Keyword-based Search campaigns are being tested in AI Mode
Google Ads Liaison Ginny Marvin has confirmed a small experiment that allows exact match and phrase match keywords in traditional Search campaigns to serve text ads inside AI Mode. The confirmation arrived on September 4, 2026 as a LinkedIn comment rather than an official announcement. Eligibility is conditional and limited to cases with explicit and direct user intent.
Restrictive match types started appearing on a new surface
The first person to spot the behaviour was Anthony Higman, founder and CEO of ADSQUIRE. On September 3, 2026 he posted on LinkedIn that ordinary Search campaigns using restrictive match types were showing up inside AI Mode, and Search Engine Land reports that he verified the pattern across more than one campaign and test. A day later Ginny Marvin, the Google Ads Liaison, confirmed on the same platform that a small experiment had recently launched allowing exact match and phrase match keywords to serve text ads in AI Mode. The form of that confirmation is worth underlining: there is no Google announcement and no updated help document behind it, only a comment on a social network. By Marvin's account the eligibility does not cover every query either, and applies where there is "explicit and direct user intent".
A door that used to open only through automation
To see why this registers at all, it helps to recall the previous state. According to PPC Land, entry into AI Mode ran through automation: Google's documentation pointed advertisers towards broad match, or towards keywordless targeting via AI Max for Search, Performance Max, Shopping and Dynamic Search Ads. The stated reasoning was technical, namely that the triggering queries tend to be long and exploratory. PPC Land also notes that the eligibility Marvin had described earlier was tied to AI Max for Search with text customisation, or to Performance Max. In other words advertisers reached this surface by choosing a product, not by selecting a placement. The experiment loosens that logic a little. One part of the picture has not moved, though: the newest ad formats such as Highlighted Answers remain exclusive to the automated products. In Google's framing, AI Max and Performance Max for Search are built to capture complex, conversational intent and to reach those newest formats.
Your ads may be running on a surface you cannot measure
The real issue is not reach, it is visibility. Google Ads currently gives advertisers no way to separate AI Mode delivery from ordinary Search delivery. The consequence is direct: in an account inside the experiment, impressions and clicks blend into existing keyword rows, and the effect cannot be isolated within those rows. The scale involved is not trivial. AI Mode reached 75 million daily active users as of February 2026, and 1 billion monthly users as of Google I/O 2026. Against that, PPC Land reports Google disclosed nothing about geographic scope, vertical restrictions, the share of accounts affected, or how long the test will run. So what advertisers have is a variable that is small in scale, indefinite in duration, and invisible in reporting. Reading it as a permanent product change would be wrong, but noting it as one possible explanation for movement in an account is fair.
Time to check unexplained movement by hand
For advertisers in Türkiye the practical conclusion is clear: accounts running restrictive match types can now appear on this surface too, and that appearance never shows up as its own line in the report. Unexplained movement in keyword performance may therefore be connected to this test. The work comes down to three items. First, run a period comparison, putting impressions, clicks and conversion rates side by side with the weeks before early September to see whether anything has drifted. Second, read the search terms report more often than usual, because unexpected terms surfacing in an exact match campaign is the earliest signal available. Third, check intent alignment on non-brand traffic manually, meaning a human look at whether an incoming query genuinely carries purchase intent. Nothing here calls for alarm, yet while a variable you cannot measure is live, a weekly reading habit is cheap insurance.
The UNALSOFT view
From where we sit the story is not that an experiment exists, it is that the distance between delivery and measurement has widened. On the accounts we run, the first thing our ad management work looks at is queries rather than totals, because once a surface stops separating out in reporting, search terms are the only verification left. Our advice for businesses in Türkiye is plain: do not change budgets or match types this week, measure first. A hurried reaction to uncertainty usually costs more than the uncertainty itself.
Seeing movement in Google Ads you cannot explain?
We can read your search terms report and period comparison together and work out what actually changed. A short conversation is enough to start.