A hint written as keywords
Context hint scheduling app, dispatch, HVAC
Ad All-in-one business software
The click opens Your home page
Below the answer No ad, or one that misses the question
A weak match, and clicks that go nowhere
What it is
What does a ChatGPT ads agency do?
ChatGPT ads are sponsored placements shown below an answer to adults on the Free and Go plans. OpenAI matches them to the conversation, the landing page and your context hints, not to keywords. An agency checks whether your offer can run, writes the hints and ads, connects tracking and shows whether the channel deserves more budget. The channel is new, and no one has years of results yet.
A buyer asks ChatGPT: “What scheduling software works for a 12-person HVAC company?” Two ways to write the hint:
A hint written as a situation
Context hint Scheduling for HVAC crews of 5 to 50
Ad Scheduling built for HVAC crews
The click opens Your HVAC scheduling page
Below the answer An ad that continues the question
A clear match, and an inquiry your sales team can qualify.
Ads appear below the answer and don’t change it. Being named inside the answer is a different job. See which service fits ↓
Step one
What we check before you spend on ChatGPT ads
The readiness check is free and takes a few business days. We look at where your company is registered, your category under the ad policy, your pages and tracking, and what a pilot budget would buy. You get a written review, then a call with Mike or Alex. If you already run ChatGPT ads, we review the account instead.
Our check format, filled in for a field service scheduling software company
01 Where your company is registered
Whether self-serve Ads Manager is open to your company, and whether ads run where your buyers are.
Check: Where your company is registered
Ready- What we saw
- The company is registered in the US, where self-serve Ads Manager is open, and sells to US companies, where ads run.
- Why it matters
- Self-serve is open only to companies registered in supported countries, and ads don’t run in every one of them. Everything else waits on that.
- What to change
- Nothing. The ad account opens in your company’s name, and our team works with the roles you assign.
- How we’ll check
- The account and its roles, at the start of the pilot.
Check it yourself. In which country is the company that pays for the ads registered, and where are your buyers?
02 Your category under the ad policy
Whether OpenAI’s ad policy allows your offer, and with what limits.
Check: Your category under the ad policy
Ready- What we saw
- Scheduling software for trades is an allowed category. Two draft claims about savings need proof on the page before they run.
- Why it matters
- Some categories are prohibited, such as dating and gambling. Housing, jobs, finance and health services run under their own rules, and some only in the US.
- What to change
- Nothing in the category. The two claims either get proof on the page or come out of the ads.
- How we’ll check
- Approval status of every ad before launch.
Check it yourself. Does your offer touch money, health, legal advice, housing or jobs?
03 The page each click opens
Whether the page answers the question the buyer asked.
Check: The page each click opens
Fix first- What we saw
- The ads would open the home page: “Everything your business needs.” Nothing on it answers a question about a 12-person HVAC crew.
- Why it matters
- A buyer who described a situation expects a page about it. A general page loses most of them.
- What to change
- One page per situation the pilot tests, starting with HVAC crews of 5 to 50.
- How we’ll check
- Inquiries per 100 clicks, page by page.
Check it yourself. Open the page your ad would link to. Does it answer the question in the first screen?
04 Tracking
Whether the pilot can tell a click from an inquiry, and count each inquiry once.
Check: Tracking
Fix first- What we saw
- Forms were counted in your analytics, but there was no OpenAI Pixel or Conversions API, and the campaign tags changed per click.
- Why it matters
- Without tracking OpenAI can read, the pilot can’t learn or be judged. Only static UTM tags work in ChatGPT ads.
- What to change
- Pixel and Conversions API with one event ID, so each inquiry counts once, static UTM tags, and a weekly check of Ads Manager clicks against your analytics.
- How we’ll check
- Ads Manager clicks against analytics visits, every week.
Check it yourself. Send a test inquiry. Does it show up once, with its source, in your analytics and your CRM?
05 Context hints
Whether the hints describe situations, not keywords.
Check: Context hints
Test- What we saw
- Your draft hints were keyword lists: “scheduling app, dispatch, HVAC”.
- Why it matters
- OpenAI matches ads to conversations. A hint written as a list misses buyers who describe a situation in their own words.
- What to change
- One plain sentence per situation, in several variants: what you offer, to whom and when.
- How we’ll check
- Click-through by hint in the first two weeks.
Check it yourself. Write one sentence: what you sell, to whom and in what situation. Would a buyer recognize their problem in it?
06 Budget and benchmark
Whether the pilot budget buys enough clicks to judge, and what it is judged against.
Check: Budget and benchmark
Fix first- What we saw
- The planned budget was $25 a day for 30 days: about 150 clicks at $5 a click, and no search benchmark.
- Why it matters
- Below 300 clicks we can’t judge whether clicks bring inquiries, and without a benchmark nobody can say whether a cost is good.
- What to change
- $2,000 to $5,000 a month for two to three months, judged against your non-brand search cost per qualified lead over the same period.
- How we’ll check
- Clicks and qualified leads against the stop rules, every week.
Check it yourself. What does a qualified lead from your search ads cost today, without searches for your name?
Every check ends as Ready, Fix first or Test. After the call, you choose: run the pilot with us, on your own or not at all.
Budget math
Is $25 a day enough for ChatGPT ads?
Usually not for a B2B pilot. $25 a day is the minimum OpenAI accepts in the US. Over two months it buys a few hundred clicks: enough to judge the clicks, rarely enough to compare costs with search. We recommend $2,000 to $5,000 a month.
| A day | Ad spend | Clicks | Qualified leads | What it’s enough for |
|---|---|---|---|---|
| $25 | $1,500 | 300 | 3.6 | enough to judge the clicks, too few to compare |
| $70 | $4,200 | 840 | 10.1 | enough to compare with search |
| $100 | $6,000 | 1,200 | 14.4 | enough to compare with search |
| $165 | $9,900 | 1,980 | 23.8 | enough to compare with search |
Ad spend per qualified lead is $417 in every row: the budget changes how much you learn, not the cost.
In the US, OpenAI’s minimum is $25 a day, averaged over seven days, and it suggests starting bids of $3 to $5 a click. Other countries have their own minimum in local currency. Prices vary by country and change as the channel grows. The two thresholds are our pilot rules, not OpenAI requirements.
How the math works
Ad spend = daily budget × pilot length. Clicks = ad spend ÷ cost per click. Inquiries = clicks × the share that become inquiries. Qualified leads = inquiries × the share that qualify. Results can be fractional: they are expected values, not a forecast.
Our pilot rules: at least 300 clicks before we judge the channel, and at least 10 qualified leads before we compare its cost with search. Below 10, one lead more or less changes the comparison too much.
Setup and the pilot
What’s included in a ChatGPT ads pilot
Eligibility and tracking come first. Setup takes 5 to 10 business days, depending on complexity. The pilot then runs for two to three months, billed monthly, and ends with a decision.
Setup · up to 10 business days
◆ You approve
Days 1–2 Eligibility and policy
Where your company is registered, your category under the ad policy and the claims your ads can make.
◆ You confirm what counts as a qualified lead.
From you View-only access to your analytics, and your site or tag manager.
Days 3–4 Buyer questions
The questions buyers ask before choosing, grouped into situations your offer answers.
From you Sales call notes or the questions your team hears most, if you have them.
Days 5–6 Tracking
OpenAI Pixel and Conversions API with one event ID, static UTM tags and a lead log, checked with a test inquiry.
From you Access to Ads Manager with the role you assign.
Days 7–9 Hints, ads and pages
Context hints by situation, ad variants and a brief for the page each click opens.
◆ You approve the hints, the ads and the stop rules.
Day 10 Launch
The pilot goes live. In the first week we check delivery every day and every new inquiry.
From you Your sales team’s ratings of new inquiries.
A simple setup takes 5 business days. The days start once we have access and your inputs.
During the pilot
- Every week
- Delivery and click-through by hint, Ads Manager clicks against your analytics, and new inquiries against your sales team’s ratings.
- Every month
- A short report, new hints and ads to test, and the hints that found no one switched off.
- At the end
- Cost per qualified lead next to your search ads, and our recommendation: scale, change or stop.
Stop rules agreed before launch
- No delivery
- 14 days without impressions after a bid increase.
- Clicks we can’t trace
- Ads Manager clicks still don’t match your analytics after the tracking fixes.
- No inquiries
- 300 to 500 clicks bring no inquiries.
- Any rule met
- We stop early and report what we learned.
Measurement
How do we measure a ChatGPT ads pilot?
Ads Manager reports views, clicks and conversions, not the conversations behind them. We follow each inquiry to your sales team’s verdict and compare its cost with your search ads. Pick a step for the details.
Delivery
- We count
- Impressions and click-through by hint, to see whether a hint reaches the right conversations.
- We leave out
- Conversions after a view without a click. OpenAI reports them separately, within one day of the view, and so do we.
- Source
- Ads Manager. It reports views, clicks and conversions, not the conversations behind them.
Clicks reconciled
- We count
- Ads Manager clicks against visits in your analytics, with static UTM tags on every ad.
- We leave out
- Clicks we can’t trace after the tracking fixes. They trigger a stop rule.
- Source
- Ads Manager and your analytics. Conversions can take 24 to 48 hours to appear.
Qualified inquiries
- We count
- Forms and calls that meet the buyer criteria you agreed on, once each, through the Pixel and Conversions API with one event ID.
- We leave out
- Duplicates, test records and inquiries sales turns down.
- Source
- Your CRM or a shared sheet with your sales team’s ratings.
Cost against search
- We count
- Ad spend per qualified lead, next to your non-brand search cost per qualified lead over the same period.
- We leave out
- Searches for your name, which would look cheaper than any new channel.
- Source
- Ads Manager, your search ads and your CRM. Customers and pipeline follow on your sales cycle.
What you get during the pilot
A short report every month and one at the end, on what was checked, changed and decided.
Our report format, filled in for six weeks of a field service software pilot.
Pilot report
Weeks 1–6
Checked
- Test inquiry through the Pixel and Conversions API
- Counted once
- Ads Manager clicks against your analytics
- Every week
- Stop rules
- None met
Changed
- The hint “scheduling app, dispatch, HVAC” switched off, it found no oneYou approved in week 3
- New hint for owners moving dispatch off spreadsheetsReview in week 8
Found and fixed
Ads Manager showed more clicks than your analytics showed visits. Fixed in week 2: static UTM tags on every ad, and the landing page no longer redirects.
So far
- Situations in the test
- Four hints, judged one by one
- Comparison with search
- At week 8
Next
Compare cost per qualified lead with your non-brand search at week 8, then recommend: scale, change or stop.
By business
Do ChatGPT ads work for your kind of business?
Too early to say in general. They fit offers with a clear use case: in a conversation, buyers describe their situation in their own words. Pick your business to see the situation a pilot would test.
B2B software
- A buyer asks
- “What scheduling software works for a 12-person HVAC company?”
- Context hint
- Scheduling for HVAC and plumbing crews of 5 to 50 that dispatch daily.
- The page the click opens
- A page for crews of that size, with setup time and pricing.
- What counts as qualified
- A demo request from a company with 5 to 50 technicians.
Travel
- A buyer asks
- “Where can a team of 40 hold a three-day offsite near Denver?”
- Context hint
- Offsites for teams of 20 to 80 near Denver, with rooms and meeting space.
- The page the click opens
- A page with room blocks, meeting rooms and sample agendas.
- What counts as qualified
- A quote request with dates and a group of 20 or more.
Fintech
- A buyer asks
- “How can a 15-person company set up a 401(k) for its team?”
- Context hint
- 401(k) plans for companies of 5 to 100 staff starting their first plan.
- The page the click opens
- A page with fees, setup steps and the payroll systems it connects to.
- What counts as qualified
- A call request from a company of 5 to 100 staff with no plan yet.
Finance ads run only in the US, under extra rules.
Dental
- A patient asks
- “Which dentists in San Diego are open on Saturdays and take PPO plans?”
- Context hint
- Family dentist in San Diego with Saturday hours and PPO plans.
- The page the click opens
- A page with hours, accepted plans and online booking.
- What counts as qualified
- A first visit your front desk books.
Health services run only in the US, under extra rules. The pilot tests questions about hours and insurance, not symptoms.
Team
Who runs your pilot?
Our paid media team runs the pilot day to day. The founders set the direction, and one of them answers for your project personally.
-
-
Every project has a founder on it and a team lead for each service.
- Paid media teamRuns the pilot day to day: eligibility, tracking, context hints and ads, and the weekly check of clicks against your analytics.
If the people on your pilot change, the plan, the account history and our notes stay with the project, and so does a founder.
- 2015founded
- 100+client projects
- 15+projects for US companies
Most of our clients come through a recommendation. Meet the team →
Price
How much does ChatGPT ads management cost?
We start with a pilot, not a long retainer: setup from $X, then pilot management from $X a month for two to three months. You pay for media directly to OpenAI from your own account. For a B2B pilot we recommend $2,000 to $5,000 a month.
Free readiness check
$0
A few business days
- Country, ad policy, pages, tracking and click economics
- A written review: what can run, what blocks it and what a pilot would cost
- A call with Mike or Alex to go through it
Pilot
From $X setup, then $X a month
5–10 business days, then 2–3 months billed monthly
- Eligibility, Pixel and Conversions API, and the test plan with stop rules
- Context hints and ads by situation, with launch checks
- Weekly delivery and tracking checks, and a monthly report
- A decision at the end: scale, change or stop
Ongoing management
From $X a month + X% of ad spend
Only if the pilot meets the rules agreed before launch
- Weekly delivery, tracking and inquiry checks
- New hints and ads every month
- A monthly report against your search ads
How the fee works
- OpenAI bills your company directly for media. We never handle your ad budget.
- We recommend the priorities. You approve the plan, the stop rules and every change in spend.
- Weekly pilot data and your sales team’s ratings decide what changes next.
If the pilot stops
A stop rule can end the pilot early, and we report what we learned. The ad account is in your company’s name, and the account, its history and every report stay with you.
- What moves the price
- The number of situations and campaigns, markets and languages, and how complex your tracking is.
- Not included
- The ad budget, new landing pages, creative production, CRM integrations, call tracking and translations. Each is quoted separately.
Every pilot ends with a decision: scale, change or stop.
Questions buyers ask
ChatGPT ads FAQ
The short answer comes first. Open any question for the detail.
What does the free readiness check include?Whether you can run ChatGPT ads, and should.
We check where your company is registered, your category under OpenAI’s ad policy, the pages a click would open and your tracking. We estimate what a pilot budget buys at OpenAI’s suggested click costs. It usually takes a few business days. You get a short written review and a call with Mike or Alex. If you already run ChatGPT ads, we review the account instead.
How much do ChatGPT ads cost?From $25 a day in the US, plus the price of each click.
In the US, OpenAI’s minimum is $25 a day, averaged over seven days. Other countries have their own minimum in local currency, such as £15 in the UK. You can pay per thousand views, per click or for conversions. OpenAI suggests starting bids of $3 to $5 a click. Early published tests report averages of about $4 to $10, and prices vary by country. For a B2B pilot we recommend $2,000 to $5,000 a month in media, paid directly to OpenAI. Our fees are separate.
Who can advertise on ChatGPT?Companies registered in supported countries.
Self-serve Ads Manager is open to advertisers whose company is registered in one of more than 60 supported countries. Ads don’t run in every one of them, so the readiness check also looks at where your buyers are. Ads reach adults on the Free and Go plans, not on Plus, Pro, Business, Enterprise or Edu. Some categories are prohibited, such as dating and gambling. Housing, jobs, finance and health services run under their own rules, and some only in the US.
Do ads change ChatGPT’s answers?No, according to OpenAI.
OpenAI says ads appear below the answer, are labeled as sponsored and do not influence what ChatGPT says. Advertisers don’t receive conversations, and reporting is aggregated. Ads are chosen from the conversation, the landing page, the ad and your context hints. Being named inside the answer itself is the work of AI search optimization, not ads.
How are ChatGPT ads different from Google Ads?Conversations, not keywords.
There are no keywords, match types or negative keywords, and no report by conversation. You describe situations in context hints, and OpenAI matches ads to the conversation. Tracking runs through the OpenAI Pixel and Conversions API. Conversions can take 24 to 48 hours to appear, and only static UTM tags work. That makes a search benchmark and careful tracking more important.
How long should a pilot run?Two to three months, with early stop rules.
Long enough to buy 300 to 500 clicks and let inquiries qualify. We stop earlier if ads don’t deliver for 14 days after a bid increase, if Ads Manager clicks still don’t match your analytics after tracking fixes, or if 300 to 500 clicks bring no inquiries. At the end we compare cost per qualified lead with your search ads.
Do ChatGPT ads work for B2B?Too early to say in general.
OpenAI has not published benchmarks, and public B2B results are thin. One published pilot for an industrial manufacturer spent about $1,470 in 30 days and recorded no conversions. That is why we start with a two-to-three-month pilot, a benchmark and stop rules, not a long retainer.
Can you guarantee results?No. No one can.
We don’t promise a set number of leads or a cost per lead. No one can, least of all in a channel this new. Before launch we agree the question the pilot answers, the budget, the benchmark and the stop rules. At the end we report what happened and what we recommend, including stopping.
Insights
An article, a guide and a tool for testing a new channel
AI search optimization: what to check, change and measure
An AI answer can leave out a company, describe its offer incorrectly or cite a source that explains the product poorly. Each observation calls for investigation. The practical work starts by recording the answer and checking the information behind it.CheckedMQL vs SQL: what counts as a qualified lead
Marketing and sales often mean different things by “qualified.” Here is one deal followed through both marks, a one-page agreement to fill in, and a way to teach Google, Meta and LinkedIn from the second mark.Checked
Start with a free readiness check
Can your company run ChatGPT ads, and should it?
We check where your company is registered, your category under OpenAI’s ad policy, your landing pages and tracking, and what a pilot budget would buy. The check is free.
What happens next
- 01Tell us your site, your offer and the market you sell in.
- 02Within a few business days you get a written review: what can run, what blocks it and what a pilot would cost.
- 03Go through it with Mike or Alex. Run the pilot with us, on your own or not at all.
Already running ChatGPT ads? We review the account for free: delivery, context hints, tracking and what to change first.