Most agencies selling “ads in LLMs” hope you never ask where the auction lives. I ask it first. Google’s AI Overview ads run through existing Google Ads campaigns; ChatGPT ads are bought from a different seller in a separate account. If someone quotes one price for both, they have skipped the part that determines what you actually pay.

Say you run a home services company spending $20k a month on Search. Someone types “emergency water heater replacement cost” into Google and gets an AI Overview with ads underneath. Later, that homeowner asks ChatGPT what size tank they need and sees a sponsored answer. Same potential buyer, similar problem, two different buying models.

Everything else follows from that split. Google’s AI Overview and AI Mode ads use existing Search and Performance Max campaigns, so the spend sits inside your Google bids and budgets. ChatGPT ads live in OpenAI’s Ads Manager, with their own objectives, targeting and prices. Once you know where each purchase happens, you can work out what an agency claiming to manage “ChatGPT ads” must actually be doing.

Google’s AI ads use the campaigns you already run

Start with Google, because this is where the new label causes the most confusion. If you already run Search, Shopping or Performance Max, you can serve in AI Overviews and AI Mode through Google Ads, with no new campaign or separate budget required. I used to tell clients that a new placement meant a new line item to manage. I was wrong here. Google treats the AI answer as another place an existing campaign can appear, not a second account you open and fund.

Eligibility still depends on how you build that campaign. Broad match, AI Max for Search, Shopping and Performance Max can match using the full context of a query and its AI response, while exact and phrase match serve in AI Mode on explicit direct commercial intent. Your exact match for [emergency water heater replacement] will not chase every adjacent AI tangent. A broad-match version might, because it can match against more of the surrounding context.

That distinction matters more than the words “AI placement” on a sales slide. Broader matching can reach more AI-related queries, but it also gives you more variation to manage. If the water heater account starts attracting people researching tank sizes rather than booking replacements, I would inspect what matched before blaming a new channel. The practical job remains query control.

Nor is there an AI Overview rate card. Google charges for these ads through the same CPC or conversion-based bidding used by the existing campaigns, without an additional placement charge. An AI Overview click draws from the budget you have already set. It does not follow that each such click costs exactly your account’s average Search CPC; an average is not a guaranteed price for an individual click. The point is simpler: there is no separate AI Overview price to quote.

You also get no dedicated placement switch. You cannot target AI Overviews or AI Mode separately or opt out of them separately from Search. Performance folds into normal campaign reporting rather than a clean AI Mode cost-and-conversion report. I know the instinct to go looking for that breakdown. It is the first tab I would want, and it is not a separate tab you can use to settle the question.

For our $20k account, AI surface is therefore potential variation inside the campaigns you already run, not a fresh budget request. If CPA drifts as those placements appear, start with the Search levers you have: matching, negatives, bidding and the page receiving the click. A quote for “AI Overview media” as a separately priced Google product should make you stop and ask what is actually being sold.

ChatGPT requires a second account and a different plan

ChatGPT works the other way. You buy ads in a separate account through OpenAI’s self-serve Ads Manager, not through Google Ads. It is live in the US, UK, Canada, Australia and other markets and is still labeled beta. Our water heater company would open that account, install the pixel, choose an objective, set a daily budget and pay OpenAI for media separately from its Google spend.

Think of Google’s AI surfaces as rooms accessible through the Google campaigns you already rent. ChatGPT is a different building with its own lease. The analogy stops at the door: a conversation is not a search-results page, and you do not carry your Google keyword list across and expect the same buying controls.

Instead, targeting uses plain-language context hints about relevant buyer conversations, alongside conversation context, your landing page and your ad copy as signals. For the water heater company, a hint might describe a homeowner with a leaking 12-year-old tank asking about replacement cost and timing. You can also choose geography down to state, DMA and ZIP, and select iOS, Android or web placement. OpenAI then determines whether the ad fits an eligible conversation.

That changes the work. In Google, you might separate urgent replacement queries from general research and manage the matches. In ChatGPT, you have to describe the situation in which your ad belongs, then make sure the creative and landing page answer that situation. The skill is describing useful conversations, not building another keyword list. A vague hint invites a vague test.

ChatGPT also has its own pricing choices. OpenAI defines Views, priced per 1,000 impressions; Clicks, priced per valid click; and Conversions, optimized toward an eligible conversion event. You set a max bid for the objective you choose. Published starting guidance includes a $60 default max CPM for Reach and a $3 to $5 recommended starting max CPC for Clicks. The auction weighs relevance, so the highest bid alone does not settle which ad appears.

Those figures are starting bid guidance, not a promise about what our homeowner’s click will cost. Pilot pricing was reported at $60 CPM at the February 2026 launch with $200k minimums, then $25 to $45 by April 2026. Self-serve later opened with a $25-a-day US minimum, and reported CPCs ranged more widely. The sequence is useful because it shows why a launch-era price should not become a permanent line in your media plan.

For our $20k water heater account, $25 a day means about $700 over four weeks. That is a test budget, not proof that the channel works. I would use it to learn whether the chosen conversations produce useful clicks and whether the landing page can turn those clicks into the actions the business wants. Only then does a larger budget become an interesting question.

The controls tell you what management means

In Google, you control the things you already controlled: bids, budgets, match types, negatives, assets and landing pages. If AI Mode brings the water heater campaign too much tangential research traffic, you can review matches, add an appropriate negative, tighten matching or make the page answer the commercial question faster. You cannot solve it by dialing down a dedicated “AI Overview budget.” There is no such budget in this setup.

That is familiar work, but familiar does not mean automatic. Someone still has to notice when the intent shifts and decide whether the new traffic deserves the spend. For the $20k client, I would treat AI surfaces as another reason to do Search quality control well, not as a reason to buy a second Google management package.

In ChatGPT, the controls are newer and the eligible audience is narrower. You manage the budget, objective and applicable max bid, context hints, creative, landing page, geography and device surface. Ads go to logged-in adults on Free and Go plans; Plus, Pro, Business, Enterprise and Edu users do not see them. Under-18s are excluded, and ads are blocked near sensitive topics including health, mental health and politics.

That matters for the home services example. A routine replacement-cost conversation and a conversation that turns toward health or safety distress are not the same advertising opportunity. You cannot assume every relevant-sounding chat is available to buy. ChatGPT gives you a separate budget to manage, but only within its eligible conversations. The audience limit belongs in the plan before anyone promises volume.

Cost follows the account that receives the media spend

For Google AI Overviews, the invoice does not gain a new media category. The money still goes to Google through campaigns and budgets you already use; management is whatever you separately pay someone to run them. That is why percentage-of-spend fees bother me here. Suppose AI-related traffic helps lift a $20k account’s monthly spend by 12%, to $22,400. At a 15% management fee, the agency collects $360 more that month. More spend alone does not tell me what additional management work it performed.

The work can be real: reviewing queries, adjusting negatives, watching CPA and fixing pages. But it lives inside Search. I wrote about that fee mismatch in self-serve versus agency versus autonomous models. AI Mode makes the question more pointed. If the fee rises, ask what changed in the work, not just what changed on the invoice.

For ChatGPT, there are two costs you can identify separately. Media goes to OpenAI under the objective and budget you choose, with a $25-a-day US self-serve minimum, or about $700 for four weeks. Management pays for second-account work: setup, pixel and Conversions API, context hints, creative, bids and budget decisions. You can disagree about whether an agency’s fee is worth that work, but at least there is a distinct account and a distinct job to inspect.

One rule saves time: if an agency cannot show you the OpenAI account holding the spend, it has not shown you ChatGPT ad management. A Google Ads screenshot with an AI Overview annotation may show legitimate Search work. It does not show media bought from OpenAI.

Four questions that test a “we run ChatGPT ads” pitch

Ask to see the account before you ask for results. A shop managing ChatGPT inventory should be able to open OpenAI’s Ads Manager and walk through the objective, the applicable max bid, the context hints, the pixel and Conversions API status, and the media paid to OpenAI versus the fee paid to the shop. I use four questions:

  1. Which account holds the media spend? Google Ads and OpenAI Ads Manager are not interchangeable answers.
  2. What conversation are you trying to reach? Ask to see the context hints, not a slide that says “AI audience.”
  3. What did we pay OpenAI, and what did we pay you? Keep media and management on separate lines.
  4. How do you distinguish Google AI-surface work from ChatGPT clicks? The answer should respect the reporting limits on Google’s side, not claim a placement breakdown the account cannot provide.

A shop that is relabeling Google work will tend to blur those answers. It may show a Google Ads campaign with an AI Overview callout or talk about ChatGPT visibility without a separate OpenAI budget. Neither is the same as running ChatGPT ads. This is not an argument against hiring help. It is a way to make sure the help matches the product on the invoice.

Now take a new case through the same model. Say you are a SaaS team spending $30k a month on Search, and an agency pitches “ChatGPT ads” for $2,000 plus media. Where does the proposed spend go? If the agency means Google AI Mode, there is no separate ChatGPT purchase. The fee needs to justify work in your existing Google campaigns: matching, negatives, bidding and pages that keep CPA in line. Calling those tasks ChatGPT management does not make them so.

If the agency means ads bought from OpenAI, the $2,000 needs to cover a second account and a test plan. I would ask which buyer conversations the hints describe, how the creative fits the landing page, which geography and device surfaces the team chose, and what would make it stop or scale a roughly $700, four-week test at $25 a day. I would consider that test if the agency named the hints and the kill rule up front. I would pass on a logo slide about AI with no account or auction attached.

That distinction is why I track this closely at groas, where Search Ads covers Google Ads, Bing and ChatGPT ads for one flat $1,499 a month, with media paid separately. The bundle math matters: a 15% fee on $20k is $3,000. But the stronger test is whether the work stays visible. Google AI work is continuous control of existing campaigns. ChatGPT work is a separate assistant account with its own bids and hints. If your setup cannot show which job was done and what happened to CPA, you are paying for a label. Follow the auction. Then look for the execution.