies published in 2025-2026 directly measure generative AI’s effect on self-employed incomes — with a result more nuanced, and sometimes more counter-intuitive, than the marketing narrative.

A study in the journal Organization Science (Xiang Hui and Oren Reshef, Washington University in St. Louis; Luofeng Zhou), based on real data from a large freelance platform in the six to eight months after ChatGPT’s launch, measures a drop of 2% in the number of contracts and 5.2% in income for writing freelancers. The most surprising result: it is the most experienced and best-rated freelancers who lose the most income in relative terms — AI compresses precisely the premium the market previously paid for expertise and reputation, the opposite of what the “become an AI expert to charge more” argument suggests.

A second study, published in the Journal of Economic Behavior & Organization (Ole Teutloff, Johanna Einsiedler, Otto Kässi, Fabian Braesemann, Pamela Mishkin, R. Maria del Rio-Chanona; January 2025), analyses more than 3 million job posts on a global freelance platform: demand for skills directly substitutable by AI (writing, translation) falls by 20 to 50% against the expected trend, with the steepest drop on short assignments. Conversely, demand for machine learning development rises by +24%, and demand for building AI chatbots has almost tripled. The market is not collapsing: it is polarising, between skills replaced and skills reinforced by AI.

A third study, from the Stanford Digital Economy Lab (Erik Brynjolfsson, Bharat Chandar, Ruyu Chen; November 2025), draws this time on high-frequency payroll data on the employee side, not the freelance side: it measures a relative drop of 13% in employment among young workers aged 22 to 25 in the occupations most exposed to AI, while experienced workers in the same occupations stay broadly stable. The polarisation observed there is therefore the reverse of the freelance market: on the employee side, it is juniors who take the hit; on the freelance side, it is rather seniors whose expertise premium is eroding.

That polarisation confirms, through market data rather than an isolated case, what Le Recul was already observing in a more forward-looking way about RentAHuman, the service that rents humans to AI agents: the boundary between tasks reserved for humans and tasks captured by AI keeps moving fast, in both directions, and rarely in the way the marketing of “€10,000 a month” methods implies.

Two official bodies confirm, each in its own way, that no reliable overall picture yet exists. Eurostat is preparing a dedicated module on digital platform employment, but data collection only starts in 2026 — the first results will not be published before around 2027. And in France, DARES (the labour ministry’s research directorate), whose last study on platform workers dates from November 2024, has to date published no dedicated analysis combining self-employment and generative AI.

Method 1 — Selling your AI skills as a freelancer or setting up an agency

This is the most “real” version of the promise: there is a genuine market, with genuine rates, for AI skills.

Even at that level, the real monthly net income recorded by the Malt barometer stays in a range of €2,500 to €6,000+ depending on seniority — the top of that range already exceeds, as we have seen, what the micro-entreprise status legally allows, which suggests the best-paid profiles have changed status. And you have to subtract the platform’s commission (around 10% at Malt, 0 to 15% at Upwork depending on client history, a fixed 20% at Fiverr) before even applying tax.

The “AI agency” model — selling automation to small businesses rather than billed time, often through agentic systems presented as autonomous — is more fragile. One documented, costed account, published on LinkedIn by a practitioner in the sector, describes a project billed at 500 dollars that required a full week of work, an effective hourly rate below 10 dollars, and reports that over nine months of prospecting, half the prospects announced a budget below 2,000 dollars. That individual finding matches a far more solid data point, already documented by Le Recul regarding layoffs attributed to AI: an MIT study (the NANDA initiative, “The GenAI Divide: State of AI in Business 2025”, August 2025, based on 150 executive interviews and the analysis of 300 AI deployments) shows that only around 5% of AI pilots deployed in companies generate a measurable and rapid impact on results — not a 5% rate of return, but 5% of projects that show any measurable effect at all, the remaining 95% showing none. If the overwhelming majority of small businesses buying AI see no measurable return, the market an “AI agency” sells into is structurally difficult.

This model has also produced cases of outright fraud, sanctioned by the American courts:

All four cases come under the American FTC (Federal Trade Commission), as part of its dedicated “AI Comply” operation launched in September 2024. No costed French equivalent has been identified for this precise model.

Method 2 — Coding and selling a SaaS or AI app

The promise here: code an application with an AI assistant (Cursor, Claude Code, Replit, Lovable, bolt.new) in a weekend, put it online, and monetise it. Our AI coding agent ranking in fact tracks these tools regularly, and their real capabilities are advancing fast — the technical part of the promise (coding faster) is broadly true. It is the “selling” part that is the problem.

An independent analysis of 937 Indie Hackers products with revenue verified through Stripe (2022, a figure still widely recirculated today, to be dated precisely) shows that more than 54% generate no revenue at all, and that only around 5% exceed 8,333 dollars a month (100,000 dollars a year). The emblematic cases highlighted on social media (a developer generating 132,000 dollars a month with a single 40 dollar a month server) are extreme statistical exceptions, not a norm.

The viral narrative about the startup Builder.ai — 700 Indian engineers disguised as an AI named “Natasha” — deserves correction: according to a more thorough investigation (Gergely Orosz, The Pragmatic Engineer), Builder.ai genuinely used language models (GPT/Claude), developed by a small team of 15 to 30 engineers; the “700 engineers” were subcontractors on a separate outsourced development business. The company’s collapse (bankruptcy in 2025, despite a 1.5 billion dollar valuation) was a classic accounting fraud — 2024 revenue inflated by 300% — not a non-existent AI. That is a useful signal: even the most viral stories about AI failure deserve checking before being repeated.

On the technical and economic side, three solid data points:

• Security: according to Veracode (2025 GenAI Code Security Report, testing 100+ models on 80 tasks tied to known vulnerabilities), 45% of AI-generated code fails standard security tests (OWASP Top 10), with an 86% failure rate specifically against XSS flaws.
• Real conversion: according to ChartMogul (SaaS Conversion Report, January 2026, 200 SaaS products analysed), a freemium conversion rate considered “good” sits between 3 and 5%, and “excellent” between 8 and 12% — far from the image of an app that “sells itself”.
• Compressed margins: according to venture capital analyses (Bessemer Venture Partners, ICONIQ Capital), fast-growing AI startups post gross margins well below the historical standards of pure software (around 25 to 52% against 80%+ traditionally), because every request consumes paid API calls to the likes of OpenAI or Anthropic.

Method 3 — Trading and “AI bots”

This is the method most unanimously condemned by regulators, with no exception found in this research.

France’s AMF (the financial markets authority) has been warning since 2023 against investment offers through “trading robots”, which advertise returns of 5 to 15% a month, or even up to 400% a year — promises the authority explicitly describes as a warning sign. Its most recent report (“The use of AI by financial market participants in France”, February 2026, carried out with ESMA) specifies that only 1% of the use cases recorded in the regulated financial sector concerns a direct application of AI to investment services for third parties — AI is used first internally, not for consumer trading. On the saver side, 11% of French people say they use AI in their research before an investment, and the AMF points out that “publicly accessible LLMs fall outside any financial supervision”: an individual following an AI’s advice on their own loses the protection normally owed by a regulated professional.

The scale of the problem is quantified by the ACPR and the Banque de France (joint briefings with EIOPA, the EBA and ESMA, 30 January 2026): the harm from online financial fraud is estimated at at least 500 million euros a year in France, with more than 5,000 unauthorised operators listed on blacklists since 2022, and a specific alert about deepfake videos impersonating the governor of the Banque de France to tout fake investments. The annual report of the joint AMF-ACPR unit (2025) adds that 16% of French people say they were the victim of financial fraud in 2025 (32% among the under-35s), for an average declared loss of €34,000.

In the United States, the CFTC cites the Mirror Trading International case: 1.7 billion dollars stolen, more than 23,000 victims, through a “proprietary trading bot” promising a guaranteed 10% return a month. The SEC also sanctioned, in 2025-2026, several platforms using fake “AI signals”, for cumulative harm of at least 26 million dollars across two separate cases. No independent, rigorous academic study measuring the real performance of consumer AI trading bots was found — only the publishers of these bots publish performance figures, which makes them judge and jury.

Method 4 — Monetisable content: ebooks, stock, blogs

Beyond social media (already tested), three content markets attract the same promise.

Self-published books (Amazon KDP). Amazon has required, since 2023, the declaration of any AI-generated content, and has limited uploads to 3 new titles a day since September 2023, a measure taken explicitly “to protect against abuse” in the face of an influx of low-quality AI content. A recent academic study (NBER Working Paper no. 34777, Reimers & Waldfogel, May 2026, covering a representative sample of more than 10 million ebooks) shows that the share of content detected as AI-generated rose from near zero in 2022 to more than 50% of releases in 2025, while average perceived quality (measured by reader ratings) declines. On income, a survey by the Alliance of Independent Authors (2,200 respondents, 2023) gives a median of 12,749 dollars a year, far below the average of 82,600 dollars a year — the sign of a very long-tailed distribution where a minority pulls the average up. A more recent survey (Written Word Media, 1,346 respondents, 2025) is harsher still: 44% of authors earn 100 dollars or less a month.

Stock content (images, video, audio). Shutterstock and Getty Images entirely prohibit AI-generated content submitted by external contributors — while themselves selling their own AI generation tools to their customers, a contradiction those companies accept (it is impossible to guarantee the copyright of the works used to train a third-party model). Adobe Stock, by contrast, accepts labelled AI content, with a royalty rate of 33 to 35%, identical to classic content. On voice-over, ElevenLabs has paid out 22 million dollars to more than 10,400 creators through its Voice Library (May 2026) — a theoretical average of around 2,100 dollars per creator over the life of the programme, with no median or detailed distribution published by the company.

Blogs and affiliate sites. Google introduced in March 2024 its “scaled content abuse” policy, aimed at any content produced at scale with the main purpose of manipulating rankings — whether that content is generated by AI, by humans, or both. Google claims a 45% reduction in content judged unoriginal in its results. An Ahrefs study (July 2025, 600,000 pages analysed) finds a near-zero correlation (0.011) between a site’s share of AI content and its ranking position — but notes that 100% AI content with no human supervision rarely reaches the top position. No reliable costed study on the real success rate of mass AI-content affiliate sites was found: that is a genuine data gap, and it should be flagged as such rather than ignored.

The common thread across the three markets: it is the platforms themselves that capture the value — Amazon with its own free AI narration tool rather than paying third-party voices, Shutterstock banning its contributors while selling its in-house generator, Google demoting mass content while capturing the traffic through its own AI summaries at the top of results.

Method 5 — Selling courses, prompts and derivative products: the 1% law

One last family of methods deserves separate examination, because it inverts the logic of the previous ones: here, you are no longer selling a service or content, but a digital product (course, prompt, template, design) supposedly selling itself “on autopilot”. Wherever a platform publishes real data, the same pattern appears.

On Gumroad, an independent study (InsightRaider, 21 May 2026, based on crawling more than 146,000 public product pages — a third-party estimate, not official platform figures) finds a median of 72 dollars a month, with 99.5% of the platform’s total revenue captured by the top 1% of sellers, and 44% of products generating no sales at all. On Etsy, a sector trade association (Craft Industry Alliance) and a 404 Media investigation have documented since 2024 a wave of AI-generated designs (up to 3 of the first 8 results on some searches), which pushed the platform to tighten its anti-AI policy in several stages up to June 2025 — the opposite of a market easy to flood with AI content and live off. The most frequently cited seller income ranges (unofficial, but convergent) give an average of 2,965 dollars a month against a median of only 574 dollars a month, with nearly two thirds of sellers below 100 dollars a year. On Merch by Amazon, the approval rate for new applications tightened to 30-40% in 2026, in direct reaction to an influx of accounts generating low-quality AI content.

Prompt marketplaces, often cited as an “easy” way to monetise your AI skills, turn out to be partly fictional. PromptBase, the oldest (80% to the seller, 20% to the platform), has published no audience or revenue statistics since December 2022; one of the rare costed and verifiable accounts available (a seller who documented their sales month by month) reports 6 dollars in the first month, then 28 dollars in the second, before a single negative review put an end to all further sales. AIPRM and FlowGPT, often presented as equivalents, in fact have no functioning paid resale marketplace for an individual creator — one works by subscription to a curated library, the other only offers a social media influence programme, capped at 100 dollars per video. Another example to correct explicitly: several sites have been announcing since May 2026 the launch of a “Skills Marketplace” at Anthropic with a 15% commission — on checking, that marketplace does not exist for individual creators: Anthropic’s official page describes only B2B partnerships with large companies, and the public “Skills” repository is a free registry, with no payment mechanism at all.

Two last routes sometimes sold as easy — reselling white-label AI subscriptions to small businesses (GoHighLevel, ManyChat) and positioning yourself as an “AI-boosted” virtual assistant — remain, to date, insufficiently documented to judge: the only profitability figures available come from the platforms that themselves sell the resold subscription, and no independent study measures AI’s real effect on freelance virtual assistants’ rates. That lack of independent data is, in itself, a signal to weigh before investing in either of those two methods.

What Le Recul had already tested

Three strands of this promise have already been the subject of a dedicated test, with full methodology and figures. We do not repeat them here, only the verdict:

• AI courses at €1,997 — verdict partly kept: a course can be worth its price if it brings real support, but becomes indefensible if it amounts to prompts and tutorials already available free.
• AI-automated sales prospecting — verdict partly true but heavily oversold: out of 1,000 prospects contacted, public benchmarks suggest a few dozen replies and only a handful of genuinely qualified leads.
• AI-automated social media accounts — verdict partly kept: automating production genuinely works, but the test carried out by Le Recul generated €0 of revenue and 0 followers, for want of a real audience built.

The promise itself: who sells it, and how

Beyond the individual methods, the “€10,000 a month with AI” promise circulates as a marketing product in its own right, carried by accounts followed by tens of thousands of people in France, and documented as a fact-checking subject by franceinfo. The most solidly documented mechanism is income screenshots (“MRR”, Monthly Recurring Revenue) used as social proof before selling a course or coaching. The phenomenon is widespread enough that a tool for making fake Stripe screenshots circulates openly online, and that in October 2025 a recognised developer in the “indie hacker” community, Marc Louvion, launched TrustMRR.com, a revenue verification service using a read-only Stripe connection, in direct reaction to a viral report by Pieter Levels (creator of RemoteOK) on the proliferation of fake screenshots. That an “anti-cheat” tool had to be created is, in itself, proof of the scale of the problem.

On TikTok, an investigation by the outlet Indicator (Craig Silverman, 12 January 2026) documented more than 40 accounts, totalling 71 million cumulative views, claiming the platform pays for simply watching videos — a mechanism redirecting to dubious software and fraudulent affiliate commissions. That is not an AI promise strictly speaking, but it illustrates the same soil: high-audience accounts monetising the promise of easy gains more than the gain itself.

The mechanism is found among identifiable French-speaking trainers. An investigation by the Swiss daily Le Temps (26 January 2026) into Yomi Denzel — active online since 2017, with around 50,000 sign-ups to his free launch webinar — documents his pivot to AI through a programme named “Mission AAA” (AI agent agency), with paid coaching billed at between €3,000 and €7,000; the investigation does not conclude there was outright fraud but underlines the impossibility of obtaining real success rates from the seller. On his earlier e-commerce courses (Ecom Pro/Ecom Blueprint, before his AI pivot), Yomi Denzel was already using word for word the guarantee “3 months to generate 10,000 euros of turnover, or your money back” (subject to conditions) — a near-identical promise found with his associate Théo Ritzy (the “Ecom Accelerator” programme, €1,497), and in an investigation by L’ADN (9 October 2025) into other sellers of online business courses. A second franceinfo investigation, this time into courses at €75-150 resold by teenagers on TikTok and Skool, quotes the DGCCRF itself acknowledging it had received, to date, only “few reports” about this precise type of practice — consistent with the absence of targeted action noted below.

The French legal framework

Unlike the United States, France has no specific, standalone text governing commercial earnings claims — no equivalent to the American Business Opportunity Rule, which requires sellers of “business opportunities” to provide a mandatory costed evidence sheet (Earnings Claim Statement) specifying the real percentage of buyers who reached the advertised gain. The French regime relies on the general law on misleading commercial practices.

The Consumer Code (art. L121-2) classes as misleading any false claim, or claim liable to mislead, about a professional’s advantages or the scope of their commitments — a costed, unverified promise such as “€10,000 a month” falls directly within that scope. Key point: article L121-5 places the burden of proof on the seller — absent supporting evidence, the claim is deemed inaccurate. Penalties go up to 2 years’ imprisonment and a €300,000 fine for an individual, or 10% of average annual turnover. The law of 9 June 2023, known as the “influencer law”, adds specific obligations for promoting professional training (information on funding and the training body), with an implementing decree of 30 March 2026.

On the ground, the DGCCRF (France’s consumer protection and fraud control authority) imposed more than €200 million in fines and settlements in 2025 (against €81 million in 2023), against a backdrop of 460,000 consumer reports (+60% in two years). Its latest influencer review (287 accounts checked in 2024) finds irregularities in nearly half of them, with 40 warnings, 65 injunctions and 8 criminal reports. But despite an extensive search, no DGCCRF action specifically targeting “AI courses” or “AI income promises” has been identified to date — unlike the American FTC and its dedicated “AI Comply” operation, launched as early as September 2024. That is a French regulatory blind spot worth flagging as such, rather than ignoring or inventing.

The underlying mechanism — course, coaching, costed earnings promise — has, on the other hand, already led to criminal convictions in France, outside the AI field alone. The criminal court of Rennes convicted, on 23 and 24 March 2026, a “personal development coach” of abuse of weakness against several dozen complainants between 2016 and 2024: 3 years’ imprisonment including 1 year without suspension, an obligation to compensate the harm, and a ban on any training activity; her company received a €100,000 suspended fine and a ban on carrying out training activity for 5 years. The judicial risk does not run one way only, however: the commercial court of Aix-en-Provence, conversely, condemned on 9 December 2025 a French site reviewing trading courses (“Warning Trading”) for disparagement of the body it was criticising, to around €38,000 in total — a reminder that documenting or publicly denouncing a dubious course also exposes you to legal risk under French law if the statements go beyond factual criticism.

Not all the routes presented as shortcuts to “€10,000 a month” offer the same chances of success. The data show a reality far more contrasted than the marketing promises.

What is viable, what is not

Our Le Recul verdict

The promise is misleading in its most commonly sold form — accessible to everyone, fast, almost passive. None of the methods examined here, nor the three already tested by Le Recul, produces that result reliably and repeatably for a beginner. That is not the same as saying it is impossible: a small number of already qualified professionals do exceed the amount, mainly through high-rate AI freelancing. But even in that favourable case, the status most often highlighted in these adverts caps net income at a little over half the promise, and the path that remains — expertise already built, a change of legal status, several years of experience — has little to do with what is being sold.

On the other methods, the dominant statistic is not the spectacular success highlighted in the adverts, but its rarity: a minority of cases above the threshold, against a majority of incomes close to zero — and, in several cases documented by the American courts, tens of millions of dollars of harm to victims who had paid for the promise itself, not for a result obtained.

Conclusion — what to take away before paying for the method

Three simple checks let you test any version of this promise before committing money or time to it: ask for the real number of people who reached the advertised amount (not isolated examples); check whether the calculation offered accounts for the legal ceiling of the recommended status; and look into whether the seller has already been the subject of a DGCCRF report or similar proceedings. AI genuinely changes what one self-employed person can produce alone — faster, with fewer resources. It does not change tax thresholds, the maturity of a buyer market, or the statistical probability of being among the few per cent who succeed rather than the majority who never reach the promised figure.