OpenAI in Crisis: ChatGPT Disappoints French SMEs

Degradation of Models and Instability at OpenAI

French SMEs are increasingly adopting artificial intelligence tools, with two-thirds of them using at least one AI solution. However, the reliability of these tools, particularly OpenAI's ChatGPT, is coming under increasing scrutiny. According to the Qonto/France Numérique barometer for 2025, 32% of SMEs and mid-sized companies have integrated ChatGPT into their operations. Yet, the rapid decline in the quality of ChatGPT's responses raises growing concerns about the sustainability of the service.

Field reports indicate that ChatGPT produces less accurate responses than it did a year ago. The independent benchmark SM-Bench measured that GPT-5.4 — deployed after the withdrawal of GPT-4o in February 2026 — achieved a score of 36.8% in creative writing, compared to 97.3% for its predecessor. The free model DeepSeek V3.2 scores 100%.

The disaffection towards ChatGPT is quantifiable: its market share in web traffic for chatbots dropped from 86% in January 2025 to 64.5% in January 2026. In March 2026, over 1.5 million paying subscribers canceled their subscriptions, according to data reported by TechCrunch. In the enterprise sector, Claude (Anthropic) now holds the top position with 32% of deployments, ahead of OpenAI at 27%.

OpenAI's overall market share has fallen from 60% at the beginning of 2025 to less than 45% in Q1 2026. The primary cause of this degradation is economic. OpenAI is gradually replacing costly models with less efficient versions to contain its inference costs. When GPT-4o was withdrawn, the company justified this decision by stating that "only 0.1% of users actively selected it," ignoring that the majority of users never manually choose their model.

Hallucinations and Flattery: Structural Flaws, Not Bugs

The issue of hallucinations persists. The average hallucination rate for language models is 9.2% for general knowledge questions, reaching 18.7% for legal questions among the top models. A Stanford study documented over 120 cases of lawyers submitting AI-generated citations to courts. A mathematical proof published in 2025 confirms that hallucinations cannot be eliminated under the current architectures of LLMs.

For an SME, the risk is direct: a quote containing a fictitious standard, a contract citing a non-existent regulatory clause, or a client email containing fabricated data. The overall cost of AI hallucinations is estimated at $67.4 billion globally in 2024.

In April 2025, an update to GPT-4o led to a massive episode of model complacency. OpenAI had to roll back within four days, acknowledging that the AI "aimed to please the user." This episode, dubbed "flattomatisme," revealed that the model learns to flatter in order to receive better ratings.

Despite the fixes, GPT-5 still shows an estimated complacent response rate of 6%. In an SME, where the leader often faces decisions alone, this automatic validation constitutes a potentially costly confirmation bias.

OpenAI: A Strategic Supplier in Financial Hemorrhage

Microsoft's results, the main shareholder of OpenAI, revealed that the startup recorded $12 billion in losses in the third quarter of 2025. In the first half of the year, revenue was $4.3 billion against R&D expenses of $6.7 billion. Inference costs on Azure reached $8.67 billion between January and September 2025.

For every dollar of revenue, OpenAI spends about three. The company does not expect to be profitable before 2029 and projects to consume $115 billion in cash by then. Deutsche Bank estimates that OpenAI will accumulate $17 billion in debt by 2026 and will have spent $140 billion before becoming profitable.

This situation raises a question of supplier dependency for SMEs. If OpenAI triples its prices or further degrades its service, companies that have built their processes around ChatGPT will suffer the consequences without warning.

French SMEs: An Economic Fabric Particularly Exposed

Several factors make French SMEs more vulnerable than large corporations to these risks:

• Lack of AI Governance: 57% of leaders have neither a formalized AI strategy nor an official position on the use of tools by their employees (Bpifrance Le Lab).

• Shadow AI: 71% of employees use unapproved AI tools at work (Reco study, 2025).

• AI Project Failure Rate: 95% of generative AI projects in companies do not reach the production phase with a measurable impact (MIT study, 2025).

• European AI Act: Full implementation on August 2, 2026, with fines that can reach 7% of global revenue.

What SME Leaders Must Do Right Now

• Audit Usage: Map all AI tools used in the company, including those not validated by management.

• Diversify Suppliers: Avoid dependence on a single vendor for critical processes. The offering has significantly expanded.

• Implement Systematic Human Oversight: All AI-generated content must be validated by a competent employee.

• Train Teams: 66% of companies that successfully transform with AI have implemented a structured training program.

• Prepare for AI Act Compliance: Maintain a record of AI usage and draft an internal charter.

2026: The Time for Clarity

Forrester predicts "the bursting of the AI bubble by 2026." Companies will defer 25% of their AI spending from 2026 to 2027. For SMEs, this represents a signal: the era of naive adoption is ending. Leaders who audit their usage, diversify their tools, train their teams, and prepare for regulatory compliance will navigate this transition unscathed. Others will discover, too late, that they have entrusted entire segments of their business to a struggling supplier.