ChatGPT Ads Campaign: How to Build One That Actually Converts

Henry Purchase

Co-Founder

A ChatGPT Ads campaign is not a Google campaign with a different logo. The structure looks familiar, but the parts that decide whether you make money or burn it are completely different.

I have built dozens of these now, spending $10,000 of my own money to learn what works while OpenAI ships changes to the platform every week. This article is the full breakdown of how to structure, build, and launch a campaign that converts, plus the decisions most people get wrong before they even hit launch.

If you have not set up your account yet, start with how to create a ChatGPT Ads account. If your tracking is not wired up, fix that first with the ChatGPT Ads tracking guide, because launching a campaign without tracking means you cannot tell a winner from a loser. This article assumes both are done.

The three-layer structure (and what lives where)

Every ChatGPT Ads campaign has three layers. Knowing exactly what gets set at each level is the difference between a clean account and a mess you cannot read in a month.

Campaign sits at the top. This is where you set your objective, budget, start and end dates, and country targeting. One campaign per business goal.

Ad group sits inside the campaign. This is where your bid and your context hints (targeting) live, along with the landing page URL for that group. One ad group per audience or intent cluster.

Ad sits inside the ad group. This is your creative: headline, description, image, and destination URL. Multiple ads per group so you can test.

So the hierarchy reads: one campaign with a goal and a budget, containing several ad groups each targeting a different audience, each containing several ad creatives you are testing against each other.

A quick correction while we are here. Some guides call the middle layer an "ad set." That is Meta's term. On ChatGPT Ads it is an ad group, the same as Google. And some older guides list objectives like "awareness, consideration, conversion." Those do not exist on this platform. There are exactly two objectives, which we will get to now.

Step 1: Pick your objective (there are only two)

When you create a campaign, OpenAI gives you two objectives:

That is it. No "conversions" objective yet, though OpenAI has confirmed more bidding models are coming, and cost-per-conversion is the obvious next one once their pixel data matures.

Which should you pick? If your goal is measurable performance, like leads or sales, start with Clicks. You pay per click, which aligns spend with actual interest, and it is the natural fit if you came from Google Ads. If your goal is pure brand exposure inside a category, Reach makes sense, and it will feel familiar if you came from Meta.

My honest recommendation for a first campaign: run Clicks. It forces discipline because you are paying for engagement, not eyeballs, and it gives you cleaner data to read after two weeks. Once you understand your numbers, test Reach as a second campaign and compare.

How the auction actually decides who wins

This is the single most important thing to understand about ChatGPT Ads, and most setup guides skip it.

OpenAI runs a relevance-weighted, second-price auction. Your bid is one input. The other input is how relevant your ad and landing page are to the conversation the user is having. The winner is the highest combined score, not the highest bid.

What this means in practice: a $4 bid with tight, relevant targeting and on-message creative can beat a $7 bid with generic targeting. You do not win this platform by outspending people. You win it by being more relevant than they are. That is good news if you are a smaller advertiser, because relevance is something you control with effort, not budget.

Step 2: Name the campaign so you can find it later

Before you touch any settings, name the campaign properly. This sounds boring. It is the thing that saves you when you have 30 campaigns running across multiple clients and cannot remember which is which.

Use this structure:

[Brand] - [Objective] - [Geography] - [Launch date]

Example: Focal - Lead gen - US - 2026-05-14

The ChatGPT dashboard has weak filtering. You cannot slice campaigns by attributes the way you can in Google Ads. A consistent naming convention is your only defence against chaos. If you run an agency, swap the brand for the client name and you can scan your whole book at a glance.

Step 3: Set your budget and dates

Set a budget at the campaign level. There is no platform minimum spend since OpenAI dropped the $50,000 floor on 5 May 2026, but there is a practical floor: daily budgets below roughly $50 generate too little volume to learn from.

For a first campaign, $100 to $300 per day for the first two weeks is a reasonable range for most categories. That gives the platform enough volume to start finding your audience while keeping your test affordable. I am running larger numbers because I am stress-testing across multiple campaigns, but for a first test you do not need to.

Pick a start date. Decide whether you want an end date or to run open-ended. For a test, I prefer open-ended with a hard budget cap, so I control spend through the budget rather than an arbitrary end date.

Step 4: Build your ad groups around intent

This is where the campaign starts to take shape, and where the strategy lives.

An ad group is a cluster of intent. You build one ad group per audience or per type of conversation you want to show up in. Within each ad group you set the bid, write the context hints, and pick the landing page.

Here is how I think about ad group structure for a typical campaign. Say you run a lead-gen business. You might build three ad groups:

• Ad group 1: Broad intent. People asking general questions in your category. Lower bid, wider context hints, sends to a general landing page.
• Ad group 2: High intent. People asking buying-stage questions ("best X for Y", "should I hire someone for Z"). Higher bid, tighter context hints, sends to a conversion-focused landing page.
• Ad group 3: Competitor or comparison intent. People comparing options. Medium bid, comparison-focused context hints, sends to a page that positions you against alternatives.

Name each ad group [Audience] - [Theme]. Example: Business owners - high intent.

Splitting by intent like this is what lets you bid more where the money is and less where it is not. If you dump everything into one ad group, you cannot do that, and you end up overpaying for low-intent clicks and underbidding on the ones that convert.

Setting your bid

Set your max CPC bid at the ad group level. $3 to $5 is OpenAI's recommended starting range and the current sweet spot for most categories.

Premium categories bid higher. Finance, legal, and SaaS often sit at $8 to $18 because the customer is worth more. E-commerce and travel sit at $3 to $6. Your category will settle into its own range after a week of data.

OpenAI shows a "bid competitiveness" indicator while you set this. In my testing it barely moves no matter what number you enter, so treat it as a rough signal, not a guide. The real signal is your delivery and cost data after launch.

Writing context hints that convert

Context hints are your targeting. They are the most important field in the entire campaign, and they are not keywords.

The field asks you to describe the conversations, topics, and queries where your product is relevant. OpenAI's helper text is explicit: these hints guide matching but are not exact-match rules. The platform matches on the meaning of a conversation, not on literal keyword strings.

Most people paste their Google Ads keyword list in here and wonder why delivery is poor. The platform is built for natural language. Write it like you are describing your ideal customer to a colleague.

The three-part structure that works for my campaigns:

1. Describe who you want to reach. Be specific. Not "small businesses." Try: "Marketing agency owners in the US running 5 to 10 clients on monthly retainers who want to add a new service line."

2. List the things they search. Pull these from your Google Ads or Meta data if you have it. If not, think about what your buyer actually types. For an agency that might be: "how to run ChatGPT Ads, ChatGPT advertising agency, what does ChatGPT advertising cost, AI advertising for clients."

3. Describe the conversations they have. This is the part nobody else covers and it is where ChatGPT Ads gets powerful. Your buyer does not search "CRM software." They ask ChatGPT "what CRM should a 5-person agency use that won't cost a fortune." Describe those full-conversation moments.

Combine all three into one paragraph per ad group, tuned to that group's intent level. The broad ad group gets wider hints. The high-intent ad group gets tighter, buying-stage hints.

This is exactly the layer the Focal Chrome extension is built to help with. It pulls the real keywords and questions your target customers are searching in ChatGPT and Google directly into the context hints field, so you are writing targeting based on data instead of guesses. Join the waitlist if you want it when it ships.

Step 5: Pick the right landing page (this is where most campaigns die)

Each ad group needs a landing page URL. HTTPS only.

The biggest mistake I see, by a wide margin, is sending every ad to the homepage.

Think about what just happened. Someone was deep in a conversation with ChatGPT about a specific problem. Your ad showed up as a relevant answer. They clicked because it matched their need. Then they land on your homepage, which is about everything your business does, and they have to hunt for the thing they actually wanted.

Conversion rate drops 60 to 80% when you send ChatGPT traffic to a homepage instead of a matched landing page, based on what I am seeing across our campaigns. And remember most ChatGPT users are on mobile, so the page has to load fast and read clean on a phone.

Build a dedicated landing page per ad group where you can. One offer, one message, one clear action, matched to the intent of that ad group. If you can only manage one landing page for the whole campaign to start, that is fine, but it is the first thing to improve once you have data.

Step 6: Write three ads per group and let them fight

Inside each ad group, build at least three ads. Five is better. With one ad you are not testing anything, you are just guessing and hoping.

The ad has four parts, and the space is tight, so every word works hard:

Headline — under 30 characters. Past that it truncates on mobile. Lead with a number, a specific outcome, or clear category positioning. Generic headlines die in this format.

Description — 15 to 19 words, two short sentences. Include a number where you can. Make the offer obvious. Based on what we are seeing perform, the winning pattern is a clear benefit plus a clear next step like "get the free guide" or "book a call."

Image — required. Square renders best inside the chat interface. Make it catch the eye in a feed of text.

Destination URL — add UTM parameters here so you can read performance in GA4: utm_source=chatgpt&utm_medium=cpc&utm_campaign=[campaign]&utm_content=[ad variant].

Name each ad [Audience] - [Angle] - [Format]. Example: Agency owners - free guide - text+image.

Test different angles across your three ads. One direct-offer ad, one problem-solution ad, one social-proof or case-study ad. Let the data tell you which angle your market responds to, then make more like the winner.

Why your offer matters more here

ChatGPT Ads only show to Free and Go tier users. These are not your highest-intent enterprise buyers, those people are on Plus and Pro where there are no ads. The Free and Go audience responds to value first, commitment second.

Lead magnets, free trials, free guides, free discovery calls all work. Hard sells asking for a credit card on the first click do not. Build your offer for someone willing to give you an email address in exchange for something useful, then sell on the follow-up.

Step 7: Review, launch, and then leave it alone

Hit Review, check everything, and launch. Your campaign goes into the auction.

After approval, ads usually start delivering within 12 to 24 hours. There is a 7-hour lag between ads delivering and data showing in the dashboard, so a quiet first morning is normal, not a problem.

If an ad shows "Not serving," hover over the status to see why. Most issues are fixable by editing and resubmitting. Some ads get rejected for policy reasons, so check OpenAI's ad policies if you are in a borderline category.

Now the hard part: leave it alone for at least seven days.

The most common way people kill a good campaign is touching it too early. ChatGPT Ads need two to four weeks to stabilise while the platform learns which conversations and users convert for you. Three days of data is noise, not signal. Resist the urge to change bids, swap creative, or pause ad groups in the first week. Read your data on day 14, read it again on day 21, then make decisions.

How to read the results and decide what to do

Once you have two weeks of data, open the campaign and drill down through ad groups to individual ads. At each level you can see impressions, clicks, spend, CTR, average CPC, and conversions if your tracking is set up.

Here is the decision framework I use:

At the ad group level, look at cost per conversion. The ad groups with the lowest cost per conversion get more budget. The ones with high cost per conversion and enough data to trust get paused or reworked.

At the ad level, look at CTR and conversion rate together. A high CTR with low conversion means the ad promises something the landing page does not deliver. A low CTR means the creative is not earning attention. Kill the bottom performers, make more variations of the winners.

At the context hint level, if an ad group is not delivering at all, your hints are probably too narrow or your bid is too low. Widen the hints or raise the bid and give it another week.

The honest truth about reading ChatGPT Ads data: the dashboard is the weakest part of the platform. You cannot pivot cleanly, cannot compare ad groups side by side, cannot pull the granular breakdowns you are used to in Google or Meta. The workaround most operators use is exporting a CSV and pasting it into Claude or ChatGPT to find the patterns.

That manual workaround is exactly what we are turning into a product with Focal. You connect your ChatGPT Ads account and ask questions like "which ad groups are converting and where am I wasting spend" from inside Claude or ChatGPT, and get a structured answer back instead of squinting at a spreadsheet. The waitlist is here.

A realistic first-campaign timeline

Putting it all together, here is what a sane first campaign looks like:

Week 1: Build the campaign with two or three ad groups, three ads each. Launch. Do not touch it.

Week 2: First real read. Note which ad groups are delivering and which ads have the best CTR. Still resist major changes.

Week 3: Make your first decisions. Pause the clear losers. Shift budget toward the ad groups with the lowest cost per conversion.

Week 4: Refresh creative. Make new variations of your winning ads. Test a new angle. Consider a second campaign with the other objective to compare CPM against CPC.

By the end of month one you will know whether ChatGPT Ads work for your business, which is more than most of your competitors can say.

Frequently asked questions

How many ad groups should a ChatGPT Ads campaign have?

Two to three for a first campaign, split by intent level (broad, high intent, comparison). More than that and you spread your budget too thin to gather meaningful data per group. Add ad groups as you scale and learn.

What objective should I choose for my first ChatGPT Ads campaign?

Clicks (CPC) if your goal is measurable performance like leads or sales. Reach (CPM) if your goal is brand awareness. There are only these two objectives today. Start with Clicks for cleaner performance data.

How much should I budget for a first ChatGPT Ads campaign?

$100 to $300 per day for the first two weeks is a reasonable test for most categories. There is no platform minimum since 5 May 2026, but daily budgets under roughly $50 do not generate enough volume to learn from.

How long before I know if my ChatGPT Ads campaign is working?

Two to four weeks. The platform needs that long to stabilise and learn which conversations convert for you. Do not make major changes in the first seven days, and do not make scaling or killing decisions before day 14.

Why are my ChatGPT ads not serving?

Hover over the "Not serving" status to see the reason. Common causes are a bid that is too low to win the auction, context hints that are too narrow, or an ad rejected on policy grounds. Most are fixable by editing and resubmitting.

Should every ad group have its own landing page?

Ideally yes. Matching the landing page to the ad group's intent is one of the highest-leverage things you can do for conversion rate. If you can only build one page to start, that is workable, but a dedicated page per intent cluster converts far better than sending everyone to a homepage.

Can I bulk-create ChatGPT Ads campaigns?

Yes. The Ads Manager supports bulk upload via a CSV schema template under Create > Upload bulk. This is useful for agencies launching multiple campaigns or clients at once. You can also export your Google Ads campaigns and reformat them for import.

My question is not listed here.

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