This guide goes with Cited or Bought. That article explains that now there are two ways to appear in the answer your client reads, and that only one keeps working when you stop paying. This is what comes next: how to split the money between the two, with your numbers and not a case study's.
Before starting, two honest warnings.
The first: there are no AI ad prices here. Buying opened days ago and no public references exist for a small service business. Anyone giving you a figure today is making it up, us included if we tried.
The second: this does not diagnose. That is what the two-slot sheet that arrived with this guide does. If you have not filled it in, fill it in before continuing: this guide splits money, it does not diagnose.
Your ceiling, and from there your monthly cap
Before splitting you need a ceiling, and it is three numbers you already have: your average ticket, your margin on that ticket, and your close rate from inquiry to sale. If you have never calculated them, the how is in the September 7 recap, with the trap in each one. Five minutes.
What that article does not do is what is below, and that is what you need to split.
The calculation runs in two steps, because they are two different numbers and they are used for different things.
- Average ticket x margin = the maximum you can pay for a new client without losing money on that sale.
- That number x your close rate = the maximum you can pay per inquiry. If you close 2 in 10, you need five inquiries for one client, so the ceiling per inquiry is one fifth.
The second is the one you compare against an ad, because an ad brings you inquiries, not clients. Confusing the two is the most common way to shut down a channel that was making money.
From there comes your monthly cap
Ceiling per inquiry x the new inquiries you can handle this month without the quality of attention slipping.
If your ceiling per inquiry is $100 and you can handle ten new inquiries a month, your cap is $1,000. And it is not negotiable by enthusiasm. Write it down before looking at any platform.
What this is for. The bought slot is measured against that cap from day one. The cited slot is not, and that is the difference that makes them hard to compare: one gives you an invoice and the other gives you an asset. That is why the split is not decided with a calculator, it is decided with an order.
Split by stage, not by budget
The most common mistake is splitting by fixed percentage. It does not work, because the two slots do not benefit equally from money at the same moment.
How to know which stage you are in, without diagnosing again. Your sheet already told you. Translate its verdict:
| What your sheet said | Your stage |
|---|---|
| "All your visibility is rented" or "none is readable by a machine" | A |
| "Less than half of what's yours is readable" or "you rent more than you have published" | B |
| "Your base is solid" | C |
Stage A - nothing of yours is readable yet
Split: all effort to the cited slot. Zero to the bought slot cap.
And "all" here is not money: it is hours. This stage costs almost no budget, it costs someone sitting down an afternoon to organize information that already exists in your business.
Why zero on ads. An ad at this stage takes people to a page that does not answer. You are paying for a stranger to confirm you were not the answer, and that money leaves nothing behind: not a published fact, not a mention, not a better page.
When you leave this stage: when the facts your client asks before buying from you are published as plain text on your own site.
Stage B - your facts are readable but nobody mentions you
Split: almost all effort stays on the cited slot. The paid trial gets a small, capped portion - the size is fixed by your cap from Step 1, not a percentage.
Here you can buy, but as a trial and not as a channel. And a timeframe that is not chosen by habit: the one your own sales cycle takes to complete at least once. If from first inquiry to payment takes three weeks, the trial cannot run for two.
One question to answer: does the people arriving through the ad do the same as the ones arriving recommended, or do they leave sooner? If they leave sooner, the problem is not the ad: it is the page where they land. Fix that before raising the budget, because raising it just buys more confirmations of the same problem.
Stage C - they already mention you and your page converts
Split: here you can set the bought slot a stable monthly budget, with Step 1's cap and one rule: paid spending never grows faster than what you have built.
If in one month only the spending goes up, you start depending more on the budget and less on yourself. The way to prevent it is to write it down first: every budget increase gets approved with a written line on what gets published that same month. Without a written line, no increase.
The twelve-month account, with your numbers
The blog runs it as reasoning. Here it fills in with your own monthly budget, which we call P.
| I start buying | I start publishing | |
|---|---|---|
| Months 1 to 3 | P in ads each month | No media spending. Hours to organize and publish |
| Months 4 to 12 | P in ads each month | P in ads, on a base that already works |
| Media spent in one year | 12 x P | 9 x P |
| Hours spent on your end | none | the three months of organizing and publishing, and those are not free |
| What I have if I cut in month 13 | nothing on that screen | the published facts and the mentions stay |
| What I taught the system | that my page receives paid traffic | that my business has verifiable facts |
What this table makes obvious is not that one path is cheaper. It is that one gets paid entirely in money and the other gets paid three months in time, and at month thirteen only one of the two keeps something. The only reason almost nobody chooses the second is that those three months are not visible.
The caveat, which goes here and not in a footnote: this is arithmetic about what stays, not a measured case, for the same reason we say in the blog. The only thing in the table that is verifiable is the second-to-last row, and that does not depend on any case.
Five questions before signing an AI ads contract
If you decide to buy, ask these five questions to whoever is going to manage it. They are not traps: they are the five that separate someone who knows from someone selling the novelty.
1. What percentage of responses in my category carry an ad today? The honest answer is "not publicly known." If they give you a percentage, ask for the source. If they do not have it, you already know how they will measure the rest.
2. Where exactly will this land, and does that page answer the full question the person typed in the chat? If the answer is "the homepage" or "a form," the contract is premature. How to build that page is in appearing in ChatGPT does not pay the bills.
3. How will we distinguish paid traffic from traffic arriving through recommendations? Without that separation you will not be able to tell whether the ad works or whether it is simply charging you for people who would have found you anyway.
4. What do I have left if I stop for three months? The correct answer is "nothing on that screen," and whoever gives you that answer is the one telling the truth. Whoever says it accumulates, is not.
5. What part of this budget goes to work that stays? If the answer is zero, you know exactly what you are buying. That can be fine, but let it be a decision and not a discovery at six months.
Your written commitment for this month
Your sheet already gave you your first move. This is the other thing, what the sheet cannot give you: leaving the decision in writing before the enthusiasm moves it.
Write these four lines and save them where you will see them again next month:
- My monthly cap for the bought slot is ___ (from Step 1).
- This month I spend on media ___, which is equal to or less than the cap.
- And this month I publish ___, one single thing, with a date.
- If I buy, the trial closes on ___ and the question it answers is ___.
Without line 3, line 2 does not get approved. That is the whole rule.
And one thing that costs us to say
The split we just gave you is less convenient for us than the opposite, and it is the one we use. An agency gets paid every month to manage an ad budget, and paid once to organize your information so you do not need to come back.
If you want us to work through this split with you, with your numbers and your stage on the table, the call is 30 minutes, free and no commitment:
And if you prefer to stay close without a call, every Friday at 4pm we send The AI Edge, the week in AI summarized into what actually changes a decision of yours. You can subscribe at hub365.ai/newsletter.
What this guide does not cover, and where it is
- The diagnosis of your two columns: in the interactive sheet that arrived with this guide, My Two Boxes.
- How to earn the cited slot, step by step: in The lists AI recommends.
- What to do when you are already being cited and still not getting calls: in Appearing in ChatGPT does not pay the bills.
Sources
- Opening of ad buying inside ChatGPT, OpenAI, August 31, 2026 - verified at source September 8: the page is dated August 31 and states self-serve buying starts that day - https://openai.com/index/expanding-access-to-ai-with-chatgpt-ads/
- Designation of ChatGPT search as a very large online search engine, European Commission, September 1, 2026 - https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1772
