A small team following misplaced credit
Atelier Source Clair studies how AI engines attribute French business information — where a claim can travel from a first-party page to a directory, press note or bilingual mirror before the model names its source.
i. Where the work began
The founding source-mismatch case was small enough to miss. A French manufacturer's own technical note had the full explanation. A regional directory carried a clipped version. The AI answer used the richer detail, yet the citation landed on the directory. Nothing in the sentence looked dramatic. The mismatch sat in the credit line, like a label stuck to the wrong drawer.
The group formed around that gap. Maëlle Riston traced where the claim had first appeared. Adrien Solac compared whether the same behavior returned in other engines. Lise Varnier looked at the French and English versions of the source trail. Tomas Elian named the error pattern instead of smoothing it away. Together they built Atelier Source Clair as a methodological lab for attribution, not a ranking desk or a page-repair shop.
The lab works at the seam between public evidence and visible citation. Its material follows prompts, cited pages and source paths closely enough to show when credit held, slipped or disappeared.
ii. Why this layer
The lab's position is narrow on purpose. AI visibility is often discussed as presence: whether a business is mentioned, recommended or summarized. Atelier Source Clair watches the next layer down. When an answer names a source, that source gains authority in the reader's eyes.
If the credited page only copied, shortened or contradicted the original carrier of the claim, the public memory of the business has shifted a little. The lab's work is to make that shift visible.
iii. What we do not do
The lab does not rank companies, declare one engine the winner or rewrite business pages as a service. Its materials are observations and methodological notes, meant to help readers understand citation behavior, not to sell a guaranteed visibility fix. Focus: citation and attribution in French AI answers. Method: repeated query sets, model comparison and source-path review.
Masthead · four researchers
Studies how AI answers move credit from first-party French sources to directories, media pages or institutional summaries. She previously edited business documentation, compared public company profiles and reviewed source trails for research summaries. Her work keeps the original carrier of a claim in view.
Asks whether citation patterns are shared across AI engines or remain specific to one answer system. He previously built repeatable desk-research workflows, maintained query logs and reviewed conflicting public descriptions of small companies. He looks for patterns that survive more than one run.
Examines how engines choose between French and English versions of the same business page or topic explanation. She previously worked on bilingual editorial audits, terminology alignment and public-facing service descriptions. Her review catches the quiet drift when two languages carry different versions of a fact.
Works on the qualitative patterns behind misattribution, uncited absorption, copied facts and source displacement. He previously reviewed factual consistency across business pages, local listings, press fragments and sector directories. He turns messy cases into clear error categories without pretending every path is provable.
The lab studies where credit moves in the answer; new cases appear in the index as the work advances.