We spent a month on ChatGPT Ads. Our best ads brought in nobody.

There is no shortage of walkthroughs for setting up ChatGPT Ads. Account creation, verification, context hints, bulk upload, the mistakes that waste budget. Search any of it and you will find a decent explanation inside two results. What I could not find anywhere was somebody saying what the channel actually returned once they had paid for it.

We launched on 17 September and spent real money. This is the first month: what we built, what it returned, and what I would want to know before deciding whether to do the same. Not a guide. The part the guides cannot give you, which is whether it is worth it.

What we actually built in the ChatGPT Ads Manager

The setup is simpler than Google Ads, which helps at the start and hurts later. Fewer things to configure means you can be live in an afternoon. It also means that when a campaign underperforms there is very little to pull on. No keyword lists to prune, no match types, no negative terms. Your copy and your context hints are close to the whole of it, so a campaign that is not working is usually a copy problem and there is nowhere else to look.

One campaign. Objective set to clicks. Thirteen countries: US, UAE, Australia, Canada, India, Italy, Japan, South Korea, Malaysia, New Zealand, Philippines, Singapore and Vietnam. We named it for India and US because that is where we expected the volume, which turned out to be roughly right.

Two ad groups. One broad, aimed at B2B SaaS growth generally. One narrow, aimed at AI visibility specifically.

Seven ads. Four in the broad group, three in the narrow one. Each is a headline and a line of body copy. No images, no video. The ad appears inside the conversation, so it reads more like a suggestion than a banner.

Three levels, the same shape as Google and Meta. The campaign sets objective, budget, dates and countries. The ad group holds your context hints and your max bid. The ads sit under that.

India + US | SaaS growth | ChatGPT        campaign, objective: clicks
│
├── SaaS growth, broad                    ad group
│   ├── SG | Named by AI          19,675 impressions · 428 clicks · 3 leads
│   ├── SG | Agency for SaaS      11,683 impressions · 197 clicks · 2 leads
│   ├── SG | Leads dropped         7,810 impressions · 110 clicks · 0
│   └── SG | Two founders          added later, nothing in this window
│
└── Am I visible in AI                    ad group
    ├── AV | Named or not          4,298 impressions · 115 clicks · 0
    ├── Tools say you are invisible 3,433 impressions ·  94 clicks · 0
    └── AV | Shortlist leak         1,670 impressions ·  34 clicks · 0

Every ad served. The spread is the thing to look at: the top two took most of the delivery and produced all five leads, and the bottom four produced none between them despite 353 clicks.

A dedicated landing page. We did not send traffic to the homepage. We built a page on a separate subdomain, so the traffic was clean and we could tell what came from where without untangling it from everything else.

A budget ceiling. About $12.50 a day (₹1,200). Billing runs on a threshold rather than a monthly invoice: you are charged when you hit the threshold or when the period ends, whichever comes first.

1. Context hints are the whole game

If you take one thing from this post, take this. Context hints are how ChatGPT Ads decide where your ad belongs, and almost nothing else is adjustable. Get them wrong and no amount of budget rescues the campaign.

They are not keywords. The field even says so: describe the conversations, topics or keywords where your product may be relevant, and these guide matching rather than acting as exact-match rules. You write sentences describing a person and what they are asking. The model does the rest.

I am not going to publish ours, for reasons that will be obvious to anyone who sells what we sell. The shape is the useful part anyway, and the shape is easy to hand over. Every line names a person and the situation they are in, roughly like this:

[the role you sell to] asking how to get more [the outcome they want]. Asking why [the thing going wrong] started happening this year. Asking how to do [your category of work] with a small team. Comparing [category] tools for a [type of company]. Writing a [plan or document] for a [type of company]. Asking how competitors manage [the thing they cannot].

Notice the shape. Every line is a person plus a situation, not a term. “Asking why website traffic or organic leads dropped this year” is a moment, and it is a moment where a relevant ad makes sense. “SEO agency” on its own would be a keyword, and it would match the wrong conversations entirely: students, job hunters, people researching the industry.

1.1 How to write context hints for your B2B SaaS

The hints that work describe the conversation you want to be in. So start from the conversation, not from your product.

If you sell something people search for by name, your hints should cover the comparison conversation rather than the brand one. Someone who already knows your category is asking which option fits, not what the category is.

If you sell something people do not know exists, your hints have to describe the symptom. Nobody asks for a product they have not heard of. They describe the problem and hope the model knows what to suggest. Ours do a lot of this: “asking why organic leads dropped this year” is a symptom, not a solution.

If you sell to a specific role, name the role in the hint. “B2B SaaS founder or marketer asking” is doing real work. It keeps us out of conversations with agencies, students and people in unrelated industries asking the same surface question.

If your buying moment is seasonal or triggered, describe the trigger. Planning a budget, writing a go-to-market plan, preparing for a fundraise. These are moments, and moments are what the system matches on.

A warning from our own results. Our second ad group used narrower hints aimed at AI visibility specifically. It served fine and collected 243 clicks, and not one of them became a lead. The broader group, pointed at the symptom rather than the solution, produced all five. Narrow hints found us people curious about the topic. Broad hints found us people with the problem.

2. Bidding: what we capped, and what we actually paid

We set a manual maximum CPC of about $4.20 (₹400). Our actual cost per click came in between $0.47 and $0.69 (₹45 to ₹66).

That gap is worth understanding. The max bid is a ceiling for winning auctions, not what you pay. Setting it high enough to be competitive did not mean spending it. If we had set fifty cents to feel safe, we would have lost most of the auctions and concluded there was no inventory.

For context on whether that is cheap: B2B SaaS paid search in the US typically runs five to twenty-five dollars a click. We were paying under a dollar.

I would not read too much into that yet. It is a new auction with few advertisers in our category, and prices in a young ad market go one way. But right now the gap is roughly an order of magnitude, and that is the single strongest argument for testing this channel before everyone else does.

Rupee figures converted at roughly ₹95 to the dollar, the rate in late October 2026. Our account bills in INR, so the rupee numbers are the exact ones and the dollar figures are the approximation.

That is the whole build. Looking back at it, the striking part is how little of it is configuration and how much is writing. Two paragraphs of hints and seven short ads. Everything else took an afternoon.

3. Three things about ChatGPT Ads we only found out by spending

The mechanics are documented, and the walkthroughs published since the platform opened cover them well enough. These three are not in any of them, because they only show up once your card is attached.

Your objective is permanent. You choose it when you create the campaign and you cannot change it afterwards. We picked clicks. If we had wanted to optimise for conversions instead, that would have meant starting again.

Verification takes a human. Business details go through Persona, and your account name and logo have to be set exactly as you want them to appear beside the ad before anything will serve. That is not a five-minute job.

Billing is postpay on a threshold. This took us a while to work out, so here it is slowly.

You are not billed on a date. You are billed when your unpaid spend reaches a set amount. That amount is your threshold, and ours is about $122 (₹11,600). Yours will be different. It is set per account, you do not choose it, and the interface does not explain how it is decided. Take the number below as an illustration of how the mechanism works, not as a figure to plan against.

So the cycle runs like this. Your ads spend, and the unpaid balance climbs. The moment it reaches your threshold, your card is charged that amount and the balance resets to zero. Then it starts climbing again. If you never reach the threshold within the billing period, you are charged whatever has accumulated at the end of it instead.

Two consequences worth planning for.

Your charge dates move with your spend. Spend faster and the threshold arrives sooner, so you might be charged twice in a fortnight. Spend slower and three weeks pass with nothing. There is no fixed billing day to put in a calendar.

The money leaves after the ads have run, not before. You cannot prepay and watch it drain. If someone in your company approves card charges, tell them in advance, because the amount and the timing will both look irregular.

What did a month of ChatGPT Ads return?

Metric	First month
Impressions	48,569
Reach	42,038
Clicks	978
Leads	5
Customers	2

Five leads from 978 clicks is a 0.51% click-to-lead rate. Judged against a paid search campaign, that is poor.

Two customers from five leads is 40%. Judged against anything, that is not.

If we had reviewed this at four weeks looking only at lead count, we would have switched it off. The volume says failure. The close rate says this is the best-converting channel we run.

1. The ads with the best click rates produced nothing

Seven ads ran. Two produced all five leads. The other four produced none between them, despite collecting 353 clicks.

That much I expected. What I did not expect was which two.

Ad	Impressions	Clicks	CTR	Leads
Competitors get named by AI	19,675	428	2.18%	3
A growth agency built only for B2B SaaS	11,683	197	1.69%	2
Tools say you are invisible in AI	3,433	94	2.74%	0
Your buyers ask ChatGPT first	4,298	115	2.68%	0
Not in the AI shortlist?	1,670	34	2.04%	0
Fewer inbound leads this year?	7,810	110	1.41%	0

Look at the click-through column. The two best-performing ads in the whole campaign, at 2.74% and 2.68%, produced nothing at all. The ad that brought in two customers had the second-worst click rate on the list.

If we had optimised on click-through, which is the obvious move and what every dashboard nudges you toward, we would have killed the ad responsible for 40% of our customers and put budget behind two that produced none.

2. What the two converting ads had in common

Here is the ad copy from both, exactly as it ran. Headline first, then the line underneath it.

Competitors get named by AI. You do not. Yet.
Find out which three companies ChatGPT recommends in your category and what it takes to be one.

A growth agency built only for B2B SaaS.
SEO, AI visibility, sales, retention. Two founders, no account managers. Three clients a quarter.

One is a hook. One is a plain description of the company. Stylistically they have almost nothing in common.

What they share is that each tells you exactly what you get if you click. A specific answer in the first case, a specific kind of company in the second.

And here is the ad copy that earned the best click rate in the campaign and produced nothing:

Tools say you are invisible in AI. We fix it.
SEO, AEO and GEO for B2B SaaS. No dashboard, no retainer. A fix, handed over working.

I still think that is the sharpest line of the seven, and the click rate says readers agreed. It just did not bring anybody who wanted to buy. My best guess after a month is that it is interesting rather than useful. It makes you curious. It does not describe your situation back to you.

So: write ads that name the reader’s situation, and judge them on what closes rather than what gets clicked. For us those two measures pointed in opposite directions all month.

Will ChatGPT Ads work for you?

This is the part where our numbers stop being useful to you, because whether this channel works at all turns on something neither of us controls. Check it before you spend anything. We did not, and we were lucky.

Ads do not appear in every ChatGPT conversation. They appear on exactly two plans: Free, and Go, which is the eight-dollar tier. Plus, Pro, Business, Enterprise and Education accounts never show ads. Nor does any account belonging to someone under 18. Free users can also turn ads off entirely in exchange for lower daily message limits. Our buyers are B2B SaaS founders and operators who live in ChatGPT and are often on configurations where ads show. That is lucky, and it is specific to us.

Read that again, because it is a harder filter than it looks. Your buyer has to be on the free plan or the cheapest paid one. Anyone who has expensed a Plus subscription, which describes most people who use ChatGPT seriously for work, cannot see your ad at all.

And being on an eligible plan is not sufficient. Financial Times reporting in March 2026 put the share of eligible users who actually see an ad in any given session at roughly 20%.

Geography narrows it again, though less than it did. Ads ran in nine countries in August 2026 and reached 89 by 21 October, so this part is moving quickly in your favour.

If your buyer is an enterprise security officer on a locked-down deployment, or a developer working through an API, or anyone whose organisation has turned this off, they will never see your ad no matter how good your copy is.

So before you budget for this, answer three things honestly:

Does your buyer use ChatGPT to think through this kind of problem? Not “do they use AI”. Do they open a conversation and describe the problem your product solves, in their own words? Our buyers do. “Why are our inbound leads down” is a thing a founder types at eleven at night.

Is your category one a model can name? The winning ad worked because ChatGPT genuinely does recommend vendors in our category when asked. If your category is too new, too niche, or not something people ask for recommendations in, the premise does not hold.

Can you survive a month of spend producing nothing? At three weeks we had almost nothing. If about $300 of spend with no result would hurt, wait.

So did ChatGPT Ads deliver a return?

For us, yes, and not marginally.

Two customers from roughly a month of spend at a ceiling of about $12.50 a day. Our engagements are twelve weeks and priced accordingly. Against our deal sizes, one customer from ChatGPT ads covered the channel several times over. The second made it the most efficient line in our acquisition budget this quarter.

That is the honest answer for us. It is not an answer for you, and I want to be careful about the difference, because the thing I keep seeing in posts about new channels is a single company’s result presented as a benchmark.

Here is what I would actually look at.

Your contract value against your cost per click. Ours ran between $0.47 and $0.69. At five leads per thousand clicks and a 40% close rate, our cost per customer landed somewhere a twelve-week engagement absorbs easily. Run the same arithmetic on a $5,000-a-year product and it falls apart immediately. This channel suits higher contract values, and the lower your price point the harder the maths gets.

Your close rate, not your lead count. If your leads typically close at 2%, five leads is nothing and this channel will disappoint you. Our 40% is a function of how qualified people are when they arrive, which is itself a function of the buyer having already articulated the problem to a model before they ever saw us. If your sales process cannot convert a well-informed inbound lead at a high rate, the channel’s main advantage is wasted.

Whether you can wait. Leads did not arrive evenly. They came in clusters, and in the eight days after this reporting window we spent on another 130 clicks and got nothing. If you need weekly predictability, this is the wrong channel. It is a channel that produces a small number of unusually good conversations at irregular intervals.

What I would do differently on the next campaign

Write more ads, not better ones. We wrote seven and one worked. Writing seven more and finding another winner is a better use of a week than polishing the four that did not.

Set the volume expectation before starting. We went in with search-ad instincts and spent two weeks reading the numbers wrongly. Knowing this is low-volume and high-intent would have saved that.

Ask every lead where they came from. We did not take a clean baseline of our inbound before launch, which made the early weeks harder to read than they needed to be. Ask on the form. Ask again on the call.

Expect it to look strange in analytics. It arrives as chatgpt / cpc, and for a stretch it was the second-largest source on our site behind direct. If you are not expecting it, it reads as a tracking error.

Would I spend on ChatGPT Ads again?

Yes, with the budget capped where it is.

Not because the volume is impressive. It is not. Because five conversations with people who had already worked out their problem before they found us produced two clients, and because one ad taught me more about what our buyers actually respond to than any amount of positioning work had.

The second thing may turn out to be worth more than the first.

If you have read this far and still cannot tell whether your buyers are on the other side of that ad, that is the right question to be stuck on, and it is a short conversation rather than a long one. We run ChatGPT Ads for clients now, but the first thing worth doing is working out whether you should. Ask us and we will tell you straight, including when the answer is no.

Common questions