Why AI App Publishers Are Routing Around Google and Legacy Networks Entirely

AI app publishers need monetization infrastructure that reads live conversation context, supplies relevant advertiser demand, and serves ads inside their own interfaces. They are routing around Google and legacy networks because page-based targeting and click-based measurement do not map onto an evolving dialogue. Publishers need a native conversational SSP and exchange.

Why the gap between AI publishers and legacy ad infrastructure is structural, not a preference

Google Ad Manager and legacy ad networks serve the medium they were built for, namely observable webpages with identifiable content, ad placements, and measurable clicks. Their strengths depend on those stable inputs. A conversational interface has a thread that changes in meaning as the user adds more information. The relevant unit is not a persistent page or an isolated query but the conversation in progress. Publisher defection therefore reflects a category mismatch, rather than dissatisfaction with an incumbent’s feature set.

A different auction cannot by itself recover intent that the targeting system never reads. Changing the attribution window also cannot show a decision made entirely within an assistant. Targeting, serving, and measurement all need to start from the medium’s actual signals.

How large conversational surfaces have become in terms of user reach

Amazon’s Rufus assistant had hundreds of millions of users in a single year, and the number of monthly users grew sharply year-over-year. Conversational interfaces have grown too large for their revenue model to stay an experiment forever. They carry audiences comparable to major ad-supported properties, but much of their monetization infrastructure is absent or adapted from another medium. That gap makes native ad serving an economic requirement. Publishers need a way to turn activity inside the conversation into revenue without depending on a page visit elsewhere.

How Google's AI Overviews arrangement puts publishers in a coercive position

For content publishers, the pressure begins before any decision to build an AI interface. Google’s AI Overviews uses publisher content to populate answers while reducing the need to visit the source. Bain & Company research cited in the Penske Media filing concluded that a majority of Google searches now end without a click. The publisher can consequently contribute to the answer without receiving the visit that would have generated an ad impression.

Publishers lack a meaningful way to opt out without jeopardizing search visibility, yet no equivalent revenue flows back when an AI answer replaces the visit. By folding conversational discovery into the search results page, Google keeps the ad auction close to answer synthesis. The content source absorbs the loss of traffic while the commercial opportunity remains on the search surface.

How AI crawlers take content without returning proportionate value in referrals

TollBit’s Q1 2025 State of the Bots report puts OpenAI’s scrape-to-referral ratio at 179:1 and Perplexity’s at 369:1. These figures describe a second imbalance, extending beyond Google. AI systems consume publisher content far more frequently than they send users back to it. When a business earns money from ads on source pages, the use of its content no longer reliably gives it a chance to earn revenue.

The ratios expose the weakness of referrals as compensation for content access. Publishers cannot assume that increased use of their material will produce proportionate audience growth. Hundreds of publishers have begun organizing around new monetization models in response. Publishers need a way to receive value when discovery ends inside an answer without continuing to the source.

Why keyword bidding, cookie retargeting, and click attribution cannot be ported into conversational interfaces

Keyword bidding starts by matching a single search string against a bid list. In a conversation, though, what a user means may depend on earlier turns. The question “what about vegetarian options?” depends on the preceding discussion of meal planning. A long exchange about how to coordinate a team can also show purchase intent for project management software without naming that category.

Cookie retargeting needs browser storage, domains, and a way to keep track across sessions. Those are not properties of the conversational thread itself, even when a browser hosts the interface. OpenAI uses conversation-level context as its primary ad-matching mechanism. Its April 2026 privacy policy also formalized data-sharing with marketing partners and introduced a Measurement Pixel. That distinction matters because contextual relevance does not require accumulated browsing history, but contextual targeting does not mean that every surrounding marketing or measurement process is free of identifiers.

Last-click attribution fails at a different boundary. It assumes that a measurable click connects an impression to a conversion. In conversational discovery, users can research, compare, and form a preference inside the thread before navigating directly to a brand site. They may go several prompts deep before moving to the open internet, separating the decision from the observable visit. A click report can therefore describe the exit without capturing the conversation that made it commercially meaningful.

What conversation-level intent signals make possible that page-level signals cannot

Early questions that seek information, later comparisons, and final decision signals show how ready a user is to buy. Their sequence matters because a comparison means more when the system knows what the user has already ruled out. Conversation-level targeting can read that progression.

Pace supplies another dimension. Quick follow-up questions can show urgency, while slower exchanges can suggest that the user is taking time to research. Microsoft Copilot uses the full session context rather than only the latest query, giving targeting a view of the conversation’s direction. The opportunity is therefore not merely more text to analyze but a more coherent account of intent. Because that intent changes with each turn, the signal is also perishable and calls for real-time matching.

What the technical infrastructure of real-time conversational ad serving actually requires

Native conversational ad serving must read each new query in light of the earlier thread and keep state across turns. It must then evaluate potential ads, score their relevance, run an auction, and serve the result within milliseconds without disrupting the exchange. This resembles an inference pipeline more than a keyword lookup followed by a bid. The difficult work happens before the auction, when the system determines what the conversation means now.

The advertiser’s targeting input changes accordingly. A freeform natural-language description of the conversations where an ad belongs gives a semantic system more useful direction than a comma-separated topic list. Keyword-list thinking discards the relationships the matcher needs to understand. Infrastructure built for this medium, including Thrad’s real-time prompt analysis and contextual ad infrastructure, works from live conversational meaning rather than requiring a page, cookie, or keyword match. Adding a conversational placement to an existing campaign does not, by itself, supply those capabilities.

Measurement still needs work of its own before it can be treated as complete. OpenAI expanded its measurement infrastructure as of October 5, 2026, with Pixel and Conversions API, an Event Quality Score, a one-day view-through window, and attribution partnerships. These additions provide more ways to connect exposure and outcomes. They do not remove the separation between an assistant-mediated decision and a later conversion.

How the major AI surfaces have each resolved — or declined — the monetization question differently

OpenAI launched ChatGPT ads on February 9, 2026, after years of publicly expressing reservations about advertising. Ads reach logged-in adults on Free and Go tiers, while Plus, Pro, Business, Enterprise, and Education remain ad-free. This makes the available audience narrower than ChatGPT’s total audience. For B2B advertisers, the distinction matters when intended buyers are disproportionately on paid tiers.

Microsoft Copilot instead monetizes through existing Microsoft Advertising campaigns. Eligible campaign types are automatically opted in, and advertisers cannot opt out of Copilot serving. Microsoft states that Copilot placements outperform traditional search ads on click-through and conversion rates, a vendor claim that remains independently unaudited. Copilot’s position inside Microsoft 365 workflows also gives it a distinct relationship to business audiences.

Google AI Overviews and AI Mode keep ads within the search results. Their auction logic, keyword inputs, and campaign types predate the conversational experiences they now serve. Perplexity took another path, launching sponsored answers in November 2024, taking no new advertisers from October 2025, and winding the program down by February 2026. As of February 2026, these choices show an unsettled market, with different answers about audience access, campaign control, and whether advertising belongs in the product at all.

What monetization infrastructure do independent AI app publishers need

An independent publisher needs a supply-side system for an interface it controls. ChatGPT Ads Manager lets advertisers buy placements but does not monetize an independent healthcare assistant, specialized research tool, or niche LLM product. Major platforms’ own ad programs therefore address a different buyer. They do not supply the missing infrastructure for publishers outside their surfaces.

The requirement is a conversational SSP connected to advertiser demand deep enough to fill varied inventory. It must understand context natively, work across the publisher’s interface, and apply brand safety standards suited to AI responses. Google AdSense, header bidding stacks, and display networks can serve webpage inventory, but that does not make them native monetization systems for a dialogue. Higher CPMs for native conversational inventory indicate an opportunity when the right infrastructure is present. Publishers still need sufficient demand and meaningful fill rates to turn that pricing opportunity into revenue.

What a purpose-built monetization layer for conversational interfaces actually looks like, and where Thrad fits

Thrad operates as both a DSP and SSP, with its own exchange and direct publisher relationships across AI apps. Its contextual matching reads what a conversation is about rather than relying on a static page, an isolated keyword, or a browsing-derived profile. For publishers, its SSP and exchange provide infrastructure built for conversational inventory. For advertisers, its buying role connects demand to multiple AI surfaces.

Thrad uses the term “conversational AI advertising” for paid, targeted commercial messages inside AI assistant and LLM interfaces that match the intent of an ongoing conversation. This distinguishes the category from chatbots that sell products, AI-generated creative, and search ads placed beside AI summaries. It also stresses brand safety and transparency, both of which matter when ads appear near model-generated responses.

Any capable platform in this category must meet the same underlying requirements. The prompt supplies an intent signal, while the response context supplies a placement opportunity. Matching must stay within the latency limits of a live conversation, and publishers must have enough demand for their supply to achieve fill rates that bring in revenue. Measurement must distinguish what view-through windows and incrementality testing can establish from what remains unresolved. Brand safety must also account for adjacent model-generated content that the advertiser does not control.

What publishers and advertisers should assess before moving budget toward conversational AI channels

Advertisers should first check whether their existing Google and Microsoft campaigns already serve ads in AI surfaces, since eligible campaigns are automatically opted in. They should then check which audiences they can reach, including ChatGPT’s tier restrictions, before assuming that ads can reach everyone who uses the platform. Context-hint writing requires a different skill from assembling keyword lists, so early campaigns should be learning exercises with measurement plans attached. Criteo observed higher conversion rates for LLM-referred traffic across a sample of retailers, but the finding remains directional and independently unaudited at scale. A unified attribution view across conversational surfaces does not yet exist.

Publishers should check demand depth, fill rate, contextual matching, and brand safety before choosing an SSP or exchange. Higher native CPMs are relevant, but they do not settle the economics without delivery and measurement. Blocking crawlers can recover leverage over content access, but it does not replace referral revenue. Publishers must choose infrastructure that can monetize the conversation that exists, since a page visit may never happen.