Insurance has always been a business of words and numbers. Policies, endorsements, exclusions, claim forms, adjuster notes, medical reports, loss histories: the industry runs on mountains of documents that humans have spent decades reading, summarising and re-keying. That is exactly why generative AI, a technology built to read, write and reason over language, is landing in insurance with unusual force.
For the last two years most insurers treated GenAI as an experiment. In 2026 that is changing. Carriers are moving from scattered pilots to enterprise-wide programmes, and the impact is starting to show up in claims cycle times, underwriting capacity and customer satisfaction. Here is where the change is happening, and what to watch out for.
Claims: where GenAI landed first
Claims is the moment of truth for any insurer, the point where a promise on paper becomes money in a customer’s account. It is also where generative AI has made its most visible mark.
A typical property or motor claim involves photos, repair estimates, police reports, emails and phone transcripts. GenAI models can ingest all of it, summarise the file for the adjuster, flag missing documents and draft the customer letter. In property and casualty lines, systems can now produce assessment letters and settlement offers from structured claims data in seconds rather than days.
The adjuster does not disappear. Instead, they spend less time on paperwork and more time on judgement calls: complex losses, disputed liability and customers who need a human voice. Claims went first because the work is high-volume, document-heavy and easy to measure, which makes the return on investment simple to prove.
Underwriting: slower to change, bigger prize
Underwriting is the heart of an insurer, and it has proved harder to automate. Commercial submissions arrive as messy bundles of broker emails, schedules, loss runs and inspection reports. Underwriters can spend most of their day just extracting and organising information before any real risk assessment begins.
GenAI “copilots” change that equation. They read a submission, pull out the key exposures, compare them against underwriting guidelines and highlight anything unusual. Some analysts estimate this can roughly triple an underwriter’s throughput without loosening discipline. Industry research suggests the technology can shrink quoting timelines from weeks or days to hours in some cases.
The reason underwriting has moved more slowly is not model quality. It is accountability. Pricing a risk is a regulated, high-stakes decision, and insurers rightly want a human signing off.
A new kind of customer experience
Most people interact with their insurer twice: when they buy a policy and when something goes wrong. Both moments are often frustrating. Policy wording is dense, call centres are busy, and simple questions can take days to answer.
Generative AI assistants can explain coverage in plain language, answer “am I covered for this?” questions instantly and guide customers through filing a claim at two in the morning. Swiss insurer Helvetia was among the first to launch a customer service bot built on ChatGPT technology, and it has since expanded AI across claims, fraud and underwriting.
For agents and brokers, GenAI drafts quotes, comparison summaries and renewal emails, letting a small team serve far more clients without adding headcount.
Fraud: a double-edged sword
Insurance fraud is enormous. The Coalition Against Insurance Fraud estimates it costs the US industry around $308 billion a year. GenAI helps by scanning claims narratives, documents and images for inconsistencies that a tired human reviewer might miss.
But the same technology that detects fraud also makes it easier. Fake invoices, doctored photos of vehicle damage and synthetic medical records can now be generated in minutes. Insurers are responding with detection tools that check image metadata, look for tell-tale generation artefacts and cross-reference claims against external data. Expect this arms race to define the next few years of claims security.
The risks insurers cannot ignore
GenAI brings real dangers alongside the efficiency gains.
- Hallucination. A model that confidently invents a policy clause or misreads a medical report can cause real harm to a customer.
- Bias and fairness. Regulators are watching closely for discrimination, opacity and data misuse in AI-assisted underwriting and claims decisions.
- Oversight. Researchers have raised concerns about thin human review in AI-driven decisions, such as prior authorisation in health insurance.
- New liabilities. Insurers are also underwriting the risk of their customers’ AI. New ISO endorsements for 2026 let commercial liability policies exclude claims arising from generative AI, such as IP infringement or defamatory outputs.
The insurers getting this right treat governance as a foundation, not an afterthought: clean data, access controls, audit trails and a clear rule that a human owns every consequential decision.
What comes next
The next wave is agentic AI: systems that do not just summarise a claim but take actions, such as requesting documents, scheduling an inspection or triggering a payment within set limits. Surveys from Celent suggest roughly one in five insurers expect an agentic solution in production by the end of 2026.
The prize is significant. McKinsey estimates GenAI could unlock $50 to $70 billion in additional annual revenue for the industry. But the winners will not simply be those who deploy the most models. They will be the insurers who pair speed with trust, using GenAI to make insurance faster and clearer while keeping people accountable for the decisions that matter.
Insurance has always been about managing risk. With generative AI, the industry now has to manage its own.
