# Insurance: which agencies do ChatGPT and Gemini recommend, town by town? > When someone asks ChatGPT or Gemini for “an insurer in Laon” or “an agency to insure my shop in Nîmes”, does the AI name a nearby agency, a national brand or a comparison site? We measured where each insurance network's agencies stand in nearly 7,000 answers to local questions, town by town, against mutual insurers, banks, brokers and comparison sites. Canonical page: https://directory.wispra.com/ai-barometer/insurance Edition: 1 · October 2026 · Answers collected from July 1, 2026 to October 2, 2026 All barometers: https://directory.wispra.com/ai-barometers ## Key figures - 56% of answers to a local question recommend a specific agency first: its name, address, phone number or agency page. - 57% of local answers put agencies from at least two networks side by side. - 20–81% depending on the network: how often the AI names one of its agencies when it cites the brand. - 3 in 10 only: the same question, asked again a week later, gives the same first recommendation. ## Local search: AI recommends an insurance agency first In 56% of cases, the first recommendation is a specific agency of a network (95% CI 50–61). Next come a brand cited without any agency (28%) and a broker, comparison site or online insurer (11%). The gap between large cities and small towns is clear. What AI recommends first, by town size: - All local questions: A specific network agency 55.6% · A brand, no agency named 27.7% · A broker, comparison site or online insurer 11.0% · No brand 5.7% - Large cities (100,000+): A specific network agency 45.8% · A brand, no agency named 30.0% · A broker, comparison site or online insurer 13.5% · No brand 10.7% - Mid-sized towns (20,000–100,000): A specific network agency 59.0% · A brand, no agency named 27.3% · A broker, comparison site or online insurer 11.0% · No brand 2.8% - Small towns (under 20,000): A specific network agency 61.0% · A brand, no agency named 25.5% · A broker, comparison site or online insurer 9.1% · No brand 4.4% What AI recommends first, by need: - Car and motorbike: A specific network agency 48.1% · A brand, no agency named 29.5% · A broker, comparison site or online insurer 21.6% · No brand 0.8% - Home: A specific network agency 63.5% · A brand, no agency named 27.4% · A broker, comparison site or online insurer 7.4% · No brand 1.6% - Health: A specific network agency 54.6% · A brand, no agency named 26.4% · A broker, comparison site or online insurer 12.1% · No brand 7.0% - Life and disability cover: A specific network agency 61.6% · A brand, no agency named 28.7% · A broker, comparison site or online insurer 2.6% · No brand 7.0% - Savings and retirement: A specific network agency 51.1% · A brand, no agency named 19.8% · A broker, comparison site or online insurer 7.9% · No brand 21.2% - Mortgage insurance: A specific network agency 30.5% · A brand, no agency named 17.0% · A broker, comparison site or online insurer 43.9% · No brand 8.6% - Business insurance: A specific network agency 59.8% · A brand, no agency named 33.0% · A broker, comparison site or online insurer 3.0% · No brand 4.2% - School insurance: A specific network agency 60.4% · A brand, no agency named 36.8% · A broker, comparison site or online insurer 1.1% · No brand 1.8% ## When AI cites an insurance network, does it name its agencies? On local questions, at least one agency is named in 69% of answers. But networks do not turn their visibility into agency recommendations equally: when the AI cites the brand, it names one of its agencies in 81% of cases for the best-placed network, and in 20% for the least well placed. Network: cited · agency named · conversion · agency cited first (local questions) - Allianz: 64.2% · 47.7% · 74% · 20.9% - AXA: 52.0% · 39.5% · 76% · 13.3% - Groupama: 40.9% · 28.0% · 69% · 4.2% - MMA: 30.1% · 18.8% · 63% · 3.1% - Generali: 24.4% · 15.7% · 64% · 1.8% - MAIF: 25.9% · 12.1% · 47% · 2.6% - MAAF: 20.3% · 10.8% · 53% · 0.7% - Gan: 11.8% · 9.5% · 81% · 1.8% - MACIF: 16.0% · 8.6% · 54% · 0.8% - GMF: 14.4% · 7.7% · 54% · 0.8% - Matmut: 13.3% · 6.9% · 52% · 1.0% - Abeille Assurances: 8.6% · 6.0% · 69% · 0.7% - Crédit Agricole: 9.4% · 4.6% · 49% · 0.4% - Société Générale: 4.6% · 3.3% · 72% · 0.2% - Crédit Mutuel / CIC: 9.7% · 1.9% · 20% · 0.3% - BNP Paribas / Cardif: 4.9% · 1.6% · 33% · 0.4% Conversion = share of the answers citing the network that also name one of its agencies. ## In each town, 10 networks compete for the recommendation We looked, town by town, at which networks get at least one of their agencies recommended. Competition is dense: in the median town, 10 networks have an agency named regularly (in at least 3 answers), and the leading one gathers only 22% of the agencies named. Share of towns where the network has an agency named regularly (At least 3 local answers name one of its agencies · towns with at least 20 local answers): - Allianz: 93% - AXA: 93% - Groupama: 84% - MMA: 75% - Generali: 69% - MAAF: 64% - MACIF: 61% - MAIF: 59% - Gan: 56% - Matmut: 48% - Abeille Assurances: 43% - Crédit Agricole: 43% - GMF: 41% - Crédit Mutuel / CIC: 31% - Société Générale: 26% - BNP Paribas / Cardif: 21% - 63% of answers in small towns name agencies from at least two networks, versus 46% in large cities. - The towns studied are those where Wispra clients operate, Allianz among them: the networks present in these towns are favoured. No result is published town by town. ## Insurer ranking in ChatGPT and Gemini: Allianz, AXA and Groupama lead An answer cites 4.6 brands from the sector on average. The top three account for 49% of first recommendations. Brand: cited · first - Allianz: 62.5% · 25.3% - AXA: 49.5% · 15.2% - Groupama: 38.8% · 5.2% - MMA: 28.7% · 4.3% - MAIF: 25.2% · 5.0% - Generali: 23.6% · 2.7% - MAAF: 19.6% · 1.3% - MACIF: 15.5% · 1.0% - GMF: 13.9% · 1.9% - Matmut: 13.1% · 1.8% - Gan: 11.1% · 1.9% - Crédit Mutuel / CIC: 9.9% · 1.5% - LeLynx: 9.5% · 1.6% - Crédit Agricole: 9.3% · 0.8% - LesFurets: 8.3% · 1.0% Reading: Allianz is cited in 63% of answers, and first in 25%. The towns studied are those where Wispra clients operate, Allianz among them: networks present locally are favoured (see the method). ## Mortgage, health, savings: where insurers are no longer alone - General-agent networks: 81.4% (first 51.5%) - Mutual insurers: 65.3% (first 19.6%) - Bank insurers: 29.1% (first 5.1%) - Comparison sites: 19.2% (first 4.5%) - Health and protection mutuals: 16.3% (first 5.9%) - Brokers: 11.1% (first 3.6%) - Online insurers: 9.2% (first 3.3%) The three brands most cited for each need: - Car and motorbike: Allianz 75% · AXA 57% · Groupama 57% - Home: Allianz 77% · Groupama 60% · AXA 58% - Health: Allianz 46% · AXA 35% · Harmonie Mutuelle 29% - Life and disability cover: Allianz 62% · AXA 49% · Generali 31% - Savings and retirement: Allianz 40% · AXA 38% · Crédit Agricole 16% - Mortgage insurance: Allianz 44% · Cafpi 43% · Meilleurtaux 34% - Business insurance: Allianz 66% · AXA 57% · MMA 42% - School insurance: Allianz 66% · MAE 57% · Groupama 50% - 49% of answers also cite at least one independent local broker or firm (manual coding of 240 answers). - Comparison sites: 31% / 13% ## What AI reads before recommending an insurer More than half of the sources cited by ChatGPT are insurers' own pages: their agencies' pages (28%) and their national websites (27%). 68% of answers with sources cite at least one agency page. - Network agency pages: 27.8% - Insurers' national websites: 27.1% - Public and official sites: 19.0% - Other sites (local brokers and firms…): 11.8% - Media and guides: 5.0% - Comparison sites: 4.5% - Documents (PDF, policy notices, KIDs): 3.9% - Directories and reviews: 0.6% How often the insurer is cited first (Depending on what the AI read about it before answering): - Its agency page: 32.9% - Its national site or a document: 23.6% - None of the insurer's sources: 14.0% Documents: 1 answer in 10, nearly 1 in 4 on mortgage insurance: - Car and motorbike: 7.4% - Home: 6.1% - Health: 15.2% - Life and disability cover: 10.2% - Savings and retirement: 13.2% - Mortgage insurance: 22.9% - Business insurance: 7.5% - School insurance: 3.8% ## Gemini cites more comparison sites and gives fewer addresses Gemini cites more players per answer (5.3 versus 4.6), cites a comparison site twice as often (38% versus 18%) and relies on directories and review sites in 49% of its answers with sources. It gives an address in only 18% of cases, versus 49% for ChatGPT. ## An AI recommendation is never secured - 29% of cases only give the same first recommendation as a week earlier. - 7% of cases only give the same top three. - 34% of the brands cited reappear from one week to the next, on average. ## Five levers for an agency network 1. **A complete page for every agency** — Address, opening hours, team, products, reviews: it is the source ChatGPT cites most widely, and the one that most often comes with a local recommendation. 2. **Cover every need, not just car and home** — On mortgage, health and savings, brokers, comparison sites and banks take the space left open. 3. **Publish readable documents** — Notices, key information documents, guides: as text, up to date, not blocked. AI uses them to explain cover. 4. **Exist beyond your own website** — Comparison sites, directories, reviews: Gemini relies on them far more than ChatGPT. 5. **Measure every week, agency by agency** — The first insurer recommended changes from week to week; only continuous measurement, town by town, shows where to act. ## How this barometer was built - Corpus. Nearly 13,000 AI answers collected from July 1, 2026 to October 2, 2026. Nearly 7,500 are kept, from more than 1,500 distinct questions and nearly 140 towns; nearly 7,000 of them answer a local question. - Questions. They are written as real requests from individuals and businesses: car, home, health, protection, savings, mortgage, business insurance, school insurance. Each question is asked again every week. Each one was classified individually, with the help of an AI: main product, local or not, individual or business, provider search or information. - No brand in the questions. We excluded every question that names an insurer, a bank, a broker, a comparison site or an agency (26% of search questions), as well as all comparison and brand-perception questions. - AIs queried. ChatGPT through the OpenAI API (gpt-5-nano with web search, located in France and in the town of the question). Gemini through the Google API (Gemini 3.1 Flash-Lite grounded in Google Search), since July 30, 2026. Calls are made with no account or history: answers are close to those of the consumer apps, without being identical. - How a brand is counted. A brand is “cited” if it appears in the answer text or in a cited source, and “first” if it is the first brand of the sector in the text, from a dictionary of more than 80 brands in 7 families. Manual double coding of 240 randomly drawn answers gives 99.7% precision, 99.0% recall and 96% agreement on the first brand. - Named agency. An agency is “named” when the answer designates a specific point of sale of the network: an agency page, an address or phone number right after the brand, or a localised agency name (“MAAF Poissy”, “Groupama agency in Laon”). Checked against manual coding of 90 local answers not used to set the rules: 95% precision, 93% recall. The measure is reliable for ChatGPT; for Gemini it underestimates the agencies named. - Sources. For ChatGPT, the analysis covers the exact addresses of the pages cited. For Gemini, only the domain is available: agency pages and documents cannot be measured there. - Towns and margins of error. Town size comes from official populations (geo.api.gouv.fr). 95% confidence intervals are computed by resampling grouped by territory, because answers from the same territory are not independent. - Limits and conflicts of interest. Wispra works with insurance players, Allianz among them as part of a pilot programme. The towns studied are those where these players operate, and the needs tested reflect their offers. The results therefore describe these territories, not the whole of France, and mechanically favour the networks present locally. No client-specific figure is published: no agency results, no trend. Wispra appears in 2.7% of Gemini answers and in no ChatGPT answer; it is excluded from the rankings. How to cite: Wispra, AI Insurance Barometer, edition #1, October 2026. https://directory.wispra.com/ai-barometer/insurance ## FAQ **How were the questions chosen?** They reproduce real requests from individuals and businesses looking for insurance near them, for every need: car, home, health, protection, savings, mortgage, business, school. Every question that named a brand or an agency was excluded: the questions kept name no player at all. **Why does Allianz come first?** Partly because of the panel: the towns studied are those where Wispra clients operate, Allianz among them, and the needs tested reflect their offers. The ranking describes these territories, not the whole of France. The results on local competition (agencies named, conversion, sources) remain comparable between networks, because the questions never name any brand. **What is a “named agency”?** A specific point of sale of a network, designated in the answer by its page, its address, its phone number or a localised name. A plain mention of the brand (“Allianz offers…”) does not count. On a fresh sample checked by hand, the measure finds 93% of the agencies named, with 95% precision. **Which insurers do ChatGPT and Gemini recommend most?** On the brand-free questions of this barometer, the most cited are Allianz (63% of answers), AXA (50%), Groupama (39%), MMA (29%) and MAIF (25%). This ranking describes the towns studied, where Wispra clients operate: it favours the networks present locally. **How can an insurance agency get recommended by ChatGPT?** The data points to three levers: a complete agency page (when the AI has read it, the insurer is cited first 2.4 times more often), a presence on every need (mortgage, health, savings) and readable product documents. The recommendation changes from week to week: only continuous measurement, agency by agency, shows where to act. **Can I get my network's results?** Yes. Wispra measures every week how often each agency of a network is recommended by AI, town by town and against its local competitors, and helps improve it from a central view. Book a demo to see your network's measurement.