AI assistants now sell ads. Set your rules before you set a budget
Sooner or later a client, a CMO or a platform rep will ask whether you should be advertising in AI assistants. Our answer is to decide what your brand will and won't accept first, and talk about budget afterwards.
The order matters because some of these ads can now be shaped by what people tell an assistant, or by summaries of what's in their email. If nobody in your business writes down where the line sits, the first sales pitch will draw it for you. Leaving it alone is still a decision, just one nobody made on purpose.
What changed
In March, Google said there were no ads for people who connect their apps to AI Mode, its chat-style search. "That isn't changing right now," it said, while adding that it expected ads to work in a similar way for those users in future. In mid-October Google told Search Engine Land it had started testing those ads, as Search Engine Roundtable reported.
The users in question have switched on Personal Intelligence, the setting that lets AI Mode draw on their Gmail and Google Photos. Google's help page on connecting these apps says ads won't be based directly on someone's inbox or photo library, but summaries, excerpts and inferences from those apps may be used to show them ads. Advertisers don't get access to the apps, sensitive categories such as health and religion are excluded, and connecting is the user's choice.
Google's privacy policy changed on 16 October too. Until then it said: "We don't show you personalized ads based on your content from Drive, Gmail, or Photos," as Ars Technica reported.
Google's own help page doesn't list the UK among the places where people can connect these apps, so for UK brands this is, for now, something to prepare for.
OpenAI is pushing harder. On 21 October it said ChatGPT ads now run in 89 countries and that more than 100,000 advertisers have run campaigns since launch, as The Next Web reported. That is a running total, with no figures on spend. The ads appear only for people on the Free and Go plans, are labelled and kept separate from ChatGPT's answers, and can be bought through partners or OpenAI's self-serve Ads Manager.
How big the channel is
The SEO Works, a UK agency that sells search and ChatGPT advertising services, looked at the traffic 95 websites received over the 12 months to August 2026. AI assistants sent them 2.6% of the traffic that ordinary unpaid search results (what the industry calls organic search) sent. The share sat between 2% and 2.8% until April, then ranged from 2.6% to 3.9% from May to August, a trend SEO Works calls provisional, so the channel is small but may be growing. Visitors from AI assistants converted about a fifth better on average (7.18% against 6.08%), though not on every site.
The agency's own arithmetic makes the scale plain: on the 63 sites with conversion data, a 5% lift in organic search would have delivered more extra conversions than doubling the traffic from AI assistants. SEO Works calls its conversion figures an upper bound.
Read it with its caveats. These are visits from links in AI answers, not ad clicks. The data stops in August 2026. It also leaves out Google's AI summaries at the top of search results, which can't be separated from ordinary search traffic, so the AI figure is a floor.
Why the stance matters more than the spend
A search ad responds to what someone typed. An ad in AI Mode with Personal Intelligence switched on may draw on inferences from their connected email. If a customer feels an ad knew too much, the cost shows up in how far they trust your brand, and a cheap click won't make up for that. So this is a decision for brand, legal and comms as well as the media team.
Adtrak, a Nottingham agency that sells ChatGPT ad management, argued in June: early access "genuinely means something" while the platform is uncrowded. That is a fair point, and Adtrak adds that being early only becomes an advantage with good tracking and honest measurement. We'd start one step earlier. Being early pays only if you've written down what you'll accept, and what success looks like, before you begin.
For agencies, clients will ask about AI assistant ads before there are any benchmarks. An agency that arrives with a written stance and a set of questions has a more useful conversation than one that arrives with a media plan, and it keeps search budgets from being moved on hype.
For in-house teams, someone in brand, legal or comms needs to say whether the company is comfortable with ads that can draw on what a customer told an assistant, before any test is signed off. Holding budget steady while you settle that costs little, because the channel is small on the evidence so far.
What to do this week
- Write a short position with legal or comms. Would you accept ads shaped by what people tell an AI assistant, or by summaries of their email and photos? Which topics or moments are off-limits for your brand, beyond the platforms' own exclusions?
- Prepare four questions for your agency or platform rep. Which markets and formats are open to us now? What information decides where our ad appears, and what did users agree to? How will results be measured against a group that didn't see the ad? Where can and can't our ad appear next to sensitive conversations?
- Agree who signs off a first test, and what result would justify a second, before anyone books one. Keep any test budget separate so it isn't taken from search.
- Hold search budgets where they are. On this evidence the channel is small, so there is no case yet for moving search money, bearing in mind that SEO Works counted visits from AI answers, not ad clicks.
Where would your brand draw the line on ads shaped by what someone told an AI assistant?
Jules, founder of WYNStrategy Director helping agencies and marketing teams use AI properly. More about Jules