For twenty years, a search ad lived beside the answer. In 2026, it moved inside it.
Google now runs sponsored placements in the middle of AI Mode responses. OpenAI opened a self-serve Ads Manager for ChatGPT in May and has been expanding it market by market since. Microsoft folds your existing search ads into Copilot answers whether you asked it to or not. Perplexity tried ads, then quit.
That is a lot of change in one year, and most of it happened without a clear playbook. Media teams are being asked to budget for 2027 against placements that did not exist at this time last year, with reporting that is thinner than anything they are used to.
This post is the practical read. What you can actually buy today, how the placements get chosen, what the reporting looks like, how to write for them, and how much of next year’s plan should go there. Where a claim comes from a platform’s own marketing, we say so.
What you can actually buy today
As of October 2026, there are three places a brand can pay to appear inside an AI-generated answer, and one notable place it cannot.
Surface | Status | How you buy it | Where the ad sits |
|---|---|---|---|
Live in the US, new formats in test | Existing Google Ads campaigns: Performance Max, Shopping, AI Max for Search, and Search with broad match | Above, below, or within the AI answer; new formats sit inside the response | |
Live in the US and a growing list of markets, self-serve since May 5, 2026 | OpenAI Ads Manager at ads.openai.com, or through agency and ad tech partners | Labeled sponsored card below the response, visually separate from the answer | |
Microsoft Copilot | Live | Existing Microsoft Advertising campaigns, opted in by default | Generated from your existing search assets inside Copilot responses |
Perplexity | Closed | Not buyable | None since February 2026 |
Google is the easy on-ramp. There is no AI Mode ads manager and no opt-in box. Google confirmed in May 2025 that advertisers running Performance Max, Shopping, and Search campaigns with broad match, including AI Max for Search, are eligible to appear in AI Overviews and AI Mode. If you run those campaign types, you are probably already serving there and may not know it.
At Google Marketing Live on May 20, 2026, Google announced two new AI Mode formats built with Gemini: Conversational Discovery ads, where Gemini writes creative tailored to the specific question asked, and Highlighted Answers, where a qualifying ad can appear inside a list of recommendations the AI generates. Both carry an independent AI explainer alongside the advertiser’s creative, and both are labeled Sponsored. AI-powered Shopping ads and a chat-based Business Agent for Leads are coming to standard Search in the coming months.
ChatGPT is the first true native buy. OpenAI started testing ads on February 9, 2026 for logged-in adult users on the Free and Go tiers in the US. The pilot opened with a minimum commitment reported at $200,000 or more, which dropped to $50,000 when an early ads manager arrived in April. On May 5, OpenAI opened a self-serve Ads Manager to US advertisers with CPC bidding alongside CPM, a measurement pixel, and a Conversions API. Partners include Criteo, Adobe, StackAdapt, Pacvue, and Kargo, plus the major holding companies. Rollouts have since reached Australia, New Zealand, the UK, Japan, South Korea, and India.
The structural catch: Plus, Pro, Team, and Enterprise subscribers do not see ads. If your buyers are the kind of people who pay for ChatGPT, the placement may miss them.
Copilot is automatic. Microsoft generates Copilot ads from assets already in Microsoft Advertising, including Responsive Search ads, Product ads, Multimedia ads, and Performance Max. According to reporting on Microsoft’s documentation, eligible campaigns are opted in by default and cannot opt out. There is no guarantee any given ad appears.
Perplexity is the cautionary tale. It launched sponsored follow-up questions in late 2024, stopped taking new advertisers in late 2025, and stepped away from ads entirely in February 2026. The only way into a Perplexity answer now is to be cited organically. Any plan that still has a Perplexity line item is working from last year’s facts.
How the placements get chosen
You do not target an AI answer. You make yourself eligible for one, and the model decides.
On Google, the ad has to be relevant to both the user’s query and the content of the AI response. Google Ads Help clarified in 2025 that only broad match keywords can serve within an AI Overview; exact match can serve above or below but not inside. Ads can appear in one position or the other, never both at once. In practice that means a tightly managed exact-match account can be nearly invisible inside the answer while a broad-match or AI Max account shows up constantly.
On ChatGPT, there are no keywords at all. OpenAI’s own guidance for advertisers says the system weighs conversational intent plus, when the user has personalization on, select signals from their broader ChatGPT use. OpenAI’s advice is to build many distinct title and copy variations per offering so the system has more chances to find a fit. Think of it as feeding a matching engine, not bidding on a term.
In both cases the model is also writing or framing part of the ad. Google’s Conversational Discovery format generates creative tailored to the question. The Highlighted Answers format places your ad inside an AI-written list. Google has floated AI Brief as a way to set brand guardrails on that generated copy, but agencies at GML were open about not yet knowing how well it holds.
What reporting looks like right now
Thin, and that is the honest word for it.
Google does not break out AI Overview or AI Mode as a placement in standard reporting. The ads run inside your existing campaigns and their impressions blend into the search totals. Third-party tools such as Adthena have built AI Overview tracking to fill the gap, which tells you something about how little the platform gives you directly. Google’s own September update added an AI Max reporting feature, which is worth checking if you run that campaign type.
OpenAI started further back. During the early pilot, advertisers received weekly CSV files with no dashboard login. The Ads Manager now shows spend, impressions, clicks, and CPC on a date range, and the May update added a pixel and Conversions API. Advertisers receive aggregate data only; no conversation content reaches them, which OpenAI positions as a privacy feature and which also limits what you can learn about why an ad served.
One agency that ran a live ChatGPT campaign reported a mismatch between OpenAI’s numbers and their Google Analytics data large enough to pause the test. Treat that as one data point, not a verdict, but go in expecting to reconcile.
Writing for the answer, not the results page
The ad that wins inside an AI answer is the one that reads like the next sentence of the answer.
A classic search ad competes for attention against nine other blue links. An ad inside AI Mode or below a ChatGPT response sits next to a paragraph the user just read and trusted. Punchy, vague, or hype-driven copy looks wrong in that spot. OpenAI’s guidance says it plainly: clear, specific, benefit-focused copy that explains who the product is for and when it helps outperforms a catchy line or a generic call to action.
Five things that change in practice:
- Write to the question, not the keyword. AI answers trigger on long, conversational, situational queries. Your copy should answer the situation. “Home dialysis training in under two weeks” beats “Leading home hemodialysis provider.”
- Build volume and variety. OpenAI explicitly asks for many title and copy variations per offering, each from a different angle, so the matching engine has more to work with. One tight ad group with three headlines will underserve.
- Send people to the page that finishes the thought. Both platforms reward a landing page that matches the ad and the answer. OpenAI tells advertisers to use product or content pages rather than a homepage, and to add UTM parameters to each creative by hand since automatic tagging is not available.
- Make the landing page crawlable by the AI. OpenAI requires that landing pages not block its OAI-AdsBot and OAI-SearchBot user agents. Google’s Gemini-built formats pull context from your site and feed to write the explainer. If your robots.txt or a bot blocker is shutting these crawlers out, your ads may be ineligible or served with a thin explainer.
- Clean your product data. Google’s new Shopping formats and Direct Offers draw from Merchant Center feeds and uploaded promotions. The explainer Gemini writes is only as good as the attributes you gave it.
For regulated brands there is an extra layer. Google has reportedly confirmed a live test of healthcare ads in AI Mode, and the first phase excludes ads with pinned assets or text disclaimers. That is a problem for any medical device or pharma advertiser whose compliance team requires a fixed disclaimer in the ad. Until that changes, the safe move is to keep required language on the landing page and confirm with counsel that an AI-written explainer next to your ad does not create a claims issue. We would treat that Browser Media recap as a secondary source and verify directly with your Google rep before building a plan on it.
Measuring it without fooling yourself
You will not get clean attribution from these placements in 2026, so set up measurement that does not depend on it.
On Google, isolate the campaign types that can serve in AI. If every campaign is AI Max or broad match, you cannot tell what the AI surfaces added. Keep at least one exact-match control campaign on the same products and compare search term reports. Ask your Google rep for whatever AI Overview or AI Mode impression share they can pull; it may not be in the UI, but it often exists internally.
On ChatGPT, instrument everything yourself. Install the OpenAI pixel, connect the Conversions API, and tag every creative with its own UTM. Then reconcile OpenAI’s click counts against your analytics weekly, because at least one agency has published a meaningful gap. A separate landing page or a dedicated promo code is cheap insurance.
Measure the referral, not just the ad. The bigger effect of AI surfaces may be organic. OpenAI’s ads chief said at Cannes that roughly 20% of ChatGPT queries carry direct commercial intent. Target has reported ChatGPT referral traffic growing about 40% a month. Track referrals from chatgpt.com, copilot.microsoft.com, and Google AI Mode as their own channel in analytics so a lift from the paid test is not mistaken for a lift in free citations, or the reverse.
Be skeptical of platform-funded lift numbers. Google’s claim that 75% of people make faster, more confident decisions with AI Mode comes from a Google-commissioned Ipsos survey of shoppers who already use AI Overviews or AI Mode. Criteo’s finding that LLM referrals convert about 1.5 times other channels comes from Criteo’s own retailer sample and Criteo is an OpenAI ads partner. Neither is wrong on its face, but both come from parties selling the surface.
What to put in the 2027 plan
Fund a test, not a channel. The inventory is real, the audience is real, and the measurement is not ready to justify a standing line item.
Google: audit before you spend. You are likely already serving in AI Overviews and AI Mode through Performance Max, Shopping, AI Max, and broad match. Start by finding out how much, using your rep and a third-party tracker. Then decide whether your copy and feed data are written for a conversational answer. For most accounts that is a creative and data project, not a new budget.
ChatGPT: a bounded pilot in Q1. The self-serve minimum is low enough that a test no longer needs executive sign-off on a six-figure commitment. Set a fixed budget, a 60 to 90 day window, pixel and UTM instrumentation from day one, and a success metric you can actually read, such as qualified leads from tagged URLs rather than platform-reported conversions. Remember the free-and-Go-tier limit when you set expectations about who sees it.
Copilot: check the opt-in, then leave it alone. Your Microsoft Advertising campaigns are already there. Review what is serving and make sure nothing in your search assets reads badly when a model lifts it into an answer.
Perplexity and Claude: organic only. No media line. Put that effort into answer engine optimization and the kind of content these systems cite, which is a separate post we have already written.
For regulated brands, add one gate. Before any AI Mode or ChatGPT spend, get compliance sign-off on the specific formats, confirm whether required disclaimers can appear in the ad unit, and document what the AI-generated explainer can and cannot say next to your name. In healthcare and medical device, that review is the whole project, and it needs to start now to clear a Q1 test.
The cheapest thing you can do this quarter is find out where your ads already are. The second cheapest is to rewrite them for a reader who just finished a paragraph, not one scanning a page.
