Write an insurance ad that fits the card

Write the headline in 24 characters and the description in 48. Use the asker’s words, one specific, one trust signal, one next step. Test your logo against an image of the moment. Keep the context hints to the situation. Turn off the setting that lets AI rewrite regulated copy.

Most insurance ads in ChatGPT are cut off mid-sentence. Not because the copy is bad. Because it is too long for the card.

Fix the length, and one agency saw clicks double. Fix the words, and the card reads like the next line of the conversation. Here’s how to write one.

The two numbers that matter: 24 and 48

The headline field accepts 50 characters. The description accepts 100. On a phone, where most of the spend lands, the headline is cut after about 24 and the description after about 48.

Vital Design runs campaigns on the platform. It reports that ads kept inside those cuts saw a 100% jump in click rate against ads written to the limit. An ad that ends mid-word looks broken, and the part you wanted read is the part that vanished.

The industry writes past the cut. The median insurance headline in our 404-creative sample is 25 characters. The median description is 54. So the first edit on most insurance ads is a deletion.

The formula

Four parts. Only the first is backed by a measurement. The rest is how we write them.

1. The asker’s words. Say “first apartment,” not “renters coverage solutions.” The ad is matched to a conversation, and the card that reads like the next line of it gets the tap. Pull the words from the prompts in Chapter 3.
2. One specific. A price floor, a time, a coverage amount, a state. “From $9.” “2 minutes.” “$250k.” Workshop Digital’s test found audience-specific copy (“for B2B manufacturers”) outdrew the generic version. Specifics are what make it specific.
3. One trust signal. “Ohio licensed.” “Since 1985.” The company name in the headline. Workshop saw brand-name-first headlines edge out the rest (1.1% to 1.0%), though the author flags it as confounded. For a regulated purchase, the licence line is the one I keep.
4. One next step. What happens when they tap. “Quote in 2 minutes.” “Apply online.” “Bound in a day.” Not “learn more.”

Six examples that fit, with their counts:

Every number in those examples is a placeholder for yours. Do not run a price you have not filed.

What should the image be?

Here is all the evidence that exists. It disagrees with itself.

What they agree on: the image is tiny, so detail turns to mush and a clear mark reads first. What they do not settle: whether that mark should be your logo, a picture of the moment (a front door, a car, a dog), or a face.

So test it. Logo against moment, two ads per ad group, readout on day 14. Square, at least 640 by 640, no text in the image, nothing that looks like stock.

Context hints that match the ad

Hints are free text at the ad group level. OpenAI’s pattern: offer, who it helps, the situation, one fit detail. Their example: “Same-week air-conditioning repair by licensed technicians for homeowners in Dallas.” The insurance version:

• Renters insurance for people moving into a first apartment in Ohio, starting around $9 a month, licensed in Ohio.
• Auto insurance quotes for parents adding a 16 to 19 year old driver to a Texas policy.
• Pet insurance for new dog and cat owners comparing vet bill reimbursement plans.
• Restaurant and bar insurance for owners opening a first location, general liability and liquor liability, bound within a day.

Three rules. One need per ad group, so the ad and the hint say the same thing. Keep the list short; Symphonic reports that long, comprehensive hint blocks stopped delivery. And don’t write instructions (“only Chicago,” “not for students”). The hint guides matching. It enforces nothing. Location is a setting, not a sentence.

The landing page

A page that answers the question the ad was matched to, with the quote form above the fold. Not the homepage.

Workshop Digital measured headlines and descriptions truncating earlier than expected and clicks arriving deep in a specific conversation. The page has to pick up that conversation. The licence line and the state belong on it. OpenAI reviews the landing page with the ad, and its crawler has to reach it, so no sign-in walls and no blocked bots.

Want a model? NEXT’s ChatGPT landing page, which we clicked through to in Chapter 3: a dedicated URL, an industry picker above the fold, a rating with the review count, a customer count, the same starting price the ad promised with its footnote, and every button going to the quote flow.

Expect leakage between the click and the session. Symphonic counted 53 platform clicks and 35 sessions in analytics. Workshop’s B2B software peer saw about half. A page that loads in one round trip keeps more of them. Chapter 6 is how you count the ones that arrive.

The switch to turn off

Creative enhancement (also labelled text customisation) is on by default. It lets the platform rewrite your headline and description to match the conversation, and translate it. OpenAI’s webinar recommends leaving it on.

Its own wording in Ads Manager: the rewritten copy “may be shown without your individual review or approval.” For state-filed language, rate disclosures, and anything your compliance officer signed, that is the wrong trade. Turn it off. Write two versions yourself. Let the auction choose.

Before you spend: ask ChatGPT the question ten times

Two sources, Symphonic and Search Engine Journal, give the same advice. I agree with it.

Open ChatGPT on a free account. Ask your target prompts ten times each. Read what it says about your category and your brand. If the organic answer calls you expensive or names your rival, an ad under that answer is paying for a hostile introduction.

Fix the answer first. The six steps are in our guide to getting recommended.

FAQs

Can I put the price in the headline?

If it is a filed, real starting price, yes. 11.4% of insurance creatives do. “From $9” is a specific. “Cheap” is not.

Should the headline be a question?

8.4% of insurance creatives ask one. The prompts are already questions. The card that answers reads better than the card that asks another.

How many ads per ad group?

Two to start, differing in one thing: the image, or the specific. Three, and a $25 budget cannot tell them apart in 14 days.