The ChatGPT Ads Checklist: Tick Every Box Before You Spend a Dollar

Henry Purchase

Co-Founder

I am putting $10,000 into ChatGPT Ads to find out what actually works. This is the checklist I run before every single launch.

It is not a generic "get ready for AI advertising" list. It is the specific set of things I verify in the Ads Manager, in DevTools, and on my landing pages before I let a campaign spend money, built from real campaigns and real mistakes. There are six sections: tracking, targeting, creative, funnel, benchmarks, and a first-seven-days plan.

Work through it in order. Every box you tick before launch is money you do not waste after.

Why a checklist matters more on ChatGPT than on Google

Most channels forgive a sloppy launch. You waste a bit, you notice, you fix it. ChatGPT Ads are less forgiving for two reasons.

First, the reporting is thin. The dashboard does not surface problems the way Google or Meta does, so a broken setup can run for days before you notice the numbers do not add up. Second, the platform is three months old. There are no mature guardrails, no established benchmarks to sanity-check against, and at least one piece of the default setup actively works against you. You catch these things before launch or you pay for them after.

This checklist exists to catch them before.

Section 1: Tracking

The single most important section, because if tracking is wrong, every other number you look at is fiction.

The conversion script ChatGPT gives you is broken. This is finding number one and it catches almost everyone. The copy-paste code from the Ads Manager UI fires a conversion every time someone loads any page on your site. Your dashboard fills with conversions you never got, your bids move toward the wrong actions, and you scale a campaign that is actually losing money.

The fix: keep the init pixel script global on every page, but only fire the measure() call from the conversion event itself, either from the success page only or from the JS event that confirms the action (form submit, checkout complete). Pull the current snippets from OpenAI's official ads documentation rather than the in-app UI, and check carefully where each call sits. The full walkthrough is in the ChatGPT Ads tracking guide.

Your tracking checklist:

• The measure() call fires only on the conversion event, never on page load. (Trap: the copy-paste code does the opposite.)
• Verified in DevTools: with debug: true, complete a real conversion and confirm the [oaiq] event queues once. Reload normal pages and confirm it does not log again.
• The event you fire on conversion is the one that maps to revenue. (Trap: firing on email signup when the real conversion is a paid trial.)
• Your landing page does not block OpenAI's crawlers, OAI-AdsBot and OAI-SearchBot. Check your robots.txt and any WAF rules. (Trap: a default bot-blocking rule silently breaks your ads.)

Section 2: Targeting (context hints)

A hint that is too narrow will not deliver. A hint that is too broad will not qualify. The trick is knowing which one yours is before you spend.

OpenAI calls them context hints. They describe the conversations, topics, and situations where your ad is eligible to appear. They are thematic signals, not exact-match keywords. Two things to know before you launch.

Volume of distinct ads beats one perfect ad. OpenAI's own guidance is explicit: a high and diverse volume of ads, meaning multiple title and copy variations per offering, lets the system match across a wider range of conversations. Launch one hint with two ads and you have under-supplied the auction. Ship several distinct angles per hint, not several wordings of the same line.

You cannot feel a hint's reach by reading it. One of my early ad groups got 50 times fewer impressions than another with identical keywords, purely because the audience description was tighter. The lesson was not "broad is always better" (OpenAI's guidance is to prioritise relevance over coverage). The lesson was that your gut is unreliable here, so you have to check. Run your hint through the Focal Reach Calculator to see estimated audience size and CPC before launch, or build one from scratch with the Hint Generator.

Your targeting checklist:

• Hint validated through the Reach Calculator before launch, so you know its estimated audience size and CPC.
• At least four ad variations live, each a different angle (FOMO, social proof, founder story), not different wordings of the same line.
• Delivery threshold set: under 100 impressions in 48 hours means the hint is too narrow. Broaden it.

Section 3: Creative

Your ad sits inside a ChatGPT response. If it sounds louder than the response it sits within, the tone is wrong.

The numbers here come from Adthena's analysis of 40,000+ ChatGPT ad placements, reported via Search Engine Land. Headlines average 30 characters with a ceiling around 36. Body copy averages 19 words across two sentences. The strongest CTAs name what the user is about to do and include the offer where there is one ("Start Free Trial," "Get 20% Off"), not "Learn More."

The dominant headline pattern across top performers is "Brand: Benefit", the brand name, a colon, a specific value claim. Specific numbers in body copy (a dollar figure, a percentage, a count) beat vague claims. Exclamation marks, superlatives, and pattern-interrupt hooks are uncommon among the strongest ads, which makes sense when your ad is sitting inside a calm, helpful answer.

Your creative checklist:

• Headline uses a "Brand: Benefit" pattern and stays under 36 characters.
• Body copy is two sentences, around 19 words, with at least one specific number.
• CTA names the action or offer ("Start Free Trial," "Get 20% Off"), not "Learn More."

Section 4: Funnel and economics

Whoever can spend the most to acquire a customer wins. Your funnel is the real ceiling, not your ads.

If your funnel converts at 1% and a competitor's converts at 4%, they can outbid you four times over and still stay profitable. No amount of clever targeting fixes a leaky funnel. Fix the page before you fund the ads.

One thing that surprises people: most ChatGPT use is on mobile, but most ChatGPT referral traffic, around 94%, comes from desktop (per BrightEdge and Search Engine Journal). The audience uses both, so both have to work. Open your landing page on a phone and on a desktop, time how long it takes to actually click the button on each, and fix whichever is slower before you spend.

Your funnel checklist:

• Landing page tested on mobile and desktop: above-the-fold CTA, loads in under 3 seconds, no more than 3 form fields. (Trap: sending paid traffic to your homepage. OpenAI's own guidance is to use product, collection, or content pages instead.)
• Daily budget small enough that a bad first 72 hours is survivable. Rule: under 2% of your monthly target, or under $100/day if it is your first time.
• Kill rule defined. Rule of thumb: pause if CPA exceeds 2x target after 50 clicks.
• Scale rule defined. Rule of thumb: raise budget 30% if CPA stays under target across 50 clicks.

Section 5: Benchmarks

You need something to measure against. Here is the industry data, from Similarweb's Q1 2026 worldwide ad performance dataset.

My personal bar is higher than the average. In my campaigns, winners run above 1.5% CTR and losers stay below 1%. I pause anything under 1% and scale anything above 1.5%.

One warning: the widely cited "1.3% industry CTR" number floating around is probably skewed by survivor bias from a handful of early advertisers. Do not benchmark against it. Use the table above, then set your own bar from your own results as you gather data.

Section 6: Your first seven days

The checklist does not end at launch. The first week is where you separate signal from noise.

Day 1. Verify the pixel fires only on the conversion event. Do this before you spend a dollar.

Day 3. Compare delivery across your hint variants. Kill the narrow one. Double the budget on whichever is actually serving impressions.

Day 7. Read your CTR against your own results so far. Pause anything below 1%. Scale anything above 1.5%. Iterate creative on whatever sits in between.

After day 7, you are out of setup mode and into the longer rhythm: let campaigns run two to four weeks to stabilise, read the data on a weekly cadence, and make scale-or-kill decisions on real numbers. The full campaign-management approach is in the ChatGPT Ads campaign guide.

Naming conventions

If you are running more than three campaigns, name them now. You cannot rename mid-flight without losing your reporting, and the platform does not enforce any structure, so your future self has to.

The patterns I use:

Campaign: Brand_Objective_Geo_LaunchDate — e.g. Acme_Clicks_US_2026-05-08. The objective matches the buying mode (Clicks for CPC, Reach for CPM, which the OpenAI UI sometimes calls "Views"). The full date prevents collisions when you launch several campaigns in a week.

Ad group: Audience_Theme_Stage — e.g. Agency_BroadCAC_TOFU. One ad group equals one hint equals one match logic. Funnel stage is TOFU, MOFU, or BOFU.

Ad: Audience_Angle_Version — e.g. Agency_FOMO_v1. The angle is the creative concept (FOMO, SocialProof, Discount).

The PDF includes the matching UTM template so the whole hierarchy lines up in GA4. OpenAI explicitly supports custom UTM parameters on destination URLs, so you can roll up by audience, by angle, or by campaign without leaving Google Analytics.

Where this checklist comes from

I am not writing this from the outside. I am spending $10,000 on ChatGPT Ads right now, tracking every dollar, and publishing what I find. The findings in this checklist, the broken default pixel, the 50x impression gap between near-identical hints, the CTR bars, all come from real campaigns, not theory.

If you want the running data behind these rules, the ChatGPT Ads ROAS study tracks the actual spend and returns as they land. And if you want to follow along as I learn, the FutureProof community is where I post the weekly updates first.

When you are ready to run ChatGPT Ads at scale, Focal is the platform we are building for it: connect your Google or Meta accounts, see what competitors are running, and ship campaigns built for the new auction.

Frequently asked questions

What should I check before launching a ChatGPT Ads campaign?

Six things, in order: that your conversion pixel fires only on the conversion event (not every page), that your context hint is validated for reach before launch, that you have at least four distinct ad angles live, that your landing page works on both mobile and desktop and is not your homepage, that you have kill and scale rules defined, and that you know your CTR benchmark. The full checklist is above and in the downloadable PDF.

What is a good CTR for ChatGPT Ads?

Per Similarweb's Q1 2026 data, the average is 0.68%, top quartile is 1.0%, and top 10% is 1.57%. Aim for the top quartile at minimum. My personal bar is higher: winners above 1.5%, losers below 1%. Ignore the widely cited "1.3% industry" figure, which is likely skewed by survivor bias.

Why is my ChatGPT ad not delivering impressions?

Almost always a context hint that is too narrow. The rule of thumb: under 100 impressions in 48 hours means the hint is too tight, so broaden it. Validate reach before launch with the Focal Reach Calculator so you catch this before you spend.

How many ad variations should I launch?

At least four per hint, each a genuinely different angle (FOMO, social proof, founder story), not four wordings of the same line. OpenAI's auction rewards a high and diverse volume of ads because it can match them across more conversations.

What is the biggest ChatGPT Ads launch mistake?

The broken default pixel. The copy-paste tracking code from the Ads Manager UI fires a conversion on every page load, so your dashboard shows conversions you never got and you scale a losing campaign. Verify in DevTools that the conversion event fires only on conversion before you spend a dollar.

Should I send ChatGPT ad traffic to my homepage?

No. OpenAI's own guidance is to use product, collection, or content pages, not the homepage. Homepage traffic converts far worse because it makes the visitor hunt for the thing the ad promised. Build a dedicated, intent-matched landing page.

My question is not listed here.

Ask in the FutureProof community or email support@tryfocal.com.