Every guide to ChatGPT Ads pricing quotes the same figure: a $3 to $5 maximum bid. That is what OpenAI recommends you set as a ceiling, not what you pay. Across the accounts we run, the blended cost per click is $2.42, and the spread between the cheapest vertical and the dearest is roughly nine times. Read the method before the numbers — a benchmark without its method is a rumour.
How this was compiled
The figures come from Bytown's own advertiser reporting across the ChatGPT Ads accounts we manage. They were hand-compiled from that reporting, not exported through a platform integration — no account has been connected through Bytown's ChatGPT Ads connector, so nothing here synced automatically. Somebody read the accounts and typed the numbers into a sheet.
Every rate is computed from summed raw counts: total spend divided by total clicks, total conversions divided by total clicks, and so on. None of them is an average of per-account rates. That distinction matters more than it sounds. Averaging the rate of a small test account against the rate of one spending five times as much gives the two equal weight, and produces a number that describes no advertiser at all.
Aggregation is by vertical, with no client names and no account-level figures. Where a vertical has few accounts, the account count sits in the table next to it rather than being hidden — you can see exactly how thin any row is and discount it accordingly. Nothing is suppressed: every vertical we have data for is in the table, including the ones with two accounts behind them.
On the restricted categories. Several verticals here — dental, elective healthcare, med spa, legal, B2B fintech — sit in categories OpenAI reviews rather than opens by default. Those accounts run on a mix of the two routes the policy allows, account by account: some as approved advertisers, some advertising an offer that is not in the regulated class to begin with. We are not attributing a route to any particular vertical, because doing so would identify clients. If you are in one of these categories, read the two routes through before reading the numbers — which route applies to you changes what these figures mean for your own account.
What this is not: an industry benchmark. It is one agency's book of business, and the verticals in it are the verticals our clients happen to work in. A vertical backed by two accounts tells you what those two accounts paid. It does not tell you what the category costs.
| Vertical | Accts | Acct-mo | Spend | CPC | CPM | CTR | CVR | CPA |
|---|---|---|---|---|---|---|---|---|
| B2B fintech | 3 | 8 | $5,621 | $5.40 | $47.64 | 0.88% | 1.73% | $312 |
| Automotive | 2 | 4 | $5,222 | $1.61 | $58.56 | 3.65% | 3.63% | $44 |
| Healthcare (elective procedures) | 3 | 7 | $4,486 | $4.99 | $187.86 | 3.76% | 4.12% | $121 |
| Education | 3 | 8 | $4,272 | $3.28 | $106.93 | 3.26% | 4.38% | $75 |
| B2B SaaS | 2 | 4 | $4,084 | $4.31 | $45.62 | 1.06% | 2.32% | $186 |
| Non-profit | 3 | 6 | $4,007 | $0.74 | $22.56 | 3.06% | 6.09% | $12 |
| Events | 3 | 6 | $3,915 | $2.61 | $104.86 | 4.02% | 4.59% | $57 |
| Engineering | 3 | 8 | $3,843 | $6.65 | $101.69 | 1.53% | 1.90% | $349 |
| HVAC | 5 | 9 | $3,712 | $1.84 | $104.04 | 5.65% | 6.45% | $29 |
| Architecture | 3 | 9 | $3,281 | $5.72 | $100.89 | 1.76% | 2.26% | $252 |
| Dental | 4 | 12 | $2,748 | $2.50 | $117.46 | 4.71% | 5.36% | $47 |
| Legal | 3 | 8 | $2,674 | $4.27 | $130.65 | 3.06% | 3.67% | $116 |
| Landscaping | 3 | 8 | $1,879 | $1.37 | $83.70 | 6.12% | 8.52% | $16 |
| Med spa | 2 | 5 | $1,350 | $2.85 | $114.15 | 4.00% | 4.02% | $71 |
| All verticals | 42 | 102 | $51,095 | $2.42 | $67.26 | 2.78% | 4.85% | $50 |
What the numbers say
The bid is not the price. OpenAI's recommended $3 to $5 maximum bid is a ceiling you set; what you actually pay is decided by a relevance-weighted second-price auction, so it usually settles below the ceiling. Our blended cost per click lands under the bottom of that recommended range. Anyone budgeting from the recommended bid is over-planning their click costs and under-planning everything downstream.
The blended figure is the least useful number on this page. A nine-fold spread between the cheapest and dearest vertical means the average describes almost nobody. What a click costs you depends far more on what you sell than on how well you run the account. Read your own row, not the total line.
The click price understates how different these verticals are. The slope chart ranks the same verticals twice, and the ordering barely changes — expensive clicks stay expensive. What changes is the scale. Cost per click spans roughly nine times from cheapest to dearest; cost per acquisition spans nearly twenty-nine, from $12 to $349. The difference between a cheap vertical and an expensive one is far larger after you divide by conversions than the click price alone suggests, because low click-through rates and low conversion rates tend to arrive together. Budget from the conversion column, not the click column.
Impression pricing moves in the opposite direction to click pricing. The verticals with the highest CPM are frequently the ones with the lowest cost per click, because a high click-through rate spreads the same impression cost over more clicks. If you are buying on an impression basis, the vertical ordering you should expect is close to inverted.
Reading the month-over-month line honestly
The trend chart plots two series, and the difference between them is the methodological point.
Accounts join mid-window. An account that started in June contributes to June and July but not to March, so a month-over-month line drawn across all accounts partly measures who was in the sample that month rather than what prices did. That is the faint line.
The solid line is a fixed cohort — only the accounts present in every month of the window. It is a smaller sample and a slower-moving one, and it is the only line here that supports a sentence beginning "costs rose" or "costs fell".
This edition is a clean example of why that matters. Read across all accounts, cost per click falls from $2.68 to $2.35 over the window — a twelve per cent decline you could reasonably write up as the auction getting cheaper. Read across the cohort that was present the whole time, it goes from $2.68 to $2.64, which is to say it did not move. The apparent decline is almost entirely newer, cheaper-vertical accounts entering the sample. Nothing got cheaper for anyone already running.
We publish both lines because publishing only the flattering one is how benchmark diaries stop being worth reading.
Against Google and Meta in the same verticals
The comparison people actually want is against what they already run, so here is the honest version rather than a table that implies more precision than exists.
Across local businesses — clinics, automotive, retail, high-billable professional services — I have generally seen cost per lead land somewhere between $30 and $150 on established search and social channels. That is a field observation from years of running those accounts, not an industry average, and the range is wide because the businesses are different.
Set that against the cost-per-conversion column here and the read is straightforward: for several consumer-facing local verticals, ChatGPT Ads is currently landing inside or below that band, and for business-to-business verticals it is landing well above it. That gap is what you would expect from a channel where the auction is still thin in some categories and crowded in none. It is also the kind of gap that closes. The reason to publish monthly is to catch the month it does.
For the full side-by-side on how the two channels differ structurally — conversational context against keyword intent, and what that does to measurement — see ChatGPT Ads vs Google Ads.
What this does not cover
Nothing here covers creative performance, because the sample is not large enough to say anything defensible about it yet. Nor does it break out by geography — every account is North American, and the location targeting available on the platform is coarse enough that a city-level read would be invented rather than measured.
If you are in a restricted category, the route in matters more than the benchmark does — start with restricted industries on ChatGPT Ads, then the two routes through.
How this gets updated
New figures land monthly, recomputed from the same source and by the same method, with the sample size restated each time. When the sample changes materially — a vertical crossing into a meaningful account count, or a category opening up — that gets noted rather than absorbed silently into a revised average.
If you want to run the channel yourself, the field-by-field setup walkthrough is how to advertise on ChatGPT, and the overview of what the channel is and who can buy it sits on ChatGPT Ads for small business.
Want these numbers for your own account?
Bytown runs ChatGPT Ads alongside Google and Meta, and reports on all three in one place. Real marketers, one flat fee, full account transparency.
Book a callOr try the app and connect an account yourself.
FAQ
What does a click actually cost on ChatGPT Ads? Across the accounts we manage the blended cost per click is $2.42, from 42 accounts and 102 account-months covering $51,095 of spend between May and July 2026. The blend hides a wide spread, so read your own vertical's row rather than the headline figure.
Why does every other page quote $3 to $5? Because that is a different thing. It is the maximum bid OpenAI recommends you set — a ceiling you choose — not what you pay after the auction runs.
How wide is the spread between verticals? Roughly nine times, from $0.74 for non-profit to $6.65 for engineering.
Where do these numbers come from? Bytown's own advertiser reporting across the ChatGPT Ads accounts we manage, hand-compiled from that reporting rather than exported through a platform integration. Every rate is computed from summed raw counts — total spend divided by total clicks — never as an average of per-account rates, which would give a small test account the same weight as one spending five times as much.
Is this an industry benchmark? No. It is one agency's book of business, and the verticals in it are the ones our clients happen to work in. A vertical backed by two accounts tells you what those two accounts paid, not what the category costs — which is why the account count sits in the table next to every row.
Why are restricted categories like dental, legal and med spa in the table? Those accounts run on a mix of the two routes OpenAI's policy allows, account by account. We do not attribute a route to any particular vertical, because doing so would identify clients.
