See Selected Projects

Government & public servicesWork & productivityFR

Text job seekers about a vacancy and pass their answers to an adviser

France Travail's MatchFT lets employment advisers in France send an AI agent to text selected job seekers about a vacancy, check their interest, availability and requirements, and report back before the adviser puts them forward.

Try it withClaudeChatGPT / OpenAI modelsGeminiGrok

Deployed; tested in 87 agencies in two regions by the end of 2024, with national rollout announced from winter 2025-26, and described by France Travail in June 2026 as in real, deployed use, France Travail (French public employment service), using Mistral AI models

Why it matters

France Travail, the French public employment service, has to check that the people it suggests for a vacancy still want the job and can take it. Before MatchFT, an adviser had to chase each job seeker by phone. The agent can text up to 50 people on a list of matches at once and hold the conversations in parallel, while the adviser reads the answers and decides who goes forward. In a test in 87 agencies in the Pays de la Loire and Centre-Val de Loire regions, France Travail reported a median of 19 minutes for job seekers to reply to the first text, half of introductions to employers made within 24 hours, 90% of candidates finishing the questions within five hours, and 30% of vacancies leading to an introduction. It has also been examined in public by France's national audit court, the Cour des comptes, in a January 2026 report on France Travail's use of AI.

How it works

For each vacancy, the matching system France Travail has used for about a decade (the SDR) ranks job seekers' profiles. The adviser chooses to contact the whole list, up to 50 people, or a selection, and the criteria to check. A generative AI agent, developed in house and by June 2025 already partly using Mistral AI models, with full use planned for 2025, then texts each person and asks questions based on the vacancy: interest in the job, availability, access to the workplace, agreement on hours and pay, and ability to do the tasks. The adviser can read the exchanges and uses the results to decide whether to put the person in touch with the employer, and the answers can also update the person's file, such as their availability. France Travail first tried pairing the agent with a new statistical matching model, but in early evaluations it did less well than the existing SDR, so the SDR stayed. The texts do not replace the other ways advisers contact job seekers. By June 2026, France Travail described a version that job seekers reach from their personal space on francetravail.fr, which proposes vacancies through a guided conversation, and said that MatchFT does not apply for jobs on a person's behalf: the decision to apply stays with them.

Who it's for

Public employment services, job centres, council employment teams, recruitment agencies and social enterprises that match job seekers to vacancies and spend time chasing people by phone.

The prompt

You are helping me design a text-message conversation that checks whether selected job seekers want a vacancy and can take it, before a person on our team decides who to put forward. Do not ask me for the names, phone numbers or records of any real job seeker.

Our organisation: [for example, a public job centre, a council employment support team or a recruitment social enterprise]
The vacancy: [job title, place, hours, pay, start date, and any must-have requirements such as a driving licence or certificate]
Language and tone: [for example, French or English, polite and plain]

1. Write an opening text that says who we are, that the messages come from an automated assistant, which vacancy this is about and how to stop the messages.
2. Write up to six questions, one per text, covering interest, availability and start date, getting to the workplace, hours and pay, and the must-have requirements. Keep each under 160 characters, with clear answer options where possible.
3. For each question, say which answers mean 'put forward', 'ask a person to call' or 'not a match', and how to treat unclear or off-topic replies.
4. List the replies that must go straight to a person (for example distress, a disability or health need, a complaint, or a question about benefits), with the exact handover text.
5. Write a template for the summary an adviser reads, with the answers word for word and a one-line match result at the top.
6. Suggest fairness rules: what we must never ask, how to avoid screening people out for reasons unrelated to the job, and how someone can ask to deal with a person instead.

Do not invent laws, benefit rules or phone numbers; leave placeholders. Mark every rule that an employment adviser or data protection officer must sign off before use.

Open inClaude (opens in a new tab)ChatGPT (opens in a new tab)Gemini (opens in a new tab)Grok (opens in a new tab)

Sources

The press release, news pages and interview are France Travail's own. The Cour des comptes, France's national audit court, examined MatchFT independently in its January 2026 report, though its deployment details come from France Travail. Le Monde Informatique relayed France Travail's pilot figures, and Clicanoo added a union view. No independent evaluation of job outcomes was found.

  1. 1. France Travail (press release): "contacte les candidats présélectionnés sur une offre par SMS"; "accès au lieu de travail, disponibilité"; "MatchFT et Neo utilise déjà Mistral". France Travail présente à VivaTech 2025 ses nouveautés en matière d’IA (opens in a new tab)

  2. 2. France Travail (news): "MatchFT interroge par SMS le demandeur d’emploi sur ses attendus et prérequis."; "Le conseiller peut consulter leurs échanges et intervenir ensuite"; "devait relancer l’usager par téléphone". France Travail à VivaTech : accélérons sur l’IA ! (opens in a new tab)

  3. 3. France Travail (job seeker news): "Accéder à CoachFT et MatchFT depuis votre espace personnel"; "MatchFT ne candidate pas à votre place"; "La décision de candidater reste la vôtre". Candidats : ce que l'IA change vraiment pour vous (opens in a new tab)

  4. 4. France Travail (interview with its head of AI and data): "En 2025, nous avons intégré l'IA dans les actes métiers"; "On est dans des usages réels, déployés, utiles". « L’IA n’est pas qu’un sujet technologique : c’est un enjeu de service public » (opens in a new tab)

  5. 5. Cour des comptes (French national audit court): "une brique d’IA provoque un échange de SMS pour préqualifier un profil"; "jusqu’à 50 personnes"; "capacité d'engager plusieurs discussions simultanément"; "87 agences ont fait l’objet d’un déploiement"; "s'est révélé moins efficace que le SDR"; "n’ont pas fait l’objet d’une évaluation éthique formalisée"; "de MatchFT étaient en cours de préparation". France Travail et l'intelligence artificielle (rapport d'observations définitives) (opens in a new tab)

  6. 6. Le Monde Informatique: "En test dans 87 agences France Travail"; "un temps de réponse médian de 19 minutes au 1er SMS"; "90% des candidats ont terminé leur pré-qualification en 5 heures"; "30% des offres ont donné lieu à une mise en relation". France Travail active la GenAI dans ses agences (opens in a new tab)

  7. 7. Clicanoo (Le Journal de l'île de La Réunion): "le paramétrage de l’outil prend parfois plus de temps"; "regrettent de ne plus être autant au contact des demandeurs d’emploi". Emploi : France Travail généralise une IA pour présélectionner les candidats (opens in a new tab)

Related projects