What Is LLM Ad Infrastructure? 4 Layers Explained (2026)

What Is LLM Ad Infrastructure?

LLM ad infrastructure is the stack of systems that places, targets, creates, and measures ads inside AI assistants. It is the machinery beneath the sponsored unit you see under a ChatGPT answer: the auction that picks the ad, the signals that decide whether it fits the conversation, the tools that produce the creative, and the measurement that ties the impression to a sale.

The clearest way to understand it is to compare it with the two stacks that came before. Search advertising is built around a results page: a query, keywords, an auction for positions, and a trackable click. Social advertising is built around a feed: an interest profile assembled from cookies and pixels, an auction for impressions in an endless scroll, and creative sized for that feed. Both assume a fixed surface, a targeting primitive, and a measurable click.

Advertising inside an assistant breaks all three assumptions. There is no results page and no feed, only a conversation rendered one turn at a time. Targeting starts from the meaning of that conversation rather than a keyword list. And the assistant often shapes a decision before any click happens, so a buyer sees an ad, keeps talking, and converts later. LLM ad infrastructure is what replaces the page, the keyword, and the click.

The surface is enormous and monetizing fast. OpenAI piloted ChatGPT ads in the US starting in February 2026, opened self-serve Ads Manager in May, and announced on August 31 that ChatGPT Ads had reached a $1 billion annualized revenue run rate in less than 200 days, with tens of thousands of advertisers and more than 50 technology and measurement partners. Understanding the infrastructure is now the prerequisite for spending there profitably.

Which AI Assistants Show Ads in 2026?

Three AI surfaces carry ads as of October 23, 2026: ChatGPT, Google's AI Overviews and AI Mode, and Microsoft Copilot. The Gemini app and Claude are ad-free, and Perplexity ended its ad program in February 2026.

The details behind the table matter. Google unveiled two Gemini-built AI Mode formats at Google Marketing Live in May 2026: Conversational Discovery ads tailored to a specific question, and Highlighted Answers that place relevant ads inside AI Mode recommendation lists, both with an independent AI explainer and a Sponsored label. Google keeps the Gemini app ad-free; Business Insider reported in April 2026 that the standalone Gemini experience has remained ad-free while Google focuses on AI Mode. Microsoft serves ads in Copilot from existing campaign assets, and its Copilot ads documentation confirms eligible campaigns opt in automatically. Search Engine Land reported on February 18, 2026 that Perplexity phased out the ads it began testing in 2024. And Anthropic committed on February 4, 2026 that “Claude will remain ad-free.” For a channel-by-channel playbook, see our guide to advertising across AI assistants.

Why Is a New Ad Stack Forming Inside AI Assistants?

A new stack is forming because four things changed at once: the surface became a conversation, creative became generative on both sides of the auction, targeting moved from cookie profiles to context plus first-party data, and attribution moved beyond the last click.

The surface is a conversation. A results page or feed post is a fixed slot you design once. A conversation is different every time. The ad has to fit a specific exchange, appear when it is relevant, and read as a useful next step. That requires creative produced for context, not just for placement.

Creative is generative on both sides. Each assistant renders ads in its own format, and conversational relevance rewards specificity, so the number of creatives a campaign needs explodes. Platforms responded by building generation into their own ad managers. Advertisers now need a creative system that stays on brand across every one of those channels at once.

Targeting is contextual plus first-party. Assistants match ads to the meaning of the live conversation, then layer on the advertiser's own customer lists, location, device, and exclusions. Third-party identity graphs play no role.

Attribution reaches past the click. People often see an ad, keep working through a decision, and convert later. OpenAI now reports view-through conversions by default and partners with incrementality firms to measure lift. Measurement had to be rebuilt for that path, and it has been, as Layer 4 shows.

What Are the Four Layers of LLM Ad Infrastructure?

The four layers are demand and auction, context and targeting, creative generation, and measurement and attribution. Platforms now supply a version of every layer for their own channel; advertisers own the brand, the customer data, the landing pages, and the cross-channel truth.

The layer names match our guide to building an LLM advertising stack, which turns this model into a build plan. The rest of this article explains how each layer works in October 2026.

How Does Layer 1, Demand and Auction, Decide Which Ad Shows?

Layer 1 runs a relevance-weighted auction: the platform multiplies your bid by how well your ad fits the conversation, so a precise ad at a lower bid beats a generic ad at a higher one. OpenAI's ChatGPT Ads basics describe it as “a relevance-weighted, second-price auction.”

Objectives and bidding. ChatGPT Ads offers Views (CPM), Clicks (CPC), and Conversions, which bills per valid click (oCPC) or per thousand impressions (oCPM) while optimizing toward one standard conversion event. The Maximize results strategy sets bids automatically and is the default for eligible new ad groups. For Clicks campaigns, OpenAI recommends starting with a maximum bid of $3 to $5 per click. On August 31 OpenAI said CPC and outcome-optimized bidding now account for the majority of campaigns.

Formats. The standard unit sits below a ChatGPT response with the advertiser name, favicon, title, copy, image, and landing page. Product-feed campaigns generate catalog ads with images, prices, and ratings. Sponsored Agents, announced September 16, let a user start a labeled conversation with a business-sponsored agent after clicking an ad, and are testing with select US advertisers. A new visual format that appears during image generation begins testing in the US in late October 2026.

Buying routes. Self-serve Ads Manager is open to businesses in 89 countries. Larger advertisers work with the OpenAI Ads team or agency and technology partners. Amazon opened a managed Amazon DSP pilot for ChatGPT Ads on September 10 for select US advertisers, with OpenAI still controlling delivery.

Google and Microsoft run their AI ad inventory through the campaign types advertisers already use. Google's new AI Mode formats run on AI Max and Performance Max. Microsoft automatically builds Copilot ads from existing campaign text and images and names Performance Max as the best way to increase Copilot exposure. The strategic rule holds across all three: relevance multiplies the bid, so specificity beats spend. For a deeper look at buying on the largest marketplace, read our complete guide to ChatGPT Ads.

How Does Layer 2, Context and Targeting, Replace Keywords and Cookies?

Layer 2 matches ads to the meaning of the live conversation, then narrows delivery with the advertiser's own data and platform controls. Context replaces the keyword; first-party lists and settings replace the cookie profile.

Context. OpenAI's ads system weighs the context and intent of the current conversation, the ad's landing page, title, and copy, and advertiser-provided context hints. Hints are set per ad group, describe the conversations where an offer fits, and guide matching rather than act as exact-match keywords. OpenAI's context hint guide shows the standard with “cushioned everyday running shoes for beginners training for their first 5K,” and states that hints cannot enforce geographic limits, schedules, or exclusions. Delivery controls do that work.

Targeting. The identity half of this layer is now real. When a user has ads personalization enabled, OpenAI also uses “select signals from a user's broader ChatGPT experience.” Advertisers add:

• Custom audiences from customer and prospect lists. Inclusion and bid multipliers (0.1x to 10x) need 25,000 matched users, exclusions need none, and EEA and Swiss campaigns cannot use them, per OpenAI's custom audiences guide.
• Location at country level plus supported states, cities, markets, and postal codes for standard campaigns.
• Platform targeting across Android app, Android web, desktop web, iOS app, and iOS web.
• Negative Phrases for qualifying advertisers, a hard exclusion that Adthena documented as up to 100 literal phrases matched against the user's message.

Google and Microsoft follow the same pattern with their own tools: Google's AI Mode formats run on AI Max and Performance Max matching, and Microsoft builds Copilot ads from the assets and settings already in your Microsoft Advertising campaigns. For every ChatGPT targeting control in detail, see our ChatGPT Ads targeting options guide.

Who Owns Layer 3, Creative Generation, Now That Platforms Make Ads Too?

Platforms now generate creative inside their own ad managers, so Layer 3 is shared: each platform makes a fast first ad for its own channel, and the advertiser owns the brand system, the angles, and the cross-channel program. That division is the most important change to this stack in 2026.

OpenAI moved first. On September 16, 2026 it rolled out four creative capabilities:

• AI creative suggestions in Ads Manager. When drafting an ad, advertisers receive suggested copy and imagery based on their landing page and campaign objective, then review and edit before adding them.
• The Ads Manager plugin in ChatGPT. Advertisers create, update, and analyze campaigns with natural-language prompts, turn a website or brief into a campaign, and get recommendations on what to do next.
• AI-powered text customization. An opt-in setting adapts existing headlines and descriptions to the context of each conversation and translates copy into the user's preferred language.
• Sponsored Agents. The advertiser's own agent continues the conversation after the click, answering questions until the user is ready to visit the site.

Product-feed campaigns already generate catalog ads at scale, and the ChatGPT Ads apps for Shopify and HubSpot let merchants and CRM users create and manage ChatGPT ads from inside the tools they already run. Meta is moving the same way: Marketing Dive reported on October 8, 2026 that Meta is adding agentic capacity so its Ads Manager business assistant can generate creative elements, update targeting, and change budgets through conversational prompts. Google builds AI explainers with Gemini directly into its new AI Mode formats.

Every one of these tools is built for one channel and starts from one landing page. That is exactly why the advertiser layer matters more now, not less. OpenAI's own ad creation guide asks for “a high volume and diverse set of ads,” with each variation introducing “a different angle rather than repeating the same message.” The auction rewards breadth of angles, and angles come from strategy, not from rewording one landing page. Platform tools produce a good first ad for their own auction. They do not give you a brand-trained creative system with a distinct visual per buyer angle, the same brand across ChatGPT, Meta, Google, LinkedIn, and Reddit, an experiment design that decides which angles to test next, or landing pages built for each angle.

Lapis owns that layer. OmniSense builds on-brand, correctly sized campaigns, audience versions, and variations across Meta, Google, LinkedIn, and Reddit. ChatSense produces 1,000+ ChatGPT ads tuned to how people ask ChatGPT and generates each next run from results. Both start with brand intelligence from your website URL (logo, colors, typography, product images, and voice), work in 100+ languages, and use Campaign Studio for natural-language edits. The same autonomous, goal-driven shift is reshaping the whole workflow, as our agentic ads guide explains, and creative is where volume pressure lands first.

What Does Layer 4, Measurement and Attribution, Look Like in 2026?

Layer 4 now has real tooling: Pixel and Conversions API signals, click-through and view-through attribution windows, third-party attribution partners, and geo-incrementality experiments, all expanded by OpenAI on October 5, 2026. The open problem is cross-channel truth, and that is the job Wayfinder does.

Signals. The OpenAI Pixel and Conversions API send purchases, leads, installs, and sign-ups back to ChatGPT Ads. The Pixel captures the oppref click reference, and automatic advanced matching hashes form data in the browser to recover conversions when a click identifier is missing. Event Quality Score grades the signal so advertisers can fix gaps.

Attribution. According to OpenAI's More ways to measure update, advertisers choose a click-through window and can add a one-day view-through window. In an early analysis of global campaigns, 52.7% of eligible one-day view-through conversions happened within an hour of the ad impression.

Partners. Hightouch, Tealium, and LiveRamp send conversion data from existing systems. AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge, and Tenjin support Conversions API, reporting, and click attribution. Fospha, Measured, and INCRMNTAL cover full-funnel measurement. OpenAI is exploring geo-based incrementality experiments with Haus, Measured, and WorkMagic, testing brand resonance measurement with Kantar and Cint, and developing brand suitability pilots with DoubleVerify and IAS.

The early proof points are strong. DV Rockerbox measured WeightWatchers' attributed CPA on ChatGPT Ads at 15.3% below its blended paid-search benchmark. WorkMagic found statistically significant lift for Dose, with 67% of incremental purchases from net-new customers. Triple Whale found 93% of Portland Leather's ChatGPT Ads visitors were new, per OpenAI's October 5 announcement.

What platform measurement cannot do is referee between platforms. OpenAI's conversion measurement guide says Ads Manager, your analytics tool, and other ad platforms report different totals because of different attribution windows and deduplication. Every platform grades its own homework. Wayfinder is the shared measurement layer for OmniSense and ChatSense: it connects campaign results to confirmed sales and leads from your CRM, store, or billing system, counts each conversion once, and compares every channel on the same attribution rules. Setup for the ChatGPT side lives in our Pixel and Conversions API guide.

Who Plays Where in the LLM Ad Stack?

Platforms play vertically, going deep inside their own channel at every layer. Partners plug into single layers. Lapis plays horizontally, running the advertiser's side of every layer across every channel.

The map makes the buying decision simple. You plug into each platform for inventory. You connect partners where you need a specialist. And you run one advertiser layer across all of it so the brand, the experiments, the pages, and the numbers stay consistent from channel to channel.

What Should Advertisers Build, Buy, or Hand to Lapis?

Buy inventory from the platforms, own your inputs, and hand the cross-channel program to Lapis. Never build what a platform owns, and never hand-produce what automation scales.

Demand and auction: adopt the platforms. You cannot build an ad marketplace inside someone else's model. Learn the auction rules, pick objectives deliberately, and let Lapis agents launch and reallocate budget inside the guardrails you set.

Context and targeting: own the inputs. The matching engine belongs to the platform. The context hints, customer lists, and landing pages that feed it are yours, and they are where advertiser skill compounds. ChatSense designs buyer-intent experiments; RapidDomain builds the matching pages.

Creative generation: use platform tools for a quick start, run the program on Lapis. Platform suggestions are a fine first draft for one channel. A paid growth program needs brand-trained creative, distinct angles, and next-run generation across every channel, and building that in house means competing with dedicated platforms on models and format coverage.

Measurement: adopt the signals, own the truth. Install every platform's Pixel and Conversions API and use partners where they help. Then make Wayfinder the single, deduplicated source of truth for confirmed revenue.

Start in this order: tracking first, then intent design and matched pages, then creative volume, then scale. For the full build plan, read how to build an LLM advertising stack. To see why platform tooling makes an independent advertiser layer more valuable, read will OpenAI build what Lapis does and the feature comparison in ChatGPT Ads alone vs Lapis.

Why Is Lapis the Best Way to Operate Across the LLM Ad Stack?

Lapis is the best way to operate across the LLM ad stack because it runs the advertiser's side of all four layers on every major channel at once. Lapis is the #1 AI ad generator and the operating system for paid growth.

• ChatSense (Layers 1 to 3 on ChatGPT and LLMs). Lapis agents and a dedicated strategist run ChatGPT and third-party LLM ads, including open-source chatbots: buyer-intent experiment design, 1,000+ ChatGPT ads, campaign structure, autonomous deployment, next-run generation, and daily reporting.
• OmniSense (Layers 1 and 3 across paid social and search). On-brand campaigns, audience versions, and variations for Meta, Google, LinkedIn, and Reddit, with directional forecasts of impressions, clicks, CTR, and leads, plus competitor tracking through Brand Radar and a 10,000+ ad library.
• RapidDomain (Layer 2). A dedicated, ad-matched landing page for every ad and audience, optimized on real conversions.
• Wayfinder (Layer 4). Confirmed sales and leads from your CRM, store, or billing system, each conversion counted once, every channel compared on the same attribution rules.

OpenAI gives every advertiser a fast first ad for one channel. Lapis runs the whole program: brand-trained creative with a distinct visual per angle, experiment design and next-run generation, RapidDomain pages for every ad, OmniSense across Meta, Google, LinkedIn, and Reddit, Wayfinder counting each conversion once across every channel, and a dedicated strategist. Lapis is the only company that runs creative, launch, matched landing pages, ChatGPT ads, and confirmed-conversion measurement in one system.

Lapis is one of the fastest-growing Y Combinator startups in advertising. It is part of Y Combinator's Fall 2025 batch, used by 1,500+ marketing teams and 30+ enterprises, and rated 4.9 out of 5 on G2, with eight Summer 2026 G2 badges including High Performer, Best Results, and Easiest to Use (Mid-Market). Samsung, Domino's, Dr Pepper, Mattel, Tata, Tupperware, Amway, and Bata use Lapis, along with high-growth startups including Unusual Whales, Duradry, Lightpage, and Respan. Lapis has generated 10,000+ ad campaigns across 30+ industries, and the team comes from MIT, Columbia, Duke, and Google.

ChatGPT and Reddit campaigns run on Lapis managed plans with custom pricing. Book a demo to see Lapis run all four layers for your brand. Teams starting with Google, Meta, and LinkedIn creative can begin on the $99 per month Basic plan at pricing.

Sources and Methodology

Facts and pricing checked as of October 23, 2026. Platform mechanics come from OpenAI, Google, Microsoft, and Anthropic primary sources; market events come from the named trade publications. Lapis product details come from Lapis.

• OpenAI: Building advertising for the way people use AI (October 5, 2026)
• OpenAI Ads: More ways to measure ChatGPT Ads (October 5, 2026)
• OpenAI: Reimagining advertising with AI (September 16, 2026)
• OpenAI: A milestone in expanding access to AI (August 31, 2026)
• OpenAI Help Center: Ads in ChatGPT, The Basics
• OpenAI Help Center: Ads Manager Availability
• OpenAI Help Center: Set up Custom Audiences
• OpenAI Help Center: Conversion Measurement
• OpenAI Help Center: Write Context Hints for ChatGPT Ads
• OpenAI Help Center: Create Ads for ChatGPT Ads
• Google: New ad formats built with Gemini coming to Google Search (May 20, 2026)
• Business Insider: Google says it is open to putting ads in Gemini (April 2026)
• Microsoft Learn: About ads in Copilot
• Search Engine Land: Perplexity stops testing advertising (February 18, 2026)
• Anthropic: Claude is a space to think (February 4, 2026)
• Search Engine Land: Amazon pilots ChatGPT Ads through its DSP (September 10, 2026)
• Marketing Dive: Meta introduces new AI-powered tools at Advertising Week (October 8, 2026)
• Adthena: ChatGPT Ads quietly adds negative phrases (August 21, 2026)
• Y Combinator: Lapis company profile
• G2: Lapis reviews