AI Search Advertising Compared: ChatGPT, Google AI Mode and Copilot Ad Rules

Ads now run inside the answers of all three of the largest AI answer surfaces. Only 1 of them lets you see what those ads did on their own. We read each platform's advertiser documentation for AI search advertising, and 2 of 3 say plainly that they offer no separate report for ads served inside the AI answer.

The short answer

ChatGPT ads are bought in OpenAI's own Ads Manager and appear below answers on Free and Go plans. Google AI Mode and AI Overviews ads come from existing Search, Shopping and Performance Max campaigns. Copilot ads come automatically from Microsoft Advertising campaigns. Only ChatGPT reports its answer ads as their own channel.

What counts as AI search advertising in 2026

AI search advertising here means a paid placement shown inside or beside a generated answer, rather than in a list of blue links. The three surfaces that matter by reach are ChatGPT, Google's AI Mode and AI Overviews, and Microsoft Copilot.

They look similar on a screenshot. They are not similar to buy. One is a new ad platform with its own account, its own pixel and its own bidding. The other two are new placements bolted onto ad platforms you already pay for.

That difference decides almost everything that follows. It decides whether you can choose to be there, whether you can see what happened, and who in your team owns the budget. We think it matters more than any difference in ad format.

This comparison is built only from what each company publishes for advertisers, plus trade press reporting on those documents where the original page could not be opened. Where a platform does not say something, we say it does not say it. We have not filled gaps with agency guides, which disagree with each other on basic facts such as how many ad formats ChatGPT runs.

ChatGPT ads: format, rules and the buying path

OpenAI announced ads on 16 January 2026. The first test covered logged-in adult users in the US on the Free and Go tiers, according to PPC Land's report on the announcement. Plus, Pro, Business and Enterprise subscriptions carry no ads.

The ad sits at the bottom of an answer when there is a relevant sponsored product or service based on the current conversation. It is labelled as sponsored and kept visually separate from the answer itself.

The rules OpenAI has published

OpenAI framed the launch around stated principles. The two that matter to an advertiser are answer independence and conversation privacy. The company's position is that ads do not influence the answers ChatGPT gives, and that conversations stay private from advertisers and user data is never sold to them.

The exclusions are specific. Ads do not appear near sensitive or regulated topics such as health, mental health or politics. They do not appear where a user indicates they are under 18, or where OpenAI predicts underage use.

Users get real controls. They can turn off ad personalisation, clear the data used for ads, see why an ad appeared, dismiss it and give feedback. OpenAI also committed to an ad-free paid tier.

How you buy and what you get back

On 5 May 2026 OpenAI opened a self-serve Ads Manager in beta to US advertisers, reported by Search Engine Journal. During the early pilot most buying was on a CPM basis. The beta added cost-per-click bidding, and OpenAI says both models will stay.

Measurement arrived with it: a pixel and a Conversions API that can record purchases, sign-ups and lead submissions after someone interacts with an ad. Reporting is aggregated. Advertisers do not see the conversation that produced the impression.

Agency holding companies Dentsu, Omnicom, Publicis and WPP are named partners, alongside ad tech firms Adobe, Criteo, Kargo, Pacvue and StackAdapt. If your agency belongs to one of those groups, it may already have access you do not.

What OpenAI has not published, in anything we could open, is a targeting menu. There is no documented keyword list, audience taxonomy or bid landscape. The public description is matching to the current conversation. We would not plan a budget on targeting features we have only seen described in agency blogs, and you should not either.

Google AI Mode ads run through Performance Max and AI Max

Google has no AI Mode campaign type. Ads reach its AI surfaces through campaigns you already run, and the controls are blunt. Google's help page on AI Overviews answers the two questions most advertisers ask with one word each. Can you target the placement? No. Can you opt out? Also no.

For AI Overviews, text, Shopping, local and app ads from Search, Shopping, Performance Max and App campaigns can appear above or below the overview in all 200+ markets where it runs. Inside the overview, eligibility narrows to text and Shopping ads from Search, Shopping and Performance Max campaigns, in English, in 12 countries.

The new formats announced in May 2026

At Google Marketing Live on 20 May 2026, Google announced a set of formats for AI Mode. Conversational Discovery ads use Gemini to write creative that answers a user's specific question. Highlighted Answers place an ad as an item inside the recommendation lists AI Mode generates. Both were described as testing.

Two more were promised for the following months. AI-powered Shopping ads add a Gemini-written explainer to a product. Business Agent for Leads puts a brand agent inside the ad behind a Chat button. Every new format carries what Google calls an independent AI explainer beside the advertiser's creative, and every one stays labelled Sponsored.

Direct Offers is the commerce piece. It has been in pilot since January 2026 and now surfaces inside the AI Mode response. It comes with native checkout for merchants on the Universal Commerce Protocol and travel offers from partners such as Booking and Expedia. Google's advice for all of it is the same: set campaigns up with Performance Max and AI Max. What that trade costs in keyword control is covered in our breakdown of the controls AI Max removes.

The exclusion list that hits B2B hardest

Google does not show ads in AI Overviews for sensitive verticals, naming adult, alcohol, gambling, finance, healthcare and politics. For a B2B software company selling into finance or healthcare, that is not a footnote. Fintech, insurance technology and health software advertisers should assume the AI Overview placement is closed to them until Google says otherwise.

The page covers AI Overviews only. We found no equivalent Google Ads help article for AI Mode. The topic page that lists help articles on ads in AI experiences links to one article, the AI Overviews one. Eligibility, country lists and exclusions for AI Mode formats are therefore whatever Google's blog post says, which is less than you need.

Copilot ads arrive through Microsoft Advertising automatically

Microsoft's position is the most direct of the three. Its help article states that all eligible campaign and ad types are automatically opted in to serving in Copilot, and advertisers cannot opt out. The rule of thumb in its FAQ is simple: if an ad can serve on Bing, it can serve in Copilot.

Eligible types are multimedia ads, product ads from Shopping and Performance Max, search ads that carry a logo extension or automated business logo, and property promotion and tours ads. Ads are assembled from your existing text and image assets. Negative keywords still apply, and ad copy is not truncated.

The logo rule most accounts miss

One condition is easy to overlook. A search ad must include a logo to be eligible in Copilot. A B2B account running plain responsive search ads without a business logo is, on Microsoft's own wording, not eligible for the Copilot placement at all.

That is the cheapest lever in this whole comparison. Adding a logo extension costs nothing and decides whether you are in or out. Beyond that, Microsoft's richer formats remain narrow. Showroom ads were announced in March 2025 as a pilot with select clients. Copilot Checkout lets shoppers buy inside Copilot, but only English-language merchants selling to US buyers in US dollars are eligible, through Stripe and PayPal, and Microsoft takes no commission today.

Microsoft's case for the placement rests on one number. In March 2025 it said Copilot ad relevance metrics were 25% better than traditional search, citing internal data. No sample, period or metric definition was published. We treat it as a direction of travel from the seller, not as evidence.

AI search advertising compared side by side

Put the three next to each other and the split is clear. One platform offers a separate channel with its own controls and its own measurement. Two offer reach with fewer controls and no separate measurement. Neither model is wrong. They suit different budgets.

ChatGPT rules are drawn from trade press coverage of OpenAI's January and May 2026 announcements, because openai.com refused automated access during research. Google rows cover AI Overviews from its help page and AI Mode from its 20 May 2026 blog post.

What B2B advertisers can measure in AI search advertising today

This is where the documentation is most useful, because it is most honest. Google's help page says it currently doesn't offer segmented reporting when ads show within Search AI Overviews, and those ads are reported as Top Ads. Microsoft says it does not currently support specific metrics for ads served in Copilot.

So on 2 of the 3 surfaces, an ad served inside an AI answer is indistinguishable in your reports from an ordinary top-of-page search ad. You can see that your campaign spent money and converted. You cannot see how much of it happened inside the answer.

What you can know

On ChatGPT you can know a lot. Impressions, clicks and pixel or API conversions belong to one channel, so a lead form fill traces back to a ChatGPT ad without guesswork. That is the strongest argument for testing it, and it is the reason we would start there despite its smaller targeting menu.

On Google and Microsoft you can know the totals and the direction. If AI Overview and Copilot placements are lifting or dragging your search results, it shows up in campaign-level cost per lead over time. It will not show up as a line item. Our analysis of paid search click-through rates under AI Overviews is the closest public evidence on that blended effect.

What you cannot know yet

Three things are out of reach on every platform. You cannot see which conversation or query wording produced a ChatGPT impression, by design. You cannot isolate AI Mode or AI Overview performance on Google. You cannot isolate Copilot performance on Microsoft.

For B2B the gap is worse than it is for retail. Your sales cycle runs for weeks, and the buyer who saw an ad in an AI answer often converts later through a direct visit or a branded search. That pattern is exactly what dark discovery attribution fails to catch. Last-click reporting will credit your brand campaign, not the AI placement that started the journey.

Where this comparison is weakest

Three limits apply, and the first is the one most likely to make this page wrong within a quarter.

First, the documentation moves faster than anyone can re-read it. Google's AI Mode formats were in test in May. Microsoft's help article was updated in September. OpenAI's buying terms changed between January and May. Any row in the tables above could be stale by the time you act on it. Check the source page, not this post, before you commit budget.

Second, we could not open OpenAI's own pages. Its site refused automated access, so the ChatGPT rules here come from trade press reporting on OpenAI's January and May announcements. We used outlets that quoted the company directly. We did not use the widely repeated launch pricing figures, because we could not trace them to anything OpenAI published.

Third, documentation describes what a platform permits, not what it delivers. A placement can be eligible and still rarely serve. Microsoft says outright there is no guarantee ads will display in Copilot. None of the three publishes how often ads actually appear in answers, so reach on each surface remains unknown to any outside reader.

There is also a case against our main conclusion. Blended reporting may not matter much if Google and Microsoft's AI placements are a small share of impressions. In that world, a campaign-level before-and-after test is good enough, and the separate ChatGPT channel is a measurement luxury rather than an advantage. We think the share will grow, but we cannot show it.

Which search advertising budget to move first

Our recommendation for a B2B founder or marketing director runs against the usual instinct. Do not start with the biggest surface. Start with the one you can measure.

ChatGPT is the right first test because it isolates cleanly. A small, fixed budget with the pixel installed gives you a cost per lead that belongs to AI answer advertising and nothing else. Even a disappointing result is a real number, which is more than the other two will give you.

The audience limit cuts the other way, though. Ads reach only Free and Go users. If your buyers use ChatGPT through an employer's Business or Enterprise plan, they will not see your ad there. A test that underperforms may be telling you about plan mix, not about the format.

On Google and Microsoft the decision is mostly made for you, because the placements are automatic. Your job there is hygiene, not strategy. Add logos so you are eligible in Copilot. Review your negative keywords, because they still apply. Watch campaign-level cost per lead for a shift after AI placements expand, and record the date so you can explain it later. Our coverage of how AI referral traffic converts is a useful baseline for what good looks like.

We would not buy agency packages built around AI answer ad targeting yet. No platform has documented enough targeting control for a specialist to add much beyond what a careful in-house marketer can set up.

Frequently asked questions

Where to start this week

Open your Microsoft Advertising account and check whether your search campaigns carry a logo extension. If they do not, add one. It takes minutes and it is the single setting that decides Copilot eligibility.

Then write down today's date and your current search cost per lead on Google and Microsoft, by campaign. AI placements are already serving inside those campaigns whether you chose them or not. A baseline taken now is the only way you will be able to tell, six months from now, what they changed. If you then want to test ChatGPT, give it a fixed budget and install the pixel first, so the result is yours to read. For the organic side of the same shift, our piece on the metrics that replace clicks in zero-click search sets out what to track alongside paid.

Related reading