AI search is becoming commercial. It is not becoming one marketplace.

AI search is becoming more commercial. That much is clear.

OpenAI is testing conversations with business-sponsored agents. Google has announced AI-generated ad explanations and chat-based lead experiences. Anthropic has published infrastructure for commerce agents while promising that Claude itself will remain ad-free. Perplexity tried sponsored follow-up questions, then reportedly phased the experiment out.

It is tempting to describe all of this as one market taking shape.

That would be a mistake.

The four platforms are not merely choosing different ad formats. They are making different decisions about who hosts the commercial interaction, what an advertiser can pay for, where the user’s trust should sit, who controls the agent and whether commerce occurs inside the assistant, inside an advertisement or inside a merchant’s own system.

The useful question is no longer simply whether AI assistants will carry advertising.

It is: <strong>what kind of commercial architecture is each platform building, and what can a brand actually influence inside it?</strong>

<h2>Four platforms, four commercial architectures</h2>

<div class='overflow-x-auto not-prose my-8'> <table class='w-full min-w-[820px] border-collapse text-left font-sans text-sm'> <thead><tr class='bg-[#15303C] text-[#F7FAF8]'><th class='px-4 py-3'>Platform</th><th class='px-4 py-3'>Commercial path</th><th class='px-4 py-3'>Verified status on 23 September 2026</th><th class='px-4 py-3'>What the evidence does not establish</th></tr></thead> <tbody> <tr class='border-b border-[#CAD7D3]'><td class='px-4 py-3 font-semibold'>OpenAI</td><td class='px-4 py-3'>Optional business-sponsored conversation after an ad</td><td class='px-4 py-3'>Limited alpha for selected advertisers; announced US test</td><td class='px-4 py-3'>General availability, self-service configuration, pricing, adoption or influence over independent answers</td></tr> <tr class='border-b border-[#CAD7D3] bg-[#EAF1EF]'><td class='px-4 py-3 font-semibold'>Google</td><td class='px-4 py-3'>Separate AI Mode tests, Shopping explainers, lead chat and campaign optimisation</td><td class='px-4 py-3'>A mix of tests, beta products and forthcoming formats</td><td class='px-4 py-3'>One available marketplace, guaranteed delivery, completed sales or changes to organic answers</td></tr> <tr class='border-b border-[#CAD7D3]'><td class='px-4 py-3 font-semibold'>Anthropic</td><td class='px-4 py-3'>Ad-free Claude plus merchant-deployed agent infrastructure</td><td class='px-4 py-3'>Ad-free policy plus available open-source reference blueprint</td><td class='px-4 py-3'>A hosted Claude marketplace, paid answer placement or end-to-end checkout in the reference code</td></tr> <tr><td class='px-4 py-3 font-semibold'>Perplexity</td><td class='px-4 py-3'>Historical sponsored follow-up questions and paid media beside answers</td><td class='px-4 py-3'>US experiment independently reported phased out by February 2026</td><td class='px-4 py-3'>A current buying channel, permanent official ban or sponsor control of answers and citations</td></tr> </tbody> </table> </div>

These differences matter because the same phrase, such as “conversational commerce”, can describe very different systems.

A paid conversation entered from an ad is not the same as an independent AI answer. A merchant-owned agent running on Claude is not the same as a sponsored placement in Claude. A Gemini explainer inside a Google ad is not the same as a brand changing an organic AI Mode response. A sponsored follow-up question is not the same as buying a Perplexity citation.

<h2>OpenAI is testing a paid conversation beside the independent answer</h2>

On 16 September, OpenAI said it was testing <strong>Sponsored Agents</strong> with select advertisers in the United States. The documented flow begins with an advertisement. If the user chooses, they can open a clearly labelled conversation with a business-sponsored agent, ask follow-up questions and later follow a link to the business website. [1]

That is a meaningful change from a conventional display or search ad. The paid unit can become a dialogue. A brand may be able to answer questions, clarify fit and help someone evaluate an offer before the website visit.

But the boundary is equally important. OpenAI says the sponsored-agent conversation is distinct from ChatGPT’s independent answers and separate from the original ChatGPT conversation. Its advertising documentation says advertisers cannot shape, rank or alter ChatGPT’s responses. [1][2]

As of 23 September, OpenAI’s Help Centre described Sponsored Agents as a <strong>limited alpha</strong> available only to selected advertisers and said it was not accepting early-access requests. [3]

That means marketers should not treat Sponsored Agents as a broadly executable media plan. ChatGPT Ads, Ads Manager, Shopify and HubSpot integrations may be more widely available, but none of those facts establishes that every advertiser can build or buy a Sponsored Agent.

The accurate near-term interpretation is narrower: OpenAI is testing a new paid conversational surface, while publicly maintaining a separation between that surface and the independent answer.

<h2>Google is building a family of sponsored experiences, not one product</h2>

Google’s position is more complex because several different products sit under the broad idea of AI-powered advertising.

At Google Marketing Live in May, Google said it was testing two formats in AI Mode: <strong>Conversational Discovery ads</strong> and <strong>Highlighted Answers</strong>. Google says both include an independent Gemini explainer alongside advertiser creative and remain clearly labelled Sponsored. [4]

Google separately announced <strong>AI-powered Shopping ads for Search</strong>. In the example it published, Gemini selects relevant products and writes a custom explanation of why a product may fit the shopper. Google’s product page said the format would soon be available through Performance Max and AI Max for Shopping campaigns. “Soon” is not the same as generally available. [4][5]

A third product, <strong>Business Agent for Leads</strong>, puts a Gemini-built chat experience inside a Search advertisement. Rather than completing a static form, a prospect can choose Chat and receive answers grounded in the advertiser’s website. Google officially described the product as beta in India by 9 July. [4][6]

Then there is <strong>AI Max for Shopping</strong>, which Google documents as a beta optimisation suite. It is related to these newer surfaces, but it is not another name for the AI-powered Shopping ad format. [7]

The practical problem for marketers is nomenclature. If AI Mode ads, Search Shopping explainers, in-ad lead agents, AI Overview inventory and campaign optimisation features are reported as one product, availability and performance become impossible to interpret.

Google is clearly trying to bring more evaluation into sponsored experiences. But the evidence does not show that every format is live everywhere, that a conversation produces a lead, or that an explainer produces a sale. It also does not show that paid activity changes Google’s organic AI answers.

<h2>Anthropic is separating the assistant from the merchant’s infrastructure</h2>

Anthropic has taken the clearest public position against advertising inside its assistant.

In February, the company said Claude would remain ad-free. It said users would not see sponsored links beside Claude conversations, and that Claude’s responses would not be influenced by advertisers or contain unrequested third-party product placements. [8]

Anthropic did not reject commerce. It drew a line around who initiates it. The company said commerce interactions should begin when the user asks for help, rather than when an advertiser pays to insert itself.

Its September release fits that model. <strong>Claude Commerce Agents</strong> is an open-source reference blueprint for businesses to build a customer-facing shopping agent and an operator-facing merchant agent in their own applications and systems. It is not a hosted Claude marketplace. [9][10]

The shopping agent can search a catalogue, compare products, assemble multiple items, build a cart and hand that cart to checkout. The reference code leaves payment to the merchant or payment provider. The merchant agent can analyse sales and inventory, recommend changes and stage actions, but the reference design requires human approval outside the conversation before operational changes are applied. [9][10]

This is commercial infrastructure without paid placement in Claude.

The distinction matters. A merchant can still determine its catalogue, prices, offers, product ordering and business rules. “Ad-free” does not mean economically neutral. It means Anthropic says the Claude conversation itself will not be financed by advertiser influence.

<h2>Perplexity’s sponsored-discovery experiment is a warning against assuming convergence</h2>

Perplexity was one of the earliest answer engines to announce advertising.

In November 2024, it said it would experiment in the United States with sponsored follow-up questions and paid media beside an answer. Perplexity said sponsored material would be labelled, advertiser influence would not affect answer content, and answers to Sponsored Questions would be generated by Perplexity rather than written or edited by the sponsor. [11]

The format was important because it did not necessarily buy the answer. It could buy the next avenue of discovery a person was invited to explore.

But the experiment did not become an obvious template for the market. Search Engine Land, citing Financial Times reporting, said in February 2026 that Perplexity had phased out the ads it began testing in 2024 and had no current plan to restore them, while leaving open the possibility that it could revisit advertising later. Perplexity did not publish a matching closure announcement in the official materials reviewed for this article. [12]

That is why Perplexity should be described as a historical sponsored-discovery experiment, not a current public media-buying channel and not an official permanent ban on advertising.

Its separate Merchant Program and shopping features should not be used to imply paid recommendation placement. At launch, Perplexity described its Merchant Program as free and unrelated to sponsored questions. [13]

<h2>The shared direction is commercial evaluation, not one marketplace</h2>

There is still a common thread.

All four companies are addressing some part of the commercial decision journey. OpenAI is testing a paid brand conversation. Google is testing explainers and lead qualification inside sponsored surfaces. Anthropic is helping merchants build agents inside their own systems. Perplexity tested sponsored discovery before stepping back.

But the unit of competition is different in each case.

<ul><li><strong>OpenAI</strong> is testing paid access to a separate branded dialogue.</li><li><strong>Google</strong> is testing paid evaluation inside several Search and AI surfaces.</li><li><strong>Anthropic</strong> is enabling merchant-owned assistance and operations.</li><li><strong>Perplexity</strong> historically tested paid influence over the next question a user might ask.</li></ul>

The deeper interpretation is that AI is creating several new commercial layers around the answer, not one standard marketplace inside it.

<h2>What AEO teams should do now</h2>

First, protect the distinction between <strong>earned representation</strong> and <strong>paid interaction</strong>. A brand appearing in an independent answer is an AEO outcome. A person entering a sponsored-agent conversation is an advertising outcome. They may occur in the same interface, but they require different disclosure, measurement and causal language.

Second, build a <strong>decision-context map</strong> that both teams can use. Organic AEO and paid conversational media may share upstream work: defining the customer need, the claims a brand can defend, the conditions under which it is a strong fit, and the evidence that supports those claims. This is the connection explored in AEO Updates’ earlier analysis of <a href='/articles/chatgpt-ads-paid-counterpart-to-aeo'>ChatGPT Ads as the paid counterpart to AEO</a>. Shared inputs do not make the delivery systems the same.

Third, prepare the merchant-owned data layer. Anthropic’s blueprint and Google’s website-grounded Business Agent both make one requirement visible: agents need accurate product, policy, inventory, pricing and service information. A vague marketing page is a weak substrate for either a helpful agent or a defensible recommendation.

Fourth, maintain four separate scorecards:

<div class='not-prose my-8 grid gap-4 sm:grid-cols-2'> <div class='rounded-2xl border border-[#CAD7D3] bg-white p-5'><p class='font-mono text-xs uppercase tracking-[0.18em] text-[#5B45D6]'>01 · Earned answer visibility</p><p class='mt-3 font-sans text-sm leading-6 text-[#15303C]'>Mentions, citations, recommendation position and claim accuracy in independent answers.</p></div> <div class='rounded-2xl border border-[#CAD7D3] bg-white p-5'><p class='font-mono text-xs uppercase tracking-[0.18em] text-[#5B45D6]'>02 · Paid conversational engagement</p><p class='mt-3 font-sans text-sm leading-6 text-[#15303C]'>Sponsored-agent starts, useful exchanges, qualified handoffs and cost per outcome.</p></div> <div class='rounded-2xl border border-[#CAD7D3] bg-white p-5'><p class='font-mono text-xs uppercase tracking-[0.18em] text-[#5B45D6]'>03 · Merchant-agent readiness</p><p class='mt-3 font-sans text-sm leading-6 text-[#15303C]'>Catalogue completeness, policy coverage, grounding quality, task success and escalation safety.</p></div> <div class='rounded-2xl border border-[#CAD7D3] bg-white p-5'><p class='font-mono text-xs uppercase tracking-[0.18em] text-[#5B45D6]'>04 · Downstream commercial outcomes</p><p class='mt-3 font-sans text-sm leading-6 text-[#15303C]'>Qualified leads, website visits, checkout starts, sales and incremental value.</p></div> </div>

A five-point gain in one scorecard cannot be assumed to offset a five-point loss in another. Their denominators, incentives and causal paths are different.

<h2>AEO Updates Takeaway</h2>

AI search is becoming commercial, but the evidence does not support one inevitable marketplace model.

OpenAI is testing a sponsored conversation while separating it from the independent answer. Google is testing several sponsored evaluation formats with different rollout states. Anthropic is keeping Claude ad-free while giving merchants infrastructure to build their own agents. Perplexity tested sponsored discovery and reportedly stepped back.

For brands, the strategic task is not to choose one future too early. It is to build an evidence layer strong enough to earn independent representation, a product and policy layer structured enough to support agents, and a measurement system disciplined enough to keep paid, organic and merchant-owned influence separate.

The answer may become commercial.

That does not mean every platform will commercialise it in the same way.

<h3>References</h3>

[1] OpenAI. <a href="https://openai.com/index/reimagining-advertising-with-ai/" target="_blank" rel="noreferrer">Reimagining advertising with AI</a>. 16 September 2026.

[2] OpenAI Help Centre. <a href="https://help.openai.com/en/articles/20001047-ads-in-chatgpt" target="_blank" rel="noreferrer">Ads in ChatGPT</a>. Accessed 23 September 2026.

[3] OpenAI Help Centre. <a href="https://help.openai.com/en/articles/20001524-sponsored-agents-in-chatgpt-ads" target="_blank" rel="noreferrer">Sponsored Agents in ChatGPT Ads</a>. Accessed 23 September 2026.

[4] Google Ads & Commerce. <a href="https://blog.google/products/ads-commerce/google-marketing-live-search-ads/" target="_blank" rel="noreferrer">A new generation of ads for the AI era of Search</a>. 20 May 2026.

[5] Google Accelerate. <a href="https://business.google.com/us/accelerate/announcements/ai-powered-shopping-ads/" target="_blank" rel="noreferrer">AI-powered Shopping ads</a>. Accessed 23 September 2026.

[6] Google India. <a href="https://blog.google/intl/en-in/products/google-companies/google-marketing-live-2026-delivering-the-gemini-advantage-for-indian-businesses/" target="_blank" rel="noreferrer">Google Marketing Live 2026: Delivering the Gemini advantage for Indian businesses</a>. 9 July 2026.

[7] Google Ads Help. <a href="https://support.google.com/google-ads/answer/17091277?hl=en" target="_blank" rel="noreferrer">About AI Max for Shopping campaigns (beta)</a>. Accessed 23 September 2026.

[8] Anthropic. <a href="https://www.anthropic.com/news/claude-is-a-space-to-think" target="_blank" rel="noreferrer">Claude is a space to think</a>. 4 February 2026.

[9] Anthropic. <a href="https://claude.com/blog/claude-for-commerce-agents" target="_blank" rel="noreferrer">Building commerce agents with Claude</a>. 2 September 2026.

[10] Anthropic Platform. <a href="https://platform.claude.com/docs/en/about-claude/use-case-guides/commerce-agents" target="_blank" rel="noreferrer">Commerce agent</a>. Accessed 23 September 2026.

[11] Perplexity. <a href="https://www.perplexity.ai/hub/blog/why-we-re-experimenting-with-advertising" target="_blank" rel="noreferrer">Why we’re experimenting with advertising</a>. 12 November 2024.

[12] Search Engine Land. <a href="https://searchengineland.com/perplexity-stops-testing-advertising-469452" target="_blank" rel="noreferrer">Perplexity stops testing advertising</a>. 18 February 2026, reporting attributed to the Financial Times.

[13] Perplexity. <a href="https://www.perplexity.ai/hub/blog/shop-like-a-pro" target="_blank" rel="noreferrer">Shop like a pro: Perplexity’s new AI-powered shopping assistant</a>. 18 November 2024.