Open an account at ads.openai.com and build one campaign, with one ad group for each insurance need, all in the same states. Once your tracking is live, pick Conversions, billed on impressions if Ads Manager offers it. Set $25 a day and start with Intelligent bidding. Only target states you’re licensed in, keep your hints short, run two ads, turn creative enhancements off, and on day 14 judge it by what a started quote cost you.
Update, October 30, 2026: OpenAI now turns creative enhancements on by default. Turn them off in Ads Manager (Step 5).
Three defaults need changing. The budget pre-fills at $100 a day, which turns a $350 test into $1,400. The Clicks objective never learns what a quote looks like. Creative enhancements let AI rewrite copy your compliance officer signed.
Step 1: open the account
Go to ads.openai.com and sign in with a ChatGPT account. The new onboarding asks for your website and drafts a first ad and campaign from it. Edit it or discard it.
Name the account. Pick the country, currency and time zone. Those three cannot be changed afterward, so pick the ones the business reports in.
Add a card. Billing is postpaid, charged when unpaid spend reaches a threshold or at month end, and nothing serves without a payment method. Expect a business verification step. Guides quoting OpenAI’s help text put it at up to seven business days. As an insurer, have your licensure ready, because the financial-services policy says OpenAI may ask for it. Health and Medicare plans fall under OpenAI’s health services policy instead, and Medicare ads also follow CMS marketing rules, which can require a third-party marketing (TPMO) disclaimer and HPMS submission, depending on your role and the plans you represent.
Using an agency? The advertiser owns the account and the agency is added as a user.
Step 2: build the campaign
The structure is Campaign, then ad group, then ads. Objective, budget and geography sit on the campaign. Context hints and bids sit on the ad group. Copy and image sit on the ad.
Run one campaign for the test. Because geography sits on the campaign, every ad group in it serves in the same states. One ad group per need inside those states: renters for first apartments is one, renters for people moving cities is another. Two ads in each. A different state needs its own campaign.
OpenAI recommends fewer, better-funded campaigns. Its example: a budget split across two campaigns beats the same budget across ten. On $25 a day that means one campaign and at most two ad groups. Split it further and the learning never finishes.
Step 3: objective, budget and bid
Objective. Pick Conversions. If Ads Manager offers it billed on impressions, take that: OpenAI says it also learns from people who saw the ad and acted within a day without clicking. If your account only offers Conversions billed on clicks, that works too. Your pixel and server events need to be live before launch (Chapter 6). Map the quote start to a standard event such as lead_created, because custom events cannot be the goal. Tracking not ready? Choose Clicks for the first week and switch. You cannot change the objective on a running campaign, so that means a new one.
Promotional credit. Check Billing for a sign-up offer and read its terms. With a spend $500, get $500 offer, plan 20 days instead of 14 so the test qualifies (the test plan has the math). With no offer, run the 14 days and decide.
Budget. $25 a day, the minimum. The field pre-fills at $100, so check the budget before you launch. Spend can reach $50 on a single day and $175 in a week on a $25 budget, so set the expectation with finance before the first invoice.
Bid. For the first test, start with Intelligent bidding, formerly Maximize Results. OpenAI handles the bids while you measure what a started quote costs. From November 9, 2026, new conversion campaigns will offer Target CPA instead of Max Bid if you want to set a cost target. New click campaigns will offer Target CPC.
Before choosing a target, check what you’ve recently paid for the same conversion and allow time for conversions to show up in the report. A $50 Target CPA for started quotes asks OpenAI to aim for a $50 average. Individual quotes and the average can cost more or less, so keep watching actual cost per started quote.
If you’re already using Max Bid or Max CPC, you can start switching on November 9. Plan the switch before those options retire on January 1, 2027. You still pay for clicks or impressions, depending on your billing setting. Chapter 1 explains the choices.
Step 4: targeting
- Location. Include only the states you are licensed in, then check that the campaign summary lists only those states. Location is estimated, so ask for the state on every quote form. Avoid ZIP-level targeting. Choosing neighborhoods for an insurance offer can raise unfair-discrimination questions. Brookings also found lower-income ZIP codes got more ads.
- Platform. Leave all five on (iOS app, Android app, iOS web, Android web, desktop web). Read the device split on day 7.
- Custom audiences. Skip inclusion until you have 25,000 matched people. With compliance sign-off, upload hashed policyholder emails as an exclusion list so the acquisition ad stops showing to customers. No minimum for exclusion. Check that your privacy notice covers sharing with an ad platform.
- Negative Phrases. OpenAI announced them on October 5 for qualifying advertisers, for narrow brand-policy needs. Leave them out of a first test.
Step 5: context hints, the two ads, and creative enhancements
Start with about five hints per ad group, each a sentence that describes the offer, who it helps and the situation. In a second round, test short keyword-style hints against them. Examples are in Chapter 4. Do not write a list of exclusions, and keep the block short: some advertisers report long hint blocks stopped delivery.
When Ads Manager asks for the ad format, pick a single image for this first test. Multi-image ads are new, and one image keeps the two-ad comparison clean.
Two ads per ad group. Same headline and description with two images (logo against a picture of the moment), or same image with two specifics. Headline inside 24 characters. Description inside 48. Square image of at least 640 by 640 pixels. Landing page that answers the prompt.
Then turn creative enhancements off. As of Friday, October 30, 2026, OpenAI turns creative enhancements on for new and existing campaigns unless you have opted out. They can rewrite and translate your text, resize your image and change your landing page address. Enhanced ads “can run automatically without requiring you to review or approve each version.” Turn them off for the account in Ads Manager, under Manage settings. Then check each campaign, because a campaign’s own setting overrides the account default. Opt-outs you set before October 30 carry over.
Step 6: the checks before launch
- The headline and description fit a small phone. Check them in the ChatGPT ad preview.
- The pixel fires on the quote page and the thank-you page. Watch the event stream in Ads Manager and see the events arrive.
- Server events deduplicate against the pixel (same event ID). Check the Event Quality Score and fix anything Ads Manager flags.
- HubSpot is connected with Auto tracking on, so UTMs are added and contacts attribute to the ad.
- The landing page loads without a sign-in wall, and OpenAI’s crawler is not blocked in robots.txt.
- The license line, the state and any filed disclosures are on the page. Creative enhancements are off for the account and on every campaign (Ads Manager, Manage settings).
- The daily budget reads $25, not the pre-filled amount.
- The quote form collects TCPA prior express written consent before any automated marketing call or text. The consent names the company or companies that may contact the applicant and sits apart from your general terms. Florida, Oklahoma and other states add their own rules. Life and health forms send no health answers in pixel events or URLs.
- If you write commercial lines or an MGA program with a strict appetite, put one knock-out question (business class or state) on the form before the conversion fires. Then ChatGPT learns from the risks you want, not from submissions you will decline.
Step 7: the 14-day readout
Days one to three: change nothing, no bids and no new hints. Confirm pixel and server events arrive each morning and check the spend each evening.
Day seven, the first read: impressions, clicks, cost per click, and clicks that reached the site as sessions. Check the Event Quality Score. Ad hasn’t served at all? Check Ads Manager for serving issues and confirm real conversion events are arriving. If you set a target, compare it with recent costs for the same conversion. A low target, too few chances to show the ad, or missing or poor conversion data can all leave part of the budget unspent. Check those before raising the budget. Served, but clicks are not reaching the site? The problem is the page, not the ad.
Day fourteen, the decision: cost per started quote, counted in your CRM. Ads Manager’s conversion count also includes people who only saw the ad, within a day by default. $350 buys about 70 to 117 clicks at $3 to $5 a click, and about 1 to 6 started quotes. That is a first read on the channel, not proof, and far too few quotes to pick a winning ad. Compare the two ads on click rate instead.
Compare the cost of a started quote to what Google gives you for the same line. Then decide: keep and widen to a second state, rewrite the losing ad, or stop. Write the three numbers that would trigger each decision before day one, as Search Engine Journal advises, so the readout cannot talk you into anything.
FAQs
What if I cannot get Conversions billed on impressions?
Then take Conversions billed on clicks. That works too. Conversion bidding has to be enabled on the account. If your account offers no Conversions objective at all, run Clicks with the pixel live for the two weeks. The conversion data accrues either way, and the next campaign can switch.
Can I pause for the weekend?
You can set start and end dates, but not days or hours. You can pause the campaign by hand. For a 14-day test, don’t. Weekend prompts are real, and stopping and starting muddies a 14-day read.
Can I set this up from inside ChatGPT?
Yes. OpenAI’s Ads Manager plugin lets you ask ChatGPT to draft ads, build the campaign, check results and suggest fixes. It shows you every change and waits for your OK before it saves anything, so nothing runs without your review. Billing, payment and identity checks still happen in Ads Manager, and it can’t report revenue. If it writes copy or makes an image for you, send it through compliance like any other ad.
How do I run the same test in three states?
Three campaigns with the same ad groups, which is what bulk CSV upload is for. But not on $25 a day each in the first two weeks. One state first.
Does the $500 credit cover the test?
Only if your account shows the offer, and only once you’ve spent $500. A 14-day test at $25 is $350, so it stops short. Run 20 days and you hit $500 and unlock it. Offers vary by account, so read the terms first.
Sources, checked October 27, 2026:
- Set up your ChatGPT Ads for success, OpenAI webinar (onboarding, fewer campaigns, objectives, impression billing)
- OpenAI help center, billing for ads
- Use the ChatGPT Ads Manager plugin, OpenAI help center (drafts and changes need your confirmation; billing stays in Ads Manager; no revenue reporting)
- OpenAI help center, daily budgets (twice the daily budget in a day, seven times in a week)
- OpenAI help center, conversion-optimized campaigns
- Creative enhancements for ChatGPT Ads, OpenAI help center (the October 30 default)
- OpenAI advertising policies (financial services, health services)
- OpenAI ads developer documentation (standard events, targeting, attribution windows)
- OpenAI, new ChatGPT ads format and measurement (Negative Phrases)
- Search Engine Journal, five questions (decide the criteria before launch)
- Brookings, the audit (delivery by ZIP income)
Bidding source, checked November 5, 2026:

