Advertising inside
AI experiences
Digital Traction helps businesses evaluate and test advertising inside AI products. The work covers what a US advertiser can actually buy, how a first test should be structured and how its performance will be measured.
Discuss an opportunityWhere the research
journey is going
Buyers now research inside AI answers, and ads are following them to three kinds of place. Each status answers one question: can a US advertiser buy it today?
- Google, AI ModeWhich option fits a small team?SponsoredA text or Shopping adAsk the businessChatGPTWhat matters most when comparing?SponsoredAn ad under the answerChat with the business
ChatGPT and Google
Google and OpenAI each sell ads beside AI answers, and each has an agent format that answers for the business.
Google
- Ads in AI Overviews, Live
- Ads in AI Mode, Testing
- Business Agent for Leads, Announced
OpenAI
- ChatGPT ads, Live
- Sponsored Agents, Testing
Also live: ads in Microsoft Copilot and Amazon’s sponsored prompts, both from campaigns a business may already run.
- An AI appWhat is a simple way to plan this week?Sponsored suggestionOne labeled line, matched to the chatOne network, many apps
Independent AI ad networks
Networks such as Gravity, Koah and Thrad place a labeled suggestion or card inside AI apps that other companies run, matched to the conversation. Gravity and Thrad sell self-serve, and Gravity documents its own pixel and conversions API (each network’s own site, checked September 30, 2026).
Fits offers a broad context match can find, such as consumer products, apps and software. Reach depends on which apps carry each network.
- A video platformAsk about this productA discussion platformAI summaryWhat the threads agree on
Other AI-powered media experiences
Platforms with their own audiences are adding conversational AI to their ads. Meta runs a Business Agent in WhatsApp (September 3, 2026).
Newest Google opened sign-up for a Business Agent in YouTube ads on September 24, 2026.
Drawn specimens, not screenshots · Grey examples, not real ads · Statuses from each platform’s own documentation, checked September 25 to September 30, 2026
One brief, compared across networks
Digital Traction evaluates networks one at a time. The same brief runs through each network in scope, and networks are compared on delivery, cost and click quality, since none publishes an independent audit of its measurement.
Ahead of a test, Digital Traction also reads each network’s written terms: its share of spend, invalid traffic and refunds, what data passes to bidders, and delivery by app.
What the platforms are
asking from advertisers
The newer placements write more of the ad and choose more of the audience themselves. What they work from is still the business’s own: its pages, its feed and the claims it can stand behind.
Creative is built from the landing page
Google’s AI Max and Performance Max write ad text from the website, and OpenAI’s Ads Manager now suggests copy and imagery from the landing page. The pages, the brand guidelines and the approved claims decide what those ads say.
Your website now writes part of your Google adsTargeting is consolidated
Google’s AI placements draw mostly on broad match, AI Max and Performance Max, and ChatGPT ads are matched by context, not keywords. An entry-point map, with each ad and page aligned to the question behind it, keeps the buyer’s path in the business’s hands.
The intent map
The buyer moments worth paying for, written once.
Landing pages
The pages an ad leads to. OpenAI may read one to decide when an ad fits.
Product feeds
Product data in the form each platform reads. Google’s Shopping ads can show inside AI Overviews.
Approved claims
What the business can stand behind. Under OpenAI’s terms, the advertiser answers for ads its AI tools help write and for what a sponsored agent says.
Lead generation
An agent inside the ad can answer before the form, so a lead may arrive after a conversation. Bidding still needs to learn which leads became customers.
Ecommerce
Shopping ads can show inside AI Overviews, and US Shopify merchants can run ChatGPT ads from their Shopify catalog. For a store, readiness is the feed: product titles, attributes and details in Shopify and Merchant Center, written for the questions buyers ask.
The buying model
changes by platform
Each channel has its own way in, its own controls and its own reporting, so each one gets its own read before a test is bought.
Ads in AI Overviews and AI Mode
Bought through the Search, Shopping and Performance Max campaigns a business already runs, in the same auction. Ads in AI Overviews have no opt-out, and AI Mode’s new formats are still in test.
What answer-engine advertising actually is todayChatGPT ads
Self-serve in the US through OpenAI’s Ads Manager, or through its sales team and partners, including HubSpot and a Shopify app. Not shown on Plus, Pro or business plans.
ChatGPT ads: what a US business can buy todayIndependent AI ad networks
Each network sets its own buying terms, measurement and list of apps.
Independent AI ad networks: what a US business can buy, and whose apps the files confirmSponsored business and shopping agents
The ad opens a conversation with an agent that answers for the business. OpenAI’s Sponsored Agents are in a limited US test, Google’s Business Agent for Leads is announced, and Amazon’s sponsored prompts can open its shopping assistant.
An AI agent in an ad still speaks for the businessAgent-to-agent advertising
A buyer’s agent and a seller’s agent agree a media buy between themselves. The standards are still being written, in IAB Tech Lab’s agentic protocols and the Ad Context Protocol.
Research coming
Measure the placement
against a qualified outcome
Each placement reports in its own way, and none of them sees what happens in the CRM unless the business sends it back. The measurement and analytics systems that send it back are on Marketing systems.
| Placement | What comes back | Documented limits |
|---|---|---|
| ChatGPT | A pixel and a Conversions API. Reports are totals, with nothing from the conversation. | Server events must arrive within 7 days. Reported totals may include modeled conversions. No lift study is documented. |
| Google AI Overviews and AI Mode | Normal Search, Shopping and Performance Max reporting. Ads inside AI Overviews count as top ads. | No separate AI Overviews or AI Mode report. A lift study needs a Google representative. |
| Microsoft Copilot | Normal Microsoft Advertising reporting | No Copilot-specific metrics |
| Independent networks | A pixel, a tag or a server API, depending on the network | Each defines an impression differently. |
| Agentic ads | Nothing agent-specific is published | Transcripts, metrics and lead details are not documented |
- Placement: ChatGPT, What comes back: A pixel and a Conversions API. Reports are totals, with nothing from the conversation.Documented limits: Server events must arrive within 7 days. Reported totals may include modeled conversions. No lift study is documented.
- Placement: Google AI Overviews and AI Mode, What comes back: Normal Search, Shopping and Performance Max reporting. Ads inside AI Overviews count as top ads.Documented limits: No separate AI Overviews or AI Mode report. A lift study needs a Google representative.
- Placement: Microsoft Copilot, What comes back: Normal Microsoft Advertising reportingDocumented limits: No Copilot-specific metrics
- Placement: Independent networks, What comes back: A pixel, a tag or a server API, depending on the networkDocumented limits: Each defines an impression differently.
- Placement: Agentic ads, What comes back: Nothing agent-specific is publishedDocumented limits: Transcripts, metrics and lead details are not documented
Unpaid visits from AI assistants are a separate read.
Where AI traffic actually lands in GA4Digital Traction keeps three records apart and never adds them together:
what the platform reports, including its modeled conversions;
the business’s own data, from click IDs to CRM stages;
and the results of controlled tests.
Read against a holdout
Digital Traction has found no independent study showing that ads inside AI answers produce sales that would not have happened anyway. So Digital Traction reads every test against a holdout, a region held back from the ads, not only against the platform’s report.
What can be measured
Enough qualified outcomes to read, a way to see them in the CRM, regional history for a holdout, and consent settings that let the tags run.
Campaign-level lift is not account-level liftWorking together
The work runs in order: the foundation first, then readiness for ads in AI, then the marketing systems that keep it running. Outside the search program, Digital Traction’s part is the research, the foundation, the test design and the measurement, and every change its agents propose waits for a person. Any launch happens in the business’s own ad account, under the person who runs it, and a named person at the business approves the facts and claims. The rules for access and approvals are in How Digital Traction works with your accounts.
- 1
Digital Traction:
Assesses each placement
Audience fit, market access, accounts and measurement, one placement at a time.
- 2
Your team:
Confirms the facts
Approves the claims, the offers and the qualified outcome the test is judged on.
- 3
Digital Traction:
Builds the foundation and the plan
The intent map, context hints, measurement plan and test design.
- 4
Your team:
Opens access and launches
Creates the ad accounts, shares the outcome data, and runs the launch in the business’s own account.
- 5
Digital Traction:
Reads the test
A weekly check against the CRM, and a recommendation to scale, change or stop.
Behind step 1: a decision for each placement
Decisions it supports
- which placement to try first, or none yet
- whether to fix the foundation before spending
- how large a test the question needs: a check that ads serve and clicks arrive, a cost comparison, or a lift study
- whether to scale, change or stop
- whether a sponsored agent pilot is worth seeking, and on what written terms
- Test now
- ChatGPT Ads. The buyers are likely on ChatGPT’s entry plans. Self-serve is open in the account’s country.
- Prepare first
- Google AI Overviews and AI Mode. Already reached through the existing Google Ads campaigns. Conversion data stops at the inquiry.
- Watch
- Independent networks. Few apps match the audience. One network’s supply is declared by both the network and the apps.
- Watch
- Sponsored Agents. A limited test with select US advertisers. Reporting and lead details are not published.
- Scale or stop
- Scale when a qualified outcome costs no more than it does in paid search for two reads in a row, and the holdout shows lift. Stop when it still costs more than twice as much after two rounds of changes, or when the audience does not fit.
Understanding the market behind the placement
The studio’s research follows how advertisers buy, where publishers participate and what can actually be measured.
Explore the research