ChatGPT will sell you an ad now. The math says wait

Somebody forwarded me a screenshot last week. A 1.49x return on ad spend from ChatGPT ads, with the word "finally" typed underneath.

The number is real. It came out of an account that spent about $60,000 in fifteen days.

That is $4,000 a day. Most shops I audit in Indianapolis do not spend $4,000 on advertising in a year.

OpenAI opened ChatGPT ads to any US business with a credit card on July 22, 2026. The clicks are cheaper than Google. And almost none of you should buy them this quarter, for reasons that come down to arithmetic rather than any opinion about AI.

What OpenAI actually changed on July 22

Ads have been running inside ChatGPT since February 2026. Buying one used to require a $60 flat CPM and a $200,000 minimum, which is why the early advertisers were Best Buy, Lowe's, and VistaPrint rather than anybody on College Avenue.

On July 22, 2026 OpenAI opened a self-serve Ads Manager. You register with a work email, submit your company details and service area, wait out an approval queue, add a card, and build a campaign. No minimum spend. Cost-per-click bidding now sits alongside the old CPM option.

Two campaign objectives exist right now. Reach, bought on CPM. Clicks, bought on CPC. Conversion optimization is not an objective yet, though you can install a pixel and the Conversions API to track what happens after someone clicks.

There is no keyword field. Targeting runs on what OpenAI calls context hints, which are plain-language descriptions of the situations where your service makes sense. You describe the conversation you want to appear in.

Geography is the part that should interest a local shop. You can target by state, by designated market area, or by ZIP code. ZIP-level targeting is sharper than what Google Ads hands most local advertisers by default.

Who actually sees your ad?

This is where the channel gets small.

Ads run only for logged-in users over 18 on the Free and Go tiers. Anyone paying for Plus, Pro, Business, Enterprise, or Education sees nothing. Neither does anyone browsing logged out.

OpenAI also keeps ads out of temporary chats, image-generation conversations, and any session it flags as sensitive. Sensitive covers health, mental health, and politics.

Read that last line twice if you run a clinic.

A chiropractor in Broad Ripple cannot buy the conversation "my lower back has hurt for three weeks, what should I do." ChatGPT treats that as a health conversation and runs no ads against it. Same for the dentist, the urgent care, the vet, the physical therapist. On top of that, health claims sit on the restricted-category list next to dating, financial services, and political advertising.

OpenAI carved a large share of local service businesses out of the inventory before they can even open an account.

How much does a ChatGPT ad cost?

OpenAI suggests starting CPC campaigns with a maximum bid of a few dollars. Reported 2026 CPCs land between $3 and $5 for most categories, stretching to $18 in financial services because money questions draw the heaviest bidding. CPM rates in 2026 run roughly $25 to $60 per thousand impressions.

Now compare Google. Average plumbing CPC on Google Ads in 2026 is $8.45. Home service trades run $8 to $45 a click.

So the clicks are cheaper. That part of the pitch is true.

The fifteen days of real spend I mentioned at the top produced a $1.72 CPC, a 2.35% conversion rate, and 1.49x blended return on roughly $60,000. Daily return inside that same window swung between 0.2x and 2.9x.

That swing matters more than the average does.

The math that says wait

Run your own numbers. I will use $500 a month, the low end of what practitioners recommend for a real 60-day local test.

At $5 a click, $500 buys 100 clicks. At the optimistic $1.72, it buys 290. Apply the 2.35% conversion rate and you get this.

Monthly budget	CPC	Clicks	Conversions
$500	$5.00	100	2.4
$500	$1.72	290	6.8
$1,000	$5.00	200	4.7
$2,000	$5.00	400	9.4

The people running this channel say to wait for 20 to 30 conversions before you judge it. At $500 a month and a $5 click, 20 conversions takes about eight months and $4,250. At the friendlier $1.72 click, three months and $1,500.

A test you cannot afford to finish is a donation.

The reporting works against you too. The advertiser who published that fifteen-day window found the platform's own numbers came in under what first-party tracking showed. Cost per click and cost per acquisition appear in the dashboard. Return on ad spend does not. If your website cannot already tell you which channel produced a booking, ChatGPT will not tell you either.

Now look at where else that $1,500 goes. Google Local Services Ads averaged $53 per lead across $6.72 million in spend from 888 contractors in February 2026. HVAC came in at $51, plumbing at $57, electrical at $39. Fifteen hundred dollars buys about 28 leads you can count this month instead of next spring.

One more number for scale. Analysts expect total US chatbot advertising revenue to stay under $1 billion for all of 2026. Google's ad business clears that before lunch. This is a real channel, and a very early one.

What if you want to test it anyway?

A few of you should. Do it under these conditions.

Your Google Ads or Local Services spend is already profitable and you are not defunding it to pay for this. New channels come out of surplus.

You are not in health, finance, dating, or politics. Check the restricted list before you build a single ad.

You have first-party conversion tracking that works. Pixel or server-side, tagged with UTMs, tied to actual bookings and calls. Without it you are guessing at a number the platform already undercounts.

You can put $2,000 across 60 days without it changing any other decision you make. That reaches a readable sample in one quarter instead of three.

Your context hints describe situations. "Someone comparing quotes to replace a water heater in an older home" will do more for you than "water heater replacement."

Your ZIP targeting is tight. Five to fifteen miles inside the city, twenty to forty out in the county.

Cannot tick all six? Then the answer this quarter is no, and waiting costs you nothing. The auction will not get harder to enter. It will get more expensive, which is a different problem, and by then other people will have paid to publish the benchmarks you are missing today.

I will write about this again in the spring with better numbers. The only advertiser I would send into ChatGPT right now is one who already knows, to the dollar, what a single customer is worth. Most shops I meet cannot answer that. It is the more useful thing to go fix this week.

---

Sources

#	Source	What it backed up
1	Ads in ChatGPT	OpenAI's own rules on who sees ads: logged-in US adults on Free and Go tiers only, with health, mental health, politics, temporary chats and image generation excluded.
2	OpenAI Launches Self-Serve Ads Manager for ChatGPT	The July 22, 2026 self-serve launch, CPC bidding, Conversions API and pixel measurement, aggregated-only reporting.
3	OpenAI Opens Ad Platform To CPC Bidding, Self-Serve Buys	The shift off the original CPM-only pilot, and the agency and measurement partners behind the rollout.
4	How to Advertise in ChatGPT: A Walkthrough of OpenAI Ads Manager Beta	Account setup and approval queue, the Reach and Clicks objectives, context hints replacing keywords, ZIP-level geo targeting, second-price auction.
5	ChatGPT Ads Benchmarks: ROAS, CPC, and Cost From 15 Days of Real Spend	The $1.72 CPC, 2.35% conversion rate, 1.49x blended ROAS on roughly $60,000, the 0.2x to 2.9x daily swing, and the platform under-reporting conversions.
6	Average CPC Plumbing Services Google Ads 2026	The $8.45 average plumbing CPC on Google Ads in 2026, used as the comparison against ChatGPT's $3 to $5.
7	Google Local Service Ads Cost Per Lead by Trade (2026)	The $53 average cost per lead across $6.72M in spend from 888 contractors, plus HVAC at $51, plumbing at $57 and electrical at $39.