The next ad impression might not have eyeballs.
That sounds like a joke until you realize the media industry is already trying to monetize traffic from software agents before it figures out how to replace the human clicks AI search keeps eating.
This week, Digiday reported that Time has started serving ads specifically to AI agents. Not banner ads. Not pre-roll. Not some sad display unit wedged between recipes and newsletter popups.
Machine-readable sponsored content.
Time is working with Mobian to turn pages into markdown versions that are easier for AI crawlers and agents to read. Then it places sponsored FAQ-style brand information inside that machine-friendly layer. The first buyers include Ally Bank and the Project Management Institute. Time says the ads are labeled as sponsored content, but the important part is not the label. The important part is the target.
The target is the agent.
And if you are in marketing, this is the part where the floor starts moving.
September 18 update: this is no longer only a publisher-side experiment. OpenAI has now introduced Sponsored Agents, Shopify and HubSpot integrations, AI creative tools, and natural-language campaign management for ChatGPT Ads. Google also expanded Business Agent into YouTube ads. The operator takeaway changed from “watch the weird agent-ad experiment” to “start governing conversational ad journeys before the channel turns into another unmanaged growth toy.”
This is not a gimmick
It is tempting to laugh at “ads for bots” because the phrase sounds like a cursed adtech conference panel sponsored by a company with no revenue and eleven advisors.
But the logic is brutally obvious.
People are using ChatGPT, Gemini, Claude, Perplexity, AI Mode, shopping agents, browser agents, and vertical copilots to compress research. They ask the machine what to buy, who to trust, which vendor to shortlist, which product has the least nonsense attached, and whether a brand is legit.
The machine goes out, reads what it can read, weighs sources, summarizes the market, and hands the buyer a smaller set of options.
So what do marketers do?
They try to influence the material the machine reads.
That is not science fiction. That is just SEO after the click got replaced by a paragraph.
Old SEO was about ranking on a results page. New SEO is about becoming the answer layer’s preferred source of truth. Agent ads are the paid version of that fight. They are not trying to make a human click an ad. They are trying to shape what the AI system retrieves, remembers, cites, or repeats when the human asks for a recommendation later.
That is a massive shift.
Because when you influence one human, you influence one buying journey. When you influence the answer engine, you may influence thousands or millions of downstream decisions.
That is the whole game.
OpenAI already opened the door
This is happening at the same time OpenAI is building its own advertising lane.
Back in January, OpenAI said it planned to test ads in ChatGPT for logged-in adults in the U.S. on free and Go tiers. The company said the ads would be separate, labeled, and would not influence ChatGPT’s organic answers. That distinction matters. It is also exactly where the market is going to get messy.
That test is no longer just a U.S. curiosity. OpenAI said on August 18 that ChatGPT Ads would expand to 31 European markets, including Germany, France, Spain, Italy, Sweden, Norway, Denmark, the Netherlands, and Austria. The same announcement says advertisers will initially enter through OpenAI’s Ads Solutions team, agency partners, and technology partners, with self-service Ads Manager access coming later. OpenAI also says the platform now supports conversion optimization, geo-targeting, custom audiences, the OpenAI Pixel, Conversions API, and third-party measurement integrations.
That changes the operator decision. ChatGPT Ads are still early, still controlled-access, and still separate from organic answers according to OpenAI. But they are now close enough to a real media product that brands should start evaluating budget, measurement, and governance instead of treating AI ads as a weird future rumor.
Then OpenAI made the next step explicit. On September 16, it introduced Sponsored Agents, a format that lets a person start a clearly labeled conversation with a business-sponsored agent after clicking an ad in ChatGPT. OpenAI says the Sponsored Agent conversation is distinct from ChatGPT’s independent answers and separate from the original user conversation. It is currently being tested with select U.S. advertisers.
That same announcement matters for operators because the ad workflow is moving into existing systems, not just a new media dashboard. OpenAI says U.S. Shopify merchants can now use a ChatGPT Ads app to create and manage campaigns from Shopify Catalog data, with international availability planned for markets where ChatGPT Ads are available starting September 23. HubSpot became OpenAI’s first CRM partner, and advertisers can use ChatGPT Work prompts to create, update, and analyze campaigns through the Ads Manager plugin.
That is the factual layer. The interpretation is sharper: AI ads are becoming conversational sales surfaces and workflow tools at the same time. A product brand will not only decide whether to buy a ChatGPT ad. It will decide what product data feeds the ad, who approves the sponsored conversation, what claims the agent is allowed to make, where the handoff lands, and how performance is measured after the buyer leaves ChatGPT.
There are two different ad battles forming at once.
First, there are ads inside AI products. That is the clean version. Sponsored placement appears near the answer. The platform controls the format. Users can see it. Regulators can yell at it. Everyone pretends the boundary is neat.
Second, there are ads upstream of AI products. That is the weird version. Sponsored content lives on publisher pages, markdown feeds, brand knowledge bases, structured product pages, partner sites, comparison hubs, marketplace listings, and whatever else the model or retrieval system can see.
That second category is where things get spicy, because the influence is less visible. It does not need to look like an ad by the time it reaches the user. It may show up as a product mention, a framing choice, a comparison summary, a brand description, or a “best option for X” recommendation.
There is already academic smoke around this. A new paper, The Beginning of ChatGPT Ads, studied early ChatGPT ad delivery using 91 test accounts and collected more than 3,000 ads from 186 advertisers. The researchers found lower-income accounts were more likely to receive ads, regardless of race, and noted that early ads were clearly separated from answer text.
Another 2026 paper, Generative AI Advertising as a Problem of Trustworthy Commercial Intervention, makes the bigger point: generative AI ads are not just content placement. They can become interventions in the generation process itself.
Translation: the ad unit is no longer always the thing you see. Sometimes the ad unit is the information environment your agent read before answering.
Fun, right?
Google is pushing the same pattern into retail media
This is not only an OpenAI story.
Google said on September 16 that Business Agent is moving into YouTube ads in a U.S. beta, letting viewers ask complex product questions and get tailored guidance without leaving the YouTube context. Google also made AI performance insights generally available in Merchant Center across Australia, Canada, India, New Zealand, and the U.S., and it expanded the Universal Commerce Protocol integration hub with cart transfer and checkout testing capabilities rolling out in the U.S.
Again, separate the fact from the operator read. The fact is that Google and OpenAI are both turning ads into assisted conversations around product discovery. The operator read is that the campaign brief now needs to include machine-readable product truth, approved claims, product-feed hygiene, measurement rules, and an escalation path when the assistant cannot answer cleanly.
If your ecommerce team owns the catalog, your media team owns the budget, your brand team owns the claims, and nobody owns the conversation, you have a governance gap.
Marketers are about to ruin this fast
Let’s be honest. The first wave of agent ads is going to be a landfill.
Every brand that just discovered “GEO” is going to sprint into the same dumb behavior they brought to SEO, content marketing, LinkedIn, and every other channel they eventually made unbearable.
They will publish fake FAQs nobody asked.
They will stuff comparison pages with self-serving claims.
They will create “AI-readable” pages that say the same thing twenty-seven ways.
They will hire agencies to vomit markdown at the internet and call it thought leadership.
Then platforms will adjust. Crawlers will distrust thin sponsored content. Publishers will invent disclosure schemes. AI companies will claim their systems can separate signal from manipulation. Regulators will notice five minutes too late. The best brands will quietly clean up their actual data while the desperate ones try to bribe the robot with brochure copy.
Same movie. New monster.
The hot take: agent ads are not the future of brand visibility. They are the tax you pay when your brand truth is too weak to be discovered naturally.
If your product data is clean, your assets are consistent, your pricing is sane, your dealer network is readable, your reviews are real, your comparisons are honest, and your site explains the business like an adult wrote it, agents have something useful to retrieve.
If all of that is a mess, yeah, you are going to be tempted to buy your way into the answer.
But paid influence cannot fix rotten inputs forever.
The real work is boring as hell
The brands that win in agent-driven discovery will not be the ones with the cleverest bot ads. They will be the ones with the cleanest commercial surface area.
That means:
- product pages that machines can parse without needing a séance
- structured data that matches reality
- pricing that does not collapse across marketplaces
- authorized sellers and locations that are easy to verify
- product assets that are current, approved, and consistent
- comparison content that is actually useful
- support docs that answer real buying objections
- fresh pages that do not read like they were generated from a prompt called “make me sound innovative”
This is where marketing stops being a content department and becomes an infrastructure discipline.
The answer engine does not care that your brand campaign had a nice manifesto. It cares whether it can understand your offer, trust your data, compare you against alternatives, and hand the user a confident next step.
That is why the “AI visibility” conversation keeps circling back to ops.
If you sell physical products, ToughAssets matters because agents and humans both need clean, approved product visuals and brand files. ToughMAP matters because pricing chaos becomes trust chaos the second AI shopping tools start comparing sellers. ToughLocator matters because a recommendation that dead-ends at bad dealer data is not a recommendation. It is a faceplant.
Agent ads may become a real media channel. Fine. Test it. Measure it. Keep your eyes open.
The practical move now is simple:
- audit which catalog fields, claims, images, offers, prices, and policies can feed conversational ads
- decide which product questions an agent may answer directly and which ones require a handoff
- require human review for generated ad copy, imagery, and sponsored-agent scripts before launch
- tag campaigns by assistant surface, not only by channel, so ChatGPT, YouTube, AI Mode, and traditional search do not get blended into one mushy report
- monitor the post-click path for contradictions between the ad, the agent conversation, the product page, and the retailer or dealer listing
Because once the buyer starts treating the ad like a sales assistant, sloppy handoffs become brand damage.
But do not confuse paid machine visibility with brand strategy.
The deeper move is to make your brand so legible, current, and operationally clean that the machines do not need to be tricked into understanding you.
Because the next buyer might ask an AI agent first.
And that agent is not impressed by vibes.
Disclosure: BrandWeapons and the TOUGH products share ownership. If you already have a mature DAM, MAP monitoring stack, or store-locator platform, use the system that your operators actually keep current. If you are still running the work through folders, retailer screenshots, and stale dealer spreadsheets, fix that layer before you hand budget to an AI ad channel.