ChatGPT ads for insurance: the guide

ChatGPT ads work if you sell policies to people: renters, auto, pet, life, small-business cover. They are worth a test if you recruit agents. They are not worth your money yet if you sell to insurers, because your buyers run ChatGPT on paid plans and paid plans show no ads. A clean test costs $25 a day for 14 days. You decide on quotes started, not clicks.

People now ask ChatGPT whether they need renters insurance, which pet plan is worth it, and what a 19-year-old pays for auto cover in Texas. Those are buyers. And since February, there has been an ad under the answer.

Miss that, and your competitor’s card sits under the question your customer just asked. Miss enough of them, and you lose the quote before you knew it existed.

I run ChatGPT ads for insurance and software clients, and the HubSpot behind them. Before that, we ran Rainbow’s go-to-market, which appointed 4,000+ agents across 20+ states in under a year. This guide is what I would tell an insurance CMO who asked me, straight, whether to put money into ChatGPT.

Everything in it comes from four places: OpenAI’s own documentation and webinars, three public ad libraries we read creative by creative, independent tests other agencies published with their numbers, and the Brookings audit that came out today. When a number is a vendor’s claim, I say so. When it is my judgment, I say that too.

Let’s break down what the ad is, who should run it, what insurers are doing right now, and what a test costs.

Want it run for you? See ChatGPT ads for insurance and startups.

What is a ChatGPT ad?

One card under the answer. A lead line (“This might also interest you”), then the card: an image, the advertiser’s name, a grey Ad label, a headline and a description. No web address. That’s it.

It never appears inside the answer. And the answer does not change because you paid. OpenAI says so plainly, and Search Engine Journal’s Brooke Osmundson put it in seven words: “OpenAI keeps ads separate from its answers.”

Here’s what matters: the ad is matched to the conversation, not a keyword. Someone asks about renters insurance for a first apartment. A renters insurer that described that situation in its targeting can show up. Someone asks what their policy covers after a bike theft, and the next day, while planning a weekend trip, they can see an insurer’s ad again. OpenAI matches on the current conversation, past chats, and past ad interactions.

Chapter 1 has the mechanics.

Why do insurance questions live in ChatGPT now?

Because the research moved.

OpenAI’s team calls it “collapsing the consideration layer.” The six steps of a purchase that used to happen across six sites now happen in one chat. Their own split of shopping activity in ChatGPT: 5% framing the problem, 31% discovering options, 22% comparing, 23% evaluating one option, 13% buying, 16% after the purchase.

Add it up. 76% of the activity happens before anyone pays.

Insurance is a research-heavy purchase with a confusing vocabulary. It fits this pattern better than most products. The prompts in the ad libraries read like a kitchen-table conversation: “I’m 27, got my license in Ohio last year and had one minor fender bender already.” Nobody types that into Google.

Who should run ChatGPT ads, and who should wait?

Three kinds of insurance business. Three different answers. I would rather say that up front than sell everyone the same test.

• Selling policies to people: run it. Renters, auto, pet, life, small-business cover. The questions are there in volume, and a quote form behaves like the sign-ups OpenAI reports on.
• Recruiting agents: test it. Producers ask about carriers and appointments. Keep it to $25 a day. Read the financial-services and hiring rules first.
• Carriers and distributors selling to insurers: not yet. Your buyers run ChatGPT on Business and Enterprise plans. Those plans see no ads. The two B2B tests with published numbers got 53 and 739 clicks and zero qualified leads between them.

The data behind each verdict is in Chapter 2.

What are insurers running right now?

We pulled every insurance creative from three public ad libraries and read them. 404 unique creatives from 101 advertisers, captured June to October 2026. Carriers wrote 157. Comparison sites 112. Media brands 80. Agencies 25.

On the largest library, EverQuote, GEICO and The Zebra alone took 31.5% of the 6,075 insurance ad impressions it recorded.

Two findings changed how I write targeting. First: name a brand, and the brand rarely shows up. Of 231 ad appearances on brand-naming prompts, 23 were the named brand, 68 a rival carrier, 140 a comparison site or media brand. Second: 312 of 1,279 prompts that triggered an insurance ad contained no insurance word at all.

Then we asked ChatGPT three cyber-insurance questions ourselves, signed out, on October 27. Three questions, three ads: Hiscox once, then a broker lead-gen card twice, while the answer itself named Hiscox, The Hartford, Chubb and Coalition. Chapter 3 has the counts, the examples, the real card, and the limits of the data.

What does it cost to find out?

$350 and two weeks.

The platform’s minimum is $25 a day. Fourteen days gets you through the learning period and gives you quotes started per $100 spent. Before day one you need the pixel and server events live, HubSpot connected, two ads per ad group, and the creative-enhancement setting off. Then leave it alone for three days.

Then check one thing in your account. New advertisers are often offered free ad money: spend $500, get $500 back to keep running. If you see that offer, stretch the test to 20 days. 20 days at $25 is $500, so the credit kicks in and pays for your next 20 days. That’s 40 days of results for $500. If you don’t see the offer, stop at 14 days and decide.

Don’t read anything into day two. Symphonic Digital’s test needed a week and a half of $1-a-day bid increases before it spent at all. OpenAI’s own advice is fewer, better-funded campaigns with room to learn.

Chapter 5 walks through every setting. Chapter 6 wires the tracking.

What would I fix first in most insurance ads?

Length.

The headline field takes 50 characters. The description takes 100. Almost every insurer writes to the limit. A phone shows about 24 and 48. The median insurance headline in our sample is 25 characters and the median description is 54, so most of them lose the end of the sentence.

Vital Design, which runs campaigns on the platform, saw click rate double on ads that stayed inside the cut. One change. Double the clicks.

The second fix? The image. The evidence is thin and it conflicts, so Chapter 4 lays out all three tests and tells you what to A/B, not what to believe.

What does the data say, and what doesn’t it say?

OpenAI publishes no benchmarks across advertisers. None.

Everything you will read about click rates, costs and conversions comes from one of three places: an independent test with its numbers shown, a vendor that sells the channel, or OpenAI’s customer stories. They do not agree with each other. They should not be averaged.

What they do agree on: click rates sit between 0.5% and 1%. A click costs $3 to $5 for consumer offers and more for B2B. Consumer sign-ups convert. B2B lead generation, so far, has not. And there is no public conversion data for insurance.

Chapter 7 lists every number with its label. Chapter 8 lists the claims in other guides that the platform itself contradicts.

FAQs

Do ChatGPT ads change what ChatGPT says about my company?

No. The ad is a separate card under the answer, and OpenAI says ads do not influence the answer. If ChatGPT names a competitor when someone asks for the best renters insurer in your state, that is an organic problem. How to get your insurance company recommended by ChatGPT covers it.

Can a health insurer or Medicare plan advertise?

Health insurance is restricted to approved US advertisers, and OpenAI shows no ads in health-related conversations at all. So a health plan can be approved and still find few places its ad can appear. Wait on health lines. Test on property, auto, pet or life first.

Is there really $500 of free ad money?

Maybe. OpenAI shows this offer inside Ads Manager when you sign up. It isn’t on a public page, and the amount isn’t the same for everyone: trade press reported $50 and $100 versions earlier this year. We saw $500, and OpenAI showed the same offer in its own webinar demo. No end date is shown. There’s no referral link for it, from us or anyone. Look in your own Ads Manager before you plan around it.

How is this different from being recommended by ChatGPT?

Being recommended is organic. ChatGPT names you in the answer because the web says you are good. Ads are paid placement under the answer. They are measured separately, they do not help each other, and you decide each on its own evidence.