AI Usage Trends in France – 2025 Yiaho Data

As you know, our French platform Yiaho has been offering over 70 specialized AI agents (coaches, lawyers, fun AIs, etc.) since 2023, with no account creation required.

Drawing on our teams’ expertise, we wanted to document AI adoption and practices in France.

According to our statistics, about 99% of our users are French speakers, with approximately 93% residing in France.

This article presents raw data on the use of AI agents and the tone of interactions, collected from January to June 2025 from 5 million anonymous active users on our platform.

This data highlights AI usage trends, positioning Yiaho as a key player in France for providing valuable insights into this innovative tool.

Data Collection Methodology

Yiaho implemented two complementary approaches to analyze AI usage:

• “N1” Analysis: An analysis of the usage of its 70 specialized AI agents, based on page view volume and time spent on them, excluding general AIs like GPT models.

• “N2” Analysis: A semantic analysis of interactions with all AIs, extracting semantic characteristics to categorize exchanges while avoiding access to full conversation content, in order to limit noise and respect our data privacy policy.

“N1” Analysis of specialized AI agent usage (5 million users)

• Period: January to June 2025.

• Method:
  • Each specialized AI agent has a dedicated page on Yiaho. Data was collected by measuring the number of visits to each page and the average time spent by users, based on the interactions of 5 million active users.
  • Metrics were gathered via internal tools and Google Analytics, which excludes personal data. Pages were categorized according to agent functionality (e.g., rewriting, coaching, fun games), with similar categories grouped together (e.g., aggregation of coaching agents).
  • No personally identifiable information was collected since Yiaho is freely accessible without account creation, and IP addresses were anonymized.

• Limitations: Visit and time-spent metrics reflect engagement but do not capture the exact nature of interactions.

Automated Semantic “N2” Analysis (1 million interactions)

• Period: March to June 2025.

• Method:
  • A random sample of 1 million interactions was extracted from the request streams sent to all AIs available on Yiaho.
  • An automated lexical analysis was performed using a natural language processing (NLP) model based on the spaCy framework.
  • Keywords and phrases were identified using regular expressions to detect specific patterns and a semantic clustering algorithm to group interactions by similarity in tone and content.
  • Several categories were defined. Here are a few examples (we haven’t included all defined categories).

• • • Short and useless exchanges: Words or phrases like “thanks,” “ok,” “great,” “cool,” “awesome,” “agreed,” “ciao.” These terms, often used to conclude or react briefly, are detected as conversational “noise” without a functional goal.
    • Insults and verbal violence: Terms like “shut up,” “stfu,” “you’re talking nonsense,” “you’re lying,” “stop making things up.” A predefined list of offensive words and expressions of annoyance, enriched with contextual variants, allows for the identification of these aggressive exchanges.
    • Writing requests: Phrases like “can you write,” “write a,” “draft a,” “rewrite this text,” “write me a,” “make a draft.” These phrases are identified by action verbs (e.g., “write,” “draft”) followed by an object (e.g., “a text,” “a letter”).
    • Personal exchanges: Words like “my boyfriend,” “my girlfriend,” “my husband,” “my wife,” “my partner” (but also terms like “my crush,” “I love them,” etc.). These terms, identified via relationship patterns, indicate discussions about private life.
    • Friendly exchanges: Expressions like “how are you,” “how’s it going,” “what’s up,” “hi,” “lol,” “have a nice day,” “you’re kind,” “goodnight,” “you’re right,” “see you soon,” “later.” These phrases, marked by social politeness or familiarity, reflect a warm tone.

• • No raw content was stored. Analyses were performed via an ephemeral processing pipeline, producing aggregated percentages.
  • No personally identifiable information (PII) or metadata (e.g., timestamp, location) was collected.

• Technologies used: NLP Framework (spaCy), aggregation tools (Tableau).

• Limitations: Lexical analysis may lack precision for ambiguous or multilingual requests.

Ethical Compliance

• Anonymization: No PII was collected (GDPR, Article 5).
• Security: Temporary data encrypted, deleted after aggregation.
• Transparency: Users were informed via Yiaho’s privacy policy.

AI Usage in France: Yiaho’s Raw Results

“N1” Analysis: Specialized AI agent usage (5 million users, January-June 2025)

Usage breakdown:

• • 55% in a professional context:
    • 25% rewriting and text composition (press releases, posts, professional requests, etc.).
    • 10% replying to an email or message.
    • 20% academic support (writing theses, math help, etc.).

• • 45% for fun, personal, or family use:
    • 20% coaching (parenting, nutrition, love life).
    • 15% creative content creation (stories, futuristic scenarios, dialogues with celebrities).
    • 10% fun games (name generators, quizzes, simulations).

Main uses of specialized AI agents (excluding general AIs like GPT-5):

• • 25% rewriting and text composition.
  • 10% replying to an email or message.
  • 20% academic support (writing, math, etc.).
  • 20% coaching (parenting, nutrition, love life).
  • 15% creative content creation (stories, futuristic scenarios, dialogues with celebrities).
  • 10% fun games (name generators, quizzes, simulations).

“N2” Analysis: Semantic analysis of one million interactions (March-June 2025)

Breakdown of interaction types:

• 48% fun, personal, or trivial (“personal”):
  • 50% concern human relationships, involving a partner or close relation (≈24% of the total).
  • 42% relate to well-being or coaching-style advice (≈20.16% of the total).
  • 8% include other trivial interactions, such as fun games or creative exchanges (≈3.84% of the total).
• 52% educational, professional, or technical (“pro”):
  • 85% concern rewriting, rephrasing, or text composition (≈44.2% of the total).
  • 15% include other professional tasks, such as academic support or email writing (≈7.8% of the total).

Characteristics / tones of exchanges:

• • 68% neutral, professional, personal, helpful, or factual.
  • 11% useless, with thanks, compliments, or “noise.”
  • 12% friendly, with expressions of familiarity or social politeness.
  • 9% tense, with signs of annoyance, insults, or verbal violence.

What should we take away from this data?

Data from our Yiaho platform shows AI adoption in France split between professional efficiency and personal exploration.

With 55% professional usage (writing, academic support), AI is establishing itself as a practical tool for meeting work and education demands. Meanwhile, the 45% of fun or personal interactions, particularly around coaching and relationships, reveal a need for emotional support and creativity.

The mostly neutral tone (68%), occasionally friendly (12%), and marginally aggressive (9%), reflects a varied relationship with AI, ranging from utility to familiarity. This duality suggests that AI is becoming a natural extension of daily activities, but its growing integration could sharpen the differences between those who use it to optimize productivity and those who seek a space for personal expression.

In short, key figures and trends to remember about artificial intelligence usage:

In 2025, AI is becoming almost human for the French:

• 1 in 5 users talks to AI like a human, including 1 in 8 with the familiarity of a close friend and 1 in 11 with insults or annoyance.

AI usage, between serious and lighthearted:

• 55% use AI in a professional setting,
• 20% for various personal coaching (meaning 1 in 4 people for life or relationship advice).
• Finally, 11% of exchanges have no utility, limited to thanks or cordiality.