Ads are coming to AI answers. Nobody can measure them yet.

Every new channel goes through the same phase: the audience arrives first, the ads follow, and proof that the ads work comes last. AI search is entering the middle of that story right now.

What happened

AI search is growing faster than the platforms can make money from it. One analysis this week noted that Google search queries grew at their fastest rate in five years while ad revenue growth slowed to 12%. Google and OpenAI are both expanding conversational ads to close that gap. The same piece flags the risk: 70% of regular AI search users in the US and UK say they trust the answers, and more ads could wear that trust down.

The pressure to monetize is real on the investment side too. OpenAI's latest funding round drew less than half of its $30 billion target, at a $1.25 trillion valuation instead of the $1.4 trillion it had hoped for. Investors are getting pickier.

Meanwhile, the answers themselves are keeping people inside the platform. A study of 55,198 answers in Google's AI Mode found that 57% of product links led to Google Shopping, and 46.5% of US product answers linked only to Google pages. Your product can appear in the answer without anyone visiting your site.

Reddit is moving the other way and closing the door. It will test a logged-in-only experience in select markets outside the US this quarter, to limit AI scraping, encourage data-licensing deals and collect better signals for ad targeting. Its search referrals fell by more than one million a week in October, and US daily active users dropped 2% year over year to 50.7 million.

What it means if you lead marketing

The hard part is measurement. Ads in ChatGPT and Google's AI Mode reach people at the moment they are describing what they want to buy, which sounds ideal. But as one piece this week pointed out, neither platform can yet prove those ads create sales that would not have happened anyway. Conversion tracking shows which purchases followed an ad, not which ones it caused. The same piece cites Facebook research showing that this kind of observational measurement can misstate results by three times or more.

So the first sales team pitching you AI answer ads will have impressive dashboards and no real proof. That is not a reason to avoid the channel. It is a reason to treat it as an experiment with a budget limit, not a line item that grows on its own.

The traffic picture also changes what counts as a win. If an AI answer shows your product and links to a marketplace instead of your site, a visibility metric may go up while your sessions go down. Decide now which one you report to leadership.

What I would do

• Run a holdout test before you scale any AI ad spend. One guide this week recommends geographic or audience holdouts of 20% to 30%, running for four to eight weeks, and judging channels on incremental return and new customers rather than last-click revenue.
• Audit your tracking first. Clean UTMs and consistent conversion events are the difference between a test you can read and one you argue about.
• Separate visibility from traffic in your reporting. Track whether you appear in AI answers as its own metric, next to site visits, so a gain in one does not hide a loss in the other.
• Keep your product data clean wherever answers pull from it. If the answer links to a shopping feed, that feed is now your storefront.

Worth trying

Chunk your key pages for AI retrieval. AI agents pull passages, not pages, so each section needs to make sense on its own. One guide suggests specific headings with proper HTML tags, breaking long tables and procedures into smaller pieces, describing important images in text, and keeping sections under the retrieval limit; Google's Agent Search defaults to 500 tokens. Start with your product and pricing pages.

New channels reward the teams that test early and spend carefully. AI answer ads will be no different.