French small businesses lag on AI search readiness

A new barometer from French web agency Les Créavores finds that 80.8% of 474 small-business websites are not ready for AI-powered search and answer engines. The study says the issue is not access by AI bots, but that most sites remain hard for systems like ChatGPT, Copilot and Perplexity to read and use.

Why it matters: - AI assistants and answer engines are becoming a new discovery layer for local businesses. - If small-business sites are not machine-readable, they may miss future traffic from tools that increasingly replace or complement Google search. - The study suggests the shift is already happening: on Les Créavores’ own site, assistants such as Copilot, Perplexity and ChatGPT sent 2.5 times more visits than classic Google search over the last seven measured days.

What happened: - Les Créavores published what it calls the first French barometer of TPE-PME readiness for AI search and answer engines. - The barometer analyzed 474 websites on July 13, 2026, after collecting 509 sites in total. - The sample came from the top 10 organic local Google results across 10 sectors and 15 French cities. - The study looked at homepage crawlability, site structure and trust signals.

The details: - The average readiness score was 48 out of 100, and the median was 49. - 80.8% of sites scored below 60. - Only 1.3% of sites scored above 80. - Just 1.3% of sites blocked AI robots in robots.txt, while access was open on 85.9% of sites. - 19.6% of sites had an llms.txt file, an emerging standard meant to improve AI readability. - Only 13.3% had a compliant llms.txt file. - Agentic navigation, or an AI agent’s ability to move through and act on a site, averaged 1.2 criteria out of 3. - Only 5.3% of sites met all three agentic navigation criteria. - 17.9% of sites displayed an identifiable author, a trust signal often used by answer engines. - The barometer measures readiness, not actual visibility in AI responses. - Les Créavores says the full methodology and limitations are published with the report. - The dataset is open under CC-BY 4.0 and deposited in Zenodo, Figshare, Harvard Dataverse and data.gouv.fr. - The listed permanent identifiers are DOI Zenodo 10.5281/zenodo.21381109, Figshare 10.6084/m9.figshare.32996552, and Harvard Dataverse 10.7910/DVN/IIB2AW. - The study says every company can test its own site for free with the scanner used for the analysis.

Between the lines: - The main barrier is not AI access control. The study says most sites are open to AI bots but still unreadable or poorly structured for them. - That makes technical structure, trust markers and content organization more important as AI agents influence how users find local services. - The open dataset signals an effort to make the benchmark reproducible and portable for other researchers and agencies.

What's next: - Les Créavores is positioning the scanner as a self-check tool for businesses that want to improve their AI search readiness. - The agency says it will continue publishing data and methodology openly. - Small businesses that want to stay discoverable may need to audit crawlability, add machine-readable signals and improve trust cues before AI search becomes a default behavior.

The bottom line: - French TPE-PME websites are mostly accessible to AI, but not yet legible enough for AI systems to confidently use them.