8 Places Ads Appear in ChatGPT and Other AI Assistants in 2026

What Are LLM Ads?

LLM ads are sponsored placements that a large language model product shows next to, below, or after its answer, selected from the conversation instead of a keyword. As of October 21, 2026, ChatGPT runs them in 89 countries for more than 100,000 advertisers, and Google, Microsoft, Amazon, and AI app networks sell their own versions.

More than 1.2 billion people use ChatGPT each week, OpenAI said on October 21, and most coverage still describes its ads as banners bolted onto chat. That is wrong on every count that matters. The ad is chosen from the meaning of a multi-turn conversation, it sits in a labeled slot, and ChatGPT's answer itself is never for sale.

Three shapes cover every LLM ad in market today. A sponsored unit below or beside the answer (ChatGPT cards, Copilot ads, Google ads in AI Overviews). A sponsored prompt or recommendation woven into the shopping flow (Amazon Rufus prompts, Google Highlighted Answers, Koah's in-app recommendations). A sponsored conversation that starts after the click (ChatGPT Sponsored Agents, Google Business Agent for Leads).

The infrastructure behind those shapes is broken down in LLM ad infrastructure explained, and the platform-by-platform comparison of formats and costs lives in which AI assistants have ads in 2026. This article maps exactly where each ad lands and how ChatGPT decides which one wins.

Where Do Ads Appear in ChatGPT and Other AI Assistants?

Ads appear in eight places across AI assistants in October 2026: four inside ChatGPT (below the answer, product feed units, image generation, and Sponsored Agent conversations), plus Google AI Overviews and AI Mode, Microsoft Copilot answers, Amazon Rufus prompts, and third-party AI apps that sell inventory through ad networks.

1. Below the ChatGPT Answer

The standard ChatGPT ad is a labeled card below the response, built from six parts: advertiser name, favicon, title, copy, landing page, and image. OpenAI's help center says you see one ad unit below a response, carrying one or several images. The Sprites ChatGPT Ads Library, which had logged 10,341 ads from 4,052 advertisers by its October 15, 2026 update, has seen single cards, two cards from one advertiser, two cards from two advertisers, and a card under a short AI-written heading that ties it to the answer above.

This placement catches buyers mid-comparison: "best CRM for a 12-person sales team," "which running shoes for flat feet." Titles allow 50 characters and copy 100, and the guide to creating ads for ChatGPT covers the specs, while the complete guide to ChatGPT ads covers bidding and setup. Study what already runs in the 50 best ChatGPT ad examples.

2. In ChatGPT Product Feed Ads and Carousels

Product feed ads turn a catalog into cards showing the product image, title, star rating, price, sale price, and brand. Since August 2026 they also render as a horizontal carousel of several products from one feed, labeled "Shop · Sponsored" in the Sprites library. OpenAI's help center calls feed ads "among the strongest-performing ads in our program to date."

Feed products appear only in ads during the beta, not in organic ChatGPT shopping results, and new feed campaigns target at country level only. OpenAI's engineering post adds a detail every retailer must act on: the retrieval index for product feed ads is aware of positive and negative product reviews. Fill in star_rating and review_count. The product catalog guide and the Shopify playbook cover the setup.

3. Inside ChatGPT Image Generation

OpenAI announced on October 5, 2026 a multi-image visual ad format shown during image generation for Free and Go users, labeled and separate from the image being created, with US testing starting later in October with an initial group of advertisers. The help center describes the unit as several images from one advertiser that show while ChatGPT generates the user's image and disappear once it finishes.

This is the inspiration slot. Someone generating "a small Scandinavian living room with a green sofa" is designing a purchase before they ever ask for one. The creative is pure imagery: product in use, product in context, the experience it creates. Build the image library now with the ChatGPT ad image specs guide.

4. Inside a Sponsored Agent Conversation After the Click

Sponsored Agents let a user who clicks a relevant ad start "a clearly labeled conversation with a business-sponsored agent," separate from ChatGPT's answers and from the original chat, OpenAI said on September 16, 2026. The user asks follow-ups, then follows a link to the business's site. The test runs with select US advertisers, and Angi announced it is piloting the format for home services.

The help center confirms the privacy line: if you message an advertiser through an ad, the advertiser sees only the messages you send it directly. OpenAI's October 21 engineering post names "giving users the ability to talk directly to businesses through ads" as the direction for more dynamic formats. Lapis runs Sponsored Agents for clients in OpenAI's alpha, and the Sponsored Agents playbook covers seven ways to use them.

5. In Google AI Overviews and AI Mode

Google shows ads above and below AI Overviews in more than 200 markets, and inside AI Overviews in English in the US and 11 more countries, according to Google Ads Help. Google considers both the query and the AI Overview's content, and blocks in-Overview ads for sensitive verticals such as finance and healthcare.

AI Mode is where Google tests the newer formats: Conversational Discovery ads that Gemini tailors to the user's question, Highlighted Answers that place an ad inside an AI-generated recommendation list, Direct Offers in shopping conversations, and Business Agent for Leads, a brand chat in place of a lead form. Since October 16, Google is also testing ads for users who connect apps through Personal Intelligence, with no direct Gmail or Photos targeting. There is no AI Mode campaign type: eligibility comes from Search, Shopping, Performance Max, and AI Max, which the best AI tools for Google Ads help you tune.

6. Below Microsoft Copilot Answers

Copilot places ads below its organic response, introduced by a short "ad voice" summary that explains how the ads connect to the conversation, and it considers the whole conversation in a session, not only the last prompt, Microsoft said when it launched the design. Microsoft opts every eligible campaign in automatically, with no opt-out.

Eligible formats are Multimedia ads, Product ads from Shopping and Performance Max, Search ads with logo extensions, and property and tours ads. Offer Highlights, launched April 21, 2026, pull differentiators such as free shipping into retail ads. The catch: Microsoft states it does not currently support specific metrics for ads served in Copilot, so you judge this placement on blended Microsoft results.

7. In Amazon Rufus Sponsored Prompts

Sponsored Products prompts and Sponsored Brands prompts reached general availability in the US on March 25, 2026. They appear as suggested questions in shopping results and on product pages, and a click opens a Rufus dialog or answers on the page. Campaigns enroll automatically and bill on CPC.

Amazon said in June 2026 that more than 300 million customers used Rufus in 2025, that nearly 20% of shoppers who interact with a Sponsored Brands prompt continue the conversation about that brand, and that prompts drive a 6% increase in conversions. Existing sponsored ads are also eligible inside Alexa for Shopping, which combines Rufus with Alexa+. The creative here is your product detail page and Brand Store, because Amazon generates prompts from them.

8. Inside Third-Party AI Apps and Agents

AI apps built on top of the big models sell their ad slots through networks. Koah serves action cards, image cards, polls, expandables, and catalogs inside apps such as Liner, Luzia, and DeepAI. Gravity places generated text ads in chat apps, AI search, and coding assistants, and is testing agent-to-agent ads humans never see, the shift covered in the agentic ads guide. Kontext generates native ads in real time for chatbot and companion apps, and Taboola opened its DeeperDive answer engine as an ad network on June 16, 2026.

One correction to the ranking pages: ZeroClick relaunched on July 21, 2026 as a platform for selling to AI agents, not a chat ad network. The full ranking of these networks, with buying models and budgets, is in the 10 best AI ad networks.

How Does ChatGPT Choose Which Ad to Show?

ChatGPT chooses ads with a custom, post-trained frontier LLM that OpenAI calls the orchestrator, which reads the conversation and calls tools for retrieval, ranking, relevance, and auction, according to OpenAI's October 21, 2026 engineering post by Ritendra Datta and Karthik Gopalratnam. The auction, not the LLM, sets the final price and winner.

OpenAI built this because a conversation produces no clean ad request. In a search engine, a query fires and the ad system matches it. In ChatGPT, intent forms over several turns: OpenAI's example starts with a question about training for a race and reaches running shoes several turns later. Acting on the first message retrieves the wrong intent entirely. Here is the pipeline, step by step.

1. ChatGPT opens a stateful ads session. ChatGPT sends the initial request context to the ads server, and the session remembers and updates information across turns and across conversations instead of flattening each message into a standalone query.
2. The orchestrator reads intent. It separates current intent from stale context, interprets references to earlier messages, notices when a user is only thinking out loud, and avoids "treating every mention of a product or service as a request to buy something."
3. It writes retrieval queries. The orchestrator summarizes the conversation into one or more short text queries. A chat about a two-day Yosemite trip, next week's weather, and a shoe brand's new lines becomes queries such as [shoes appropriate for hiking trip] and [new brand product lines good for travel].
4. Retrieval casts a wide net. Those summaries become semantic embeddings, and approximate-nearest-neighbor search pulls a deliberately broad pool of candidates from several indices, including a product feed index that knows positive and negative reviews.
5. Filters remove ineligible ads. Checks cover active campaigns and creatives, locale targeting, policy, remaining budget, pacing, which conversation types the ad fits, and deduplication so one advertiser cannot crowd out others.
6. Scoring narrows the field. The orchestrator calls scoring systems for semantic relevance to the user's goal, predicted engagement (view, click, and conversion), and the ad's historical performance.
7. The auction prices and picks. The orchestrator nominates plausible candidates, and the auction tool applies an admission threshold for relevance and value, pacing, bids, quality-adjusted ranking, pricing, and final selection. OpenAI's help center describes it as a relevance-weighted, second-price auction.
8. Two last gates. The orchestrator blocks an ad that would feel inappropriate in that chat even if it passed policy, and if the ad is not ready when ChatGPT's answer is, it times out. OpenAI never delays the answer to wait for an ad.
9. The ad renders with a neutral preamble. The ads system returns the creative plus a short model-generated introduction explaining how the ad relates to the conversation, and the frontend shows both below the response, labeled as sponsored.
10. Every outcome trains the system. Logs record whether a turn was eligible, whether retrieval found candidates, and whether the ad earned an interaction. OpenAI's strongest models evaluate delivery and train the smaller models, and clicks, hides, surveys, and advertiser feedback steer the direction.

The single most important line in the post: "Returning no ad is an ordinary, successful outcome." Impressions in ChatGPT are earned by relevance, not bought by bid alone. OpenAI's help center lists the inputs the system weighs: conversation context and intent, your landing page, title, copy, context hints, targeting selections, and, when ads personalization is on, past chats, memory, and how the user has interacted with ads.

How Do You Get Your Ads Picked by ChatGPT's Ad System?

Get picked by matching the orchestrator's summaries, not keywords: build one ad group per distinct buyer intent, write context hints as plain-language situations, ship many distinct angles so historical performance builds fast, keep budget and pacing healthy, and send every ad to a page that matches the conversation.

• One ad group per intent. The orchestrator splits a trip conversation into separate needs for transportation, lodging, and activities. Your account has to mirror that, with one product, theme, or customer need per ad group.
• Hints that read like summaries. Retrieval runs on short text summaries turned into embeddings, so hints written as "What, Who, When" situations land closer than keyword lists. The context hints guide has the method, and the context hints accuracy study has the evidence.
• Creative volume for coverage. Deduplication caps how many of your ads show per conversation, so the way to win more conversations is more distinct ad groups and angles. The creative volume guide sizes the library.
• Budget that survives the filter. Budget and pacing checks remove you before relevance is ever scored. A daily budget is a seven-day average capped at 2x per day and 7x per week, so a weekly allowance spent by Wednesday means no ads until Sunday.
• Reviews in the feed. The feed index is aware of review sentiment, so products without ratings start behind.
• A page that matches. Your landing page is an explicit selection signal, and OpenAI's crawler must reach it. The landing page guide covers the requirements, and targeting options covers the rest of the controls.

Do Ads Influence ChatGPT's Answers?

No. OpenAI's October 21 engineering post says ad selection runs in a separate service, with separate models and separate code, alongside answer generation. ChatGPT has no dependency on the ads system, so an ads outage leaves answers untouched, and the help center states advertisers cannot shape, rank, or alter ChatGPT's responses.

The separation runs in both directions. By default ChatGPT cannot see the ads shown to you; it learns about an ad only if you choose "Ask ChatGPT" on that ad, and advertisers cannot influence its responses about ads you share. Advertisers receive aggregated performance data such as views and clicks, never your chats, memories, name, email, or precise location.

That design is why the two ways into ChatGPT never collapse into one. Paid placement buys the labeled slot below the answer. Organic recommendation inside the answer is earned through content, reviews, and third-party mentions. Run both: paid ads vs organic mentions in ChatGPT compares them, and 15 ways to get your business recommended by ChatGPT covers the organic half.

Where Do Ads Never Appear?

Ads never appear for ChatGPT Plus, Pro, Business, Enterprise, or Edu accounts, for accounts identified as under 18, in Temporary Chats, or near sensitive topics such as personal health, mental health, and politics. Claude, Perplexity, and the standalone Gemini app carry no ads at all as of October 21, 2026.

• ChatGPT plan exclusions. OpenAI's October 21 announcement repeats that Plus, Pro, and Enterprise stay ad-free, and the help center adds Business and Edu. Free users can also switch to an ad-free experience with lower message limits.
• Logged-out and regional limits. Logged-out ChatGPT shows only all-audience ads, and personalized ads are not initially available in the EEA or Switzerland, where ads run on the current conversation alone.
• No political ads. OpenAI does not allow political advertising in ChatGPT, and regulated categories such as health and finance run only for advertisers that meet strict eligibility.
• Google's exclusions. Ads inside AI Overviews skip sensitive verticals including adult, alcohol, gambling, finance, healthcare, and politics.
• Ad-free assistants. Anthropic said on February 4, 2026 that "Claude will remain ad-free," Perplexity phased out its ads in February 2026, and Google keeps the Gemini app ad-free while it monetizes AI Mode.

The trust argument behind each choice is laid out in does Claude have ads.

The AI Ad Placement Map

The AI Ad Placement Map lists all 8 placements by surface, the moment that triggers the ad, the funnel stage it catches, the format, how you buy it, and the creative it needs. Use it to decide which placements your next brief has to cover and which assets to produce first.

Most placements on this map ignore exact keywords: ChatGPT reads conversation summaries, Google's AI Mode formats generate creative from your assets, and Koah and Kontext match context. The shared input is a clear description of who your offer is for and when it helps.

Which Placement Fits Which Funnel Stage?

Each placement catches a different buyer stage: image generation catches inspiration, ChatGPT cards and Copilot catch consideration, Sponsored Agents and Google's agent formats catch evaluation, and product feed units, Rufus prompts, and Offer Highlights catch the decision. Buy the stages your customers actually pass through in AI.

The consideration layer is where ChatGPT dominates, because that is what people use it for. OpenAI's October 21 announcement describes its own example: someone training for a first marathon starts with meal ideas for race week, narrows by diet and cooking time, and an ad introduces a grocery delivery service they had not considered. That is the mid-funnel moment search never saw, which is why ChatGPT Ads ranks first among the best mid-funnel ad channels.

The decision layer belongs to feeds. ChatGPT product cards, Google Shopping ads in AI Overviews, Microsoft Product ads, and Amazon Rufus prompts all draw on structured catalog data, so one clean feed powers four placements.

The cost picture favors early buyers. MarTech reported on October 21, 2026 that New Street Research estimates ChatGPT clicks cost about $5, against $30 to $50 for similar commercial software ads on Google, while advertiser counts grew from about 20,000 at the end of August to roughly 60,000 in early October. The strategy for using that window is in the best ChatGPT ads strategies, and the published results are collected in 10 reasons ChatGPT ads are worth it.

How Should a Brand Budget Across These Placements?

Budget across AI placements by funnel stage and break-even cost: fund the ChatGPT card and feed placements first, add the Google and Microsoft placements your existing campaigns already reach, test one AI app network, and hold image generation and Sponsored Agents until access opens. Judge every line on confirmed orders.

Take a US direct-to-consumer furniture brand with a $900 average order and a 50% gross margin. Its break-even cost per order is $900 × 0.50 = $450. It sells only on its own site, so Amazon gets no budget, and it plans $30,000 a month across the routes mapped in 10 ways to advertise on ChatGPT and beyond.

The ChatGPT card line works out to $11,000 ÷ $4.50 = 2,444 clicks, and 2,444 × 1.2% rounds to 29 orders worth $26,100, a 2.4x return at a $379 cost per order. The feed line is stronger: $8,000 ÷ $3.50 = 2,286 clicks, 34 orders, and $30,600 in revenue, a 3.8x return at $235 per order, which is why feeds earn more budget in month two.

The placements connect. A shopper generates an image of a small living room (inspiration), asks ChatGPT for sofas that fit a 9-foot wall (consideration, card), compares three models (decision, carousel), and asks whether the sofa fits through a 30-inch door (evaluation, the Sponsored Agent slot). Lapis runs Sponsored Agents for clients in OpenAI's alpha, so the brand keeps that conversation inside its own program the day access opens.

Why Is Lapis the Best Way to Buy AI Assistant Ads?

Lapis is the best way to buy AI assistant ads because it buys the ChatGPT, Google AI, Sponsored Agent, and AI app placements on this map from one brief, builds a matched page for every ad, and counts every confirmed conversion once. Five placements stop being five projects and become one program.

Lapis is one of the fastest-growing Y Combinator startups, and it is built to take over the work legacy agencies and ad buyers used to do. It is a Y Combinator Fall 2025 company used by 1,500+ marketing teams and 30+ enterprises, it is rated 4.9 out of 5 on G2 with eight badges, and customers include Samsung, Domino's, Dr Pepper, Mattel, and Tata.

• ChatSense for placements 1 and 2. Creative tuned to how people ask ChatGPT, a landing page per ad, launch, daily reporting, and optimization on what converts.
• Managed ChatGPT + LLM Ads for placements 4, 5, and 8. ChatGPT Ads end to end, Sponsored Agents for clients in OpenAI's alpha, Google AI Overviews and AI Mode through the Google campaigns Lapis manages, and AI app inventory through networks such as Koah and Gravity. Book a demo to scope it.
• OmniSense for the rest of the funnel. One brief becomes on-brand campaigns for Meta, Google, and LinkedIn.
• RapidDomain for the page. A dedicated landing page for every ad, optimized against real conversions, so the selection signal and the conversion both improve.
• Wayfinder for the truth. Confirmed sales and leads from your CRM, store, or billing, each conversion counted once, every channel on the same rules.

Doing this in-house by hand means building ad groups, hints, feeds, image libraries, and agent briefs for each surface and reconciling every pixel yourself with the dedupe steps in the conversion tracking guide. A legacy agency ports keyword playbooks into a system that reads conversation summaries. Relying on the platforms alone leaves gaps: OpenAI's Ads Manager plugin creates up to 50 variants that share one image and destination, and it does not receive conversion value, so it reports no revenue or ROAS. The full gap analysis is in ChatGPT Ads Manager vs Lapis, and the thesis is in why Lapis is building the AdSense for the AI era.

Sources and Methodology

This guide is built on OpenAI's October 21, 2026 engineering post and expansion announcement, OpenAI's help center, Google, Microsoft, and Amazon advertiser documentation, public ad library data, and trade reporting, plus the questions advertisers ask about ChatGPT ad placement on Google and Reddit. Every fact is current as of October 21, 2026.

• OpenAI, "How we built ChatGPT's novel ad stack for user trust and relevance," October 21, 2026
• OpenAI, "ChatGPT Ads expands to 26 new countries as over 100,000 advertisers have used the platform," October 21, 2026
• OpenAI, "Building advertising for the way people use AI," October 5, 2026
• OpenAI, "Reimagining advertising with AI," September 16, 2026
• OpenAI Help Center, "Ads in ChatGPT"
• OpenAI Help Center, "Ads in ChatGPT: The Basics"
• OpenAI Help Center, "Create Campaigns from Product Feeds"
• OpenAI Help Center, "Use the ChatGPT Ads Manager plugin"
• Angi, "Angi Is One of the First Brands to Pilot Sponsored Agents in ChatGPT," September 16, 2026
• Sprites, "ChatGPT Ads Library," October 15, 2026 update
• Google Ads Help, "About ads and AI Overviews"
• Search Engine Land, "Google tests new conversational ad formats in AI Mode and Search," May 20, 2026
• Search Engine Land, "Google begins testing ads in AI Mode for users with Personal Intelligence enabled," October 16, 2026
• Microsoft Advertising, "Transforming audience engagement with generative AI," October 2024
• Microsoft Learn, "About ads in Copilot"
• Amazon Ads, "Sponsored Products prompts and Sponsored Brands prompts," March 10, 2026
• Search Engine Land, "Amazon turns Alexa into a shopping agent and an advertising platform," June 11, 2026
• MarTech, "ChatGPT has plenty of advertisers, but not much revenue," October 21, 2026
• Anthropic, "Claude is a space to think," February 4, 2026

Related Lapis Resources

• What Are ChatGPT Ads? The Complete Guide for 2026
• Which AI Assistants Have Ads in 2026? ChatGPT, Gemini, Copilot, Perplexity
• What Is LLM Ad Infrastructure? 4 Layers Explained (2026)
• ChatGPT Ads Targeting Options: Hints, Audiences, Platforms, Location
• How to Turn a Product Catalog Into ChatGPT Ads: 2026 Feed Guide