Insurance: ChatGPT becomes a distribution channel. What insurers should prepare
For years, getting three insurance quotes meant filling in the same form three times. In 2026, a conversational assistant can do it for the policyholder. The first insurers and comparison sites opened a quote inside ChatGPT; supervisors are watching; brokers are asking questions. Here are the facts, and what they require to prepare.


What happened on 9 February 2026
That day, Tuio, a Spanish digital insurer, put an app online inside ChatGPT: the policyholder describes their home in the conversation, gets an indicative home insurance quote, then is sent to Tuio's website to subscribe. The infrastructure was provided by Waniwani, a young US company. The shares of the large listed brokers fell sharply during the day (Yahoo Finance France, 9 February 2026).
The market reaction rests on a simple idea: insurance distribution has long relied on how hard it is to compare. If an AI agent compares on the policyholder's behalf, that difficulty disappears. That is the analysis Waniwani's chief executive, a vendor selling this type of service, gives in an interview published by McKinsey in October 2026: the friction of distribution fades, and the centre of gravity moves back to underwriting and price (McKinsey, October 2026).
What the first movers do, in France and elsewhere
In France, the comparison site lesfurets launched its ChatGPT app on 16 April 2026: four questions, a first estimate, then a hand-over to the website for the full quote. It reports that traffic from conversational engines represents about 1% of its traffic (L'Assurance en mouvement, 21 April 2026). APRIL opened "APRIL Moto" inside ChatGPT on 15 June 2026, with a personalised rate in the conversation (press release, 15 June 2026). Leocare offers a motor quote inside ChatGPT from the registration plate, with the subscription completed on its journey (Leocare, 17 June 2026).
In the United States, Liberty Mutual offers a motor quote inside ChatGPT and Plymouth Rock followed in August 2026 (Insurance Business, 28 August 2026). At Tuio, lesfurets and Leocare the pattern is the same: estimate inside the assistant, subscription on the insurer's or comparison site's website. For APRIL and the US insurers, we did not read where the subscription is concluded. None of the launches we read describes a sale concluded inside the conversation.
A point of method: we only cite launches whose source we have read. Other names circulate in the trade press; we do not repeat them until we have read the announcement.
What it changes for an insurer or a broker
The journey must be readable by an agent. A quote funnel designed for a person (conditional fields, attachments, anti-bot checks) stops an agent without the insurer seeing it. The McKinsey interview puts it another way: tell the human visitor from the agent, and give the agent direct access to the rating engine rather than forms to fill in.
Visibility is measured on policyholders' real questions. Being cited in an assistant's answer cannot be decreed. You have to identify the questions policyholders actually ask, measure the brand's presence in the answers, then adapt content and structured data. Waniwani's chief executive stresses it in the same interview: extrapolating classic search keywords has its limits; a policyholder's question is not phrased like a Google query.
Retention changes in nature. If an agent can check prices continuously, charging a long-standing customer more becomes hard to sustain, and a customer may leave before the insurer has recouped the acquisition. Waniwani's chief executive sees it as a return to fundamentals: understanding risk and offering the right product at the right price. We have no figure on this effect; it is an analysis, not a measurement.
The number of quotes may rise without sales following. That is also a point of the McKinsey interview: when asking for a quote costs nothing, the number of requests may rise without an equivalent rise in purchases. The channel's economics must therefore be measured, quote by quote, with a log, before being generalised.
What supervisors say
In France, the ACPR (the French prudential supervisor) opened in summer 2026 a discussion paper and a consultation on artificial intelligence, closed on 30 September 2026, with guidance expected by year end (L'Assurance en mouvement, 3 August 2026). At European level, EIOPA states that its 2025 opinion on AI adds no new rule and that the Insurance Distribution Directive remains fully applicable, including customer information when interacting with an AI (EIOPA, 16 September 2025).
In practice: it is for each insurer's compliance team to qualify what a quote returned by an agent is and which information duties attach to it. The European AI regulation (Regulation (EU) 2024/1689) ranks, as EIOPA reads it, pricing and risk assessment in life and health insurance among high-risk uses (EIOPA, 16 September 2025); a motor or home quote is not designated as such in the texts we read. We give no legal opinion: it is for each insurer's compliance team to qualify its use.
And in Tunisia?
The Tunisian market counts 24 insurance companies and its net premiums reached 4,255.1 million dinars in 2025, of which 1,628.3 million for motor (Comité Général des Assurances, quoted by Managers.tn, 3 March 2026). Public digital initiatives cover online quotes and management apps: GAT offers an online quote and the MyGAT app, COMAR launched COMAR Plus in August 2024, and MAE Assurances (a Tunisian mutual insurer, unrelated to the author of this article) announced an AI-based conversational agent in September 2025 (Financial Afrik, 12 September 2025). We found no Tunisian insurer app inside a conversational assistant to date.
For a Tunisian insurer or broker, the first gain seems to us internal: reading policies and claim files, which may exist as scans, in French and Arabic, and alerting handlers on deadlines and missing documents. With one constraint to respect: organic law 2004-63 (article 52) requires the authorisation of the INPDP (the Tunisian data protection authority) for any transfer of personal data abroad. On-site processing, with an open model, avoids that transfer for the reading; the overall compliance of the processing remains for the insurer's data protection officer to qualify.
Where to start
Three entry points, depending on the situation. If you distribute online: an agent-ready assessment, which measures what an agent understands of your offer and where it gets stuck. If your rating engine can be queried and your compliance team has a position: AI agent quotes on a single product line, logged and measured. If your concern is portfolio management: policy and claims agents installed on your premises, with a trial on a batch of real policies before any commitment.
Two agents complete the set: the prospecting and follow-up agent, for inbound requests and unconverted quotes, and the renewal and retention agent, to take the initiative at renewal. In every case the rule is the same: no promise of presence in an assistant, no figure announced before measurement, and the decision left to the adviser, the handler or the policyholder. The Agentic insurance section presents the eight offers, including training, unpaid premium recovery and tender reading.
What decision-makers ask about this topic.
Is a quote returned inside ChatGPT binding for the insurer?+
In the launches we read in 2026, the assistant returns an estimate presented as indicative and sends the prospect to the insurer's or comparison site's website for the full quote and the subscription. The legal value of a quote returned by an agent is for each insurer's compliance team and terms to determine.
Do you need to be inside ChatGPT to exist?+
Nobody can guarantee it, and each platform's integration terms change. What can be prepared now: a quote journey readable by an agent, content assistants can cite, and a measurement of your presence on your policyholders' real questions.
What share of traffic comes from assistants today?+
The only public figure we read is lesfurets': about 1% of its traffic in April 2026, according to its own statements. We have no figure for insurers.
What about a Tunisian insurer?+
To our knowledge, no Tunisian insurer has opened a quote inside a conversational assistant. The first gain seems to us internal: reading of policies and claims by agents installed on site, with no data transferred abroad.
Does your program match this case?
We frame every program at Intake, with transparent budgeting. A short POC of a few weeks can be delivered before committing to the full program.
Scope an agentic insurance pilot →