← Back to insights
INSIGHTS • 10SPRING

Ads Are Now in AI Search: How B2B Businesses Need to Adjus

Ads now appear beside AI answers on Google, ChatGPT, and Copilot. Here is what B2B teams should change in paid media, creative, and measurement, and where the evidence is still thin.

Dark navy 10spring graphic titled Ads Are Now in AI Search, showing an AI search answer above a separately labeled sponsored ad and a qualified-pipeline measurement panel.

A B2B buyer who asked a chatbot for vendor options used to get an answer and nothing else. That is changing. Google now shows ads in AI Overviews and AI Mode, OpenAI sells ads below ChatGPT answers, and Microsoft places ads in Copilot. Some other AI Search Engines have said they will stay ad-free.

This article focuses on what B2B teams should change in response. It sets out where ads in AI Search stand today, why buyer behavior matters more than ad formats, and what early performance data does and does not show. It then covers how to adjust campaign structure, ChatGPT testing, creative, measurement, and governance.

It draws on platform announcements and help pages from Google, OpenAI, and Microsoft, a November 2026 academic working paper, buyer surveys from G2 and TrustRadius, and trade reporting. Many figures come from companies that sell ads, software, or research, so it notes where the evidence has limits. It reflects what was published as of November 16, 2026, and these products change often.

Where ads in AI Search stand today

Two of the six AI Search Engines, ChatGPT and Google AI Overviews, are confirmed to carry ads. Google AI Mode and Microsoft Copilot carry them too. The other AI Search Engines have declined, hesitated, or only announced plans.

Here is where the sellers stand.

  • Google AI Overviews and AI Mode. Trade coverage of Google's October 28, 2026 announcement says ads in AI Mode expanded to ten more markets. It also says multiturn ads launched in the US, which can appear in follow-up questions. Existing Search, Shopping, and Performance Max campaigns are eligible, and advertisers cannot opt out.
  • ChatGPT. OpenAI began testing ads on February 9, 2026. They appear for logged-in adults on the Free and Go plans. Plus, Pro, Business, Enterprise, and Education accounts see none. OpenAI says ads are labeled as sponsored, kept visually separate from answers, and do not influence them. Dataconomy reported that OpenAI said the platform runs in 89 countries and that more than 100,000 advertisers have used it.
  • Microsoft Copilot. In a March 2025 post, Microsoft Advertising said sponsored content is distinctly marked and never influences organic responses. PPC Land's coverage of Microsoft's later research says Copilot placements draw on existing search campaign materials.

Four others sit outside the paid inventory, each for a different reason.

  • Claude. Anthropic, which makes Claude, said in February 2026 that Claude will remain ad-free. eMarketer notes the company left open the chance of revisiting the decision.
  • Perplexity. ContentGrip reports that Perplexity began its ad experiment in November 2024. Executives told the Financial Times in February 2026 that the company would not pursue ads further, according to TechSpot. Search Engine Land framed the exit as possibly temporary.
  • Gemini app. Google's vice president of global ads said in December 2025 that the app had no ads and no current plans for them. After Alphabet's first-quarter 2026 call, Business Standard reported that Google is open to exploring ads in the app. As of the reports reviewed, the app still carries none.
  • Grok. Search Engine Land reported that Elon Musk told advertisers X would put ads inside Grok's answers. No source reviewed confirms that they have launched, so treat it as announced, not available.

These positions move. Google's stance on the Gemini app shifted in under five months. No source reviewed says how many B2B buyers use each system for research. A plan that depends on one platform's audience therefore rests on an assumption.

Why buyer behavior matters more than ad formats

The first question for a B2B team is not which ad format to try. It is where buyers form their shortlist, and whether an ad can reach them there.

G2 surveyed 1,076 B2B software buyers in March 2026 and published the results in April. It found that 51% start research with an AI chatbot more often than with Google, up from 29% in its 2025 report. In the same survey, 69% chose a vendor other than the one they first planned to consider, because it appeared in a chatbot's recommendation. About one in three bought from a vendor they had never heard of. G2 sells software review and marketing products, and the survey covers software buyers only.

Buyers do not accept those answers at face value. A TrustRadius 2026 report, summarized by its parent company HG Insights, surveyed 1,862 buyers and 444 vendors. It found that 63% of buyers used AI during their purchase process and that 94% fact-check AI-generated information at least some of the time. Another 72% verify always or very often. TrustRadius and HG Insights sell buyer reviews and data, so both have an interest in the findings.

Together, these findings point to two facts that matter for paid media.

  • Shortlists form inside answers. A vendor is either named in the recommendation or it is not. Ads sit beside that recommendation, and the platforms say ads do not change it. OpenAI says ads do not influence ChatGPT's answers, and Microsoft says the same about Copilot.
  • Buyers check what they see. Claims in an ad or in an answer will be checked against other sources. That is the point of the fact-checking figures, though neither survey measures ads specifically.

The practical conclusion is an inference. Paid placement can put a company beside an answer. It cannot buy a place inside the recommendation. A paid program that ignores how recommendations are earned leaves the most influential step to chance.

Reach is the second limit. ChatGPT ads show only on Free and Go plans, and OpenAI's post says Free users can opt out in exchange for fewer daily messages. It also says ads do not show to users it knows or predicts are under 18, and they are not eligible on sensitive topics such as health, mental health, and politics. A buyer researching on a work account with a paid plan will not see your ad. That is an inference from the plan rules, and no source reviewed measures how many B2B buyers fall into each group.

What the early performance evidence shows

Independent evidence on ads in AI Search is thin, and almost none is specific to B2B. The most useful recent study covers ChatGPT ads and ecommerce.

Christian Schulze of Frankfurt School of Finance & Management and Maximilian Kaiser of WU Vienna analyzed data from 1,898 online shops, covering June 2025 through August 2026. The Frankfurt School release of November 3 reports three findings.

  • Revenue per visit. A visit from a ChatGPT ad produced 0.89 dollars in revenue, against 2.45 dollars for a Google search ad visit.
  • Conversion rate. It was 0.37% for ChatGPT ads and 1.17% for Google search ads.
  • No purchases. Of 102 advertisers with complete revenue data, 82% recorded no purchase from their ChatGPT ads.

The release also reports that, after ads launched, ChatGPT sent about 6 to 8% fewer visitors to the shops through unpaid recommendations. Those visitors converted at the same rate as before.

Search Engine Journal's review, published November 16, lists limits that matter before anyone acts on the numbers.

  • Scope. The revenue comparison covers US traffic.
  • Descriptive, not causal. The authors call the work descriptive, so it does not show what would happen if you moved budget between channels.
  • Not a B2B result. About 20% of the sites are services, software, education, or finance businesses. The paper defines a transaction as a completed order and does not report B2B results separately.
  • A small number of sites. Only 18 of the 102 sites recorded any ChatGPT transactions.
  • An early period. The data predate OpenAI's conversion-optimized campaigns and the bidding changes that took effect this month.
  • A disclosed connection. The data were donated by Grips Intelligence, where Kaiser is a former employee and a scientific advisor. The authors say the company played no role in the research.
  • OpenAI's response. An OpenAI spokesperson said the study captures an earlier period and extrapolates from a small sample.

The release also reports a per-dollar comparison that favors Google by about 6.6 times. According to the same review, it rests on publicly reported click rates and changes with them. Do not build a budget decision on that figure.

On the Google side, the evidence is mostly the company's own. Campaign reported that Google said lower monetization in AI Mode weighed on Search growth in the third quarter. That is Google describing its own revenue, not advertiser results.

What the evidence supports is modest. Ads in AI Search are early, the formats are changing monthly, and no study reviewed shows whether they produce qualified B2B pipeline. Treat every performance claim as a hypothesis to test in your own account, with your own sales data.

Adjustment 1: Decide what Google can already show your ads in

Google does not ask you to buy AI Mode or AI Overviews placements. The same trade coverage of Google's help page says eligible campaigns join automatically, with no separate campaign or budget. It also says advertisers cannot target or opt out of these surfaces separately from Search. That makes campaign structure the main lever.

  • Map eligibility. The coverage says contextual matching, which matches ads to both the question and the AI answer, is limited to AI Max for Search, broad match, Shopping, and Performance Max. Exact and phrase match serve only when a query shows direct commercial intent. List which of your campaigns use which settings.
  • Use the controls that remain. Negative keywords, brand exclusions, and asset pinning are still available. Review them for job seekers, students, free-tool searches, and any terms you do not want your brand beside. That advice is an inference, not Google guidance.
  • Accept aggregated reporting. The coverage says AI Overviews and AI Mode results fold into standard campaign and search terms reports with no separate segment. Plan analysis at the campaign and pipeline level, not the placement level.
  • Prepare for follow-up questions. Multiturn ads can appear after a buyer's second or third question. Ad assets and landing pages should answer follow-ups such as integrations, security, implementation, and comparisons. That is an inference from how multiturn ads work.

Google has also introduced Business Agent for Leads. Search Engine Land reported that it puts a Gemini-powered agent, trained on the advertiser's website, inside lead generation ads in place of a static form, launching in open beta for US advertisers. Google's blog describes it as giving instant answers based on your website.

A Search Engine Land commentary warns that the risks are higher for lead generation advertisers without robust offline conversion data. For a B2B team, the agent is only as good as the site it learns from and the qualification rules you set. That is an inference.

Adjustment 2: Test ChatGPT Ads as an audience question, not a keyword question

ChatGPT Ads does not work like paid search, and the differences matter for B2B.

  • Targeting is conversational. OpenAI says ads are matched to the topic of the conversation, past chats, and past ad interactions. Advertisers receive aggregate performance data and do not see chats. The sources reviewed describe no query-level report, so keyword mining as practiced in paid search is not available.
  • Audience lists exist. OpenAI's help page says custom audiences are built from lists of customers or prospects, using emails, phone numbers, hashed versions, or Google Advertising IDs. PPC Land notes that one use is suppression, such as excluding existing customers and people in active CRM flows. Upload those exclusions before spending on prospecting.
  • Bidding changed this month. Search Engine Roundtable reported an OpenAI email of November 5. Target CPA and Target CPC replace Max Bid and Max CPC for new campaigns from November 9, and the older options retire January 1, 2027.
  • Signal quality matters. The same email advises confirming that the pixel or Conversions API sends the selected event. It also advises basing targets on recent actual results, because tight targets or weak conversion data can limit delivery.

That last point has a B2B consequence, and it is an inference. The event you send is the event the system learns to chase. If it is a form fill, expect optimization toward form fills. Where you can, send a downstream event such as a qualified meeting or a created opportunity.

On September 16, 2026, HubSpot announced what it calls the first CRM integration with ChatGPT Ads. OpenAI's help page says it lets advertisers create campaigns, add leads directly to the CRM, trigger follow-up workflows, and view performance beside other channels. HubSpot says closed deals can be traced to the campaign that started them. Those are vendor claims. HubSpot's announcement does not say what conversion data returns to OpenAI, and HubSpot sells CRM software.

OpenAI is also testing Sponsored Agents with select US advertisers. After an ad, a user can start a labeled conversation with a business-sponsored agent, kept distinct from ChatGPT's independent answers, as Search Engine Roundtable reported. The next adjustment covers what that means for governance.

Adjustment 3: Pair paid placement with earned visibility

If an ad cannot buy a place in the recommendation, something else has to earn it. The buyer research suggests where trust comes from.

G2's survey found that 85% of buyers think more highly of a vendor when an AI answer includes it. It also found that citations from software review sites are the top signal that raises confidence in an AI answer. Buyers also treat consistency across chatbots as a trust signal, and they dig deeper when an answer conflicts with what they already know. TrustRadius's fact-checking figures point the same way.

The Frankfurt School data adds one more observation. In that sample, unpaid recommendations from ChatGPT performed better than paid ads. It is one ecommerce study with the limits above, so it is a prompt to test, not a rule.

None of this shows that earned visibility causes sales. It shows where buyers say trust comes from. Three practical steps follow, and all are inferences.

  • Keep facts consistent. Product names, integrations, use cases, and positioning should match across your website, profiles, review listings, and third-party coverage.
  • Earn third-party mentions. Review sites and credible publications are the places buyers and AI answers check. That is the work of Digital PR and Content Development.
  • Use ads to learn, not to substitute. Paid tests can show which questions and messages draw qualified interest. They do not replace the work of being recommended.

Adjustment 4: Review creative, claims, and automated agents

Two new formats let software speak for your company. Google's Business Agent for Leads answers questions inside the ad, and OpenAI's Sponsored Agents start a conversation after a click. Both change who is responsible for what a buyer hears.

  • Build an approved claims library. List the product claims, comparisons, and proof points your company is willing to stand behind, each tied to a page that supports it. Buyers fact-check, so an unsupported claim is a liability.
  • Define the agent's scope. Decide which questions it answers, which sources it may use, and when it hands off to a person. The sources reviewed do not say what conversation data advertisers receive from either format, so ask before you commit.
  • Set the qualification rules. An agent that filters leads needs your definition of a qualified buyer. Without it, the agent will optimize for whatever it can measure.
  • Review for accuracy on a schedule. Check what the agent says against your current product, security, and compliance facts. Assign a named owner.
  • Check regulated and sensitive topics. OpenAI excludes some topics from ads. Confirm what your category allows before building campaigns.

Landing pages carry more weight in this setting. A buyer who arrives after a conversation expects the page to continue it. Conversion Rate Optimization (CRO) work should check whether pages answer the next question a buyer is likely to ask, such as comparison, integration, and security. That is an inference.

Adjustment 5: Measure paid media against pipeline

Reporting in AI Search is thin, and that raises the cost of measuring the wrong thing. Google folds AI Mode results into standard reports. OpenAI gives advertisers aggregate data. Platform-reported conversions also follow each platform's own attribution rules, so adding them across platforms can count the same buyer more than once.

A revenue-first view keeps platform metrics in their place. It uses three layers.

  • Layer 1: activity and infrastructure signals. Impressions, clicks, click-through rate, cost per click, and platform-reported conversions. Use them to diagnose delivery and tracking, not to judge the program.
  • Layer 2: conversion performance. Qualified opportunities and customers by paid source, campaign, and landing page. This is where a click becomes a sales conversation or does not.
  • Layer 3: business impact. Pipeline, customers, and recurring revenue, attributed only where reliable data supports it.

Four practices make the layers workable when platforms report little.

  • Tag every paid source. Use consistent UTM parameters on every ad, including ChatGPT and Microsoft placements, so that sessions and CRM records can be tied to a source.
  • Ask buyers how they found you. Capture it in forms and in the first sales call. Ads in AI Overviews and AI Mode will not appear in separate Google reports, so buyer-reported source helps fill the gap.
  • Send qualified outcomes back. Where a platform accepts conversion data, send qualified opportunities, not only form fills. The platforms optimize toward the signal you give them.
  • Compare over quarters. A long B2B sales cycle means a few weeks of click data says little about pipeline. Set a baseline, change one thing at a time where possible, and use hold-out groups or geographic tests where volume allows.

If clicks rise and qualified opportunities do not, diagnose the conversion pathway before adding spend. Look at landing page match, form friction, page speed, lead routing, and follow-up time. A paid click that lands on a weak page is a cost with nothing behind it, whichever surface it came from.

Set rules for testing new inventory

New ad inventory arrives monthly, and each launch invites a rushed budget decision. A short set of rules prevents that.

  • Cap the test. Set a fixed budget that you can lose without changing the plan, and fund it from experimentation, not from programs that already produce pipeline.
  • Write the hypothesis first. State what you expect, which metric would confirm it, and what would end the test.
  • Define conversion events in advance. Choose the qualified outcome before launch, and confirm it is tracked.
  • Assign an owner and a review date. One person decides whether to continue, change, or stop.
  • Log platform changes. Record changes such as bidding options retiring or new formats, because they can shift results mid-test.

Where ads in AI Search fit among the nine Execution Channels

Ads in AI Search belong to PPC & Paid Media, but their results depend on other channels.

  • PPC & Paid Media. Campaign structure, controls, tests, and budget rules decide how much of the new inventory you reach and at what risk.
  • AI Search Readiness + SEO. Consistent, accurate information helps AI Search Engines describe the company correctly and recommend it, which ads cannot buy.
  • Content Development. Approved claims, comparisons, and answers give both ads and AI answers something accurate to draw on.
  • Digital PR. Third-party citations, especially from review sites, are among the signals buyers say they trust.
  • Conversion Rate Optimization (CRO). Landing pages decide whether paid clicks and agent conversations become qualified opportunities.
  • RevOps. Lead routing, qualification rules, and CRM data decide whether paid results reach pipeline.
  • Account-Based Marketing. Account lists and suppression lists decide where paid reach is wasted or useful.
  • GTM Strategy. Positioning and the ideal customer definition decide which questions you want to be associated with.

10spring treats PPC & Paid Media as one of nine Execution Channels that work together. It performs best when it is coordinated with the others and judged on pipeline, not clicks.

Practical takeaways

  • Separate announcements from evidence. Label each claim as platform-reported, vendor-reported, or independently measured before it shapes a budget decision.
  • Start with buyers. Find out where your buyers research and whether they use ad-eligible accounts, before choosing a platform.
  • Do not expect ads to buy a recommendation. Plan earned visibility alongside paid reach.
  • Audit Google eligibility. Know which campaigns use broad match, AI Max, or Performance Max, and use negative keywords, brand exclusions, and asset pinning.
  • Plan for aggregated reporting. Expect no separate AI Mode or AI Overviews segment, and measure at the campaign and pipeline level.
  • Test ChatGPT Ads narrowly. Use suppression lists, a clearly defined conversion event, and a fixed budget.
  • Govern agents that speak for you. Set scope, sources, handoffs, an owner, and a review schedule.
  • Check claims before they run. Buyers fact-check, so each claim needs a supporting page.
  • Measure in layers. Judge paid media on qualified opportunities, pipeline, and customers, and use clicks as diagnostics.
  • Recheck platform positions often. Gemini, Perplexity, Claude, and Grok positions can change, so confirm current policies before building a plan around them.

10spring works alongside internal marketing teams. If you want to pressure-test how your paid media connects to pipeline, Book A Call. Let's accelerate your pipeline together.

Sources and further reading

All sources accessed November 16, 2026.