What to expect: every number, labeled

OpenAI publishes no benchmarks, so every number comes from an independent test, a vendor, or OpenAI’s customer stories. They should not be averaged. Across them: click rates of 0.5% to 1%, $3 to $5 a click for consumer offers, consumer sign-ups that convert, B2B tests that did not, and no public conversion data for insurance.

Every ChatGPT ads benchmark you have seen is somebody’s one campaign, somebody’s sales page, or OpenAI’s best customer.

Average them, and you get a number nobody observed. Read them with their labels on, and you can size a test. Here’s every number I could find, and what each one is worth.

How to read this chapter

Three labels. Independent: someone ran a campaign and published the numbers, or an audience-measurement firm published panel data. Vendor: an agency or tool that sells the channel reported its own result. OpenAI: a customer story OpenAI published.

All three can be true. Only the first is checkable, and even that is one account in one month. Use the ranges to size a test, not to forecast.

Click rate

Plan on 0.5% to 1%. A click rate above that is a good card, not a good campaign. The quote form decides the campaign.

Cost

Consumer insurance should price nearer Workshop than Symphonic. The $3.44 on a financial-services campaign is the closest public number to ours.

Conversions

The pattern holds on every row. Consumer sign-ups and purchases convert. B2B lead generation, so far, has not.

Note the dates too. Conversion bidding arrived in early summer. The zero-lead tests ran before or around it, click-optimised. OpenAI’s own Angi story is the switch from clicks to conversions. A test today should look better than a test in May. Only your own test will tell you how much.

What exists for insurance

No independent conversion number. Here is everything close, and why it falls short.

• An Australian vendor, Somantra, reported ads on 71% of 34,279 insurance queries run on two days in August, on desktop. That is an ad-frequency figure from one market, not a performance figure.
• Lapis lists a finance-and-insurance click rate of 0.5% to 1.1%, second lowest of seven verticals, with no basis shown.
• First Page Sage’s table of conversion rates by industry (commercial insurance 1.9%, financial services 0.2%) is a projection. Several guides quote it as measured. It is not.
• Workshop’s financial-services campaign: 0.9% click rate, $3.44 a click, no leads, B2B.

So the benchmark for insurance is the one you make. Fourteen days, $350, one line, one state. Then you own the only insurance number you can trust.

The numbers to plan a test with

Our ranges, from everything above, for a US consumer line in late 2026. Judgment, not measurement.

One more line for the budget. If your account shows the spend $500, get $500 offer, a 20-day test costs $500 and the next 20 days are paid for. That doubles what $500 buys you, so check for it before you set the dates.

Land inside those ranges, and you have a channel to grow. Land outside, and you have a $350 answer. That is cheaper than any other way to get one.

FAQs

Why not average the numbers into one benchmark?

Because a Similarweb panel figure, an agency’s $500 test and an OpenAI customer story measure different things on different dates. An average of them is a number nobody observed.

Where does the 0.21% conversion figure come from?

One advertiser, quoted against a 3.71% Google figure, repeated across several guides. Single source, no detail. I left it out of the tables for that reason.

Will you publish your own insurance results?

Only what clients clear. Everything client-specific in our accounts is under agreement, and this guide holds to the public record.