The last setting that kept a Search campaign out of Google’s AI answers stopped working on 4 September. Google is running a limited experiment in which standard Search campaigns built on exact and phrase match keywords can serve text ads inside AI Mode, a surface that until then took ads only from Performance Max, AI Max for Search, Shopping, or broad match paired with Smart Bidding.

Restrictive match types have done two jobs for years. They bought precision, and they bought exclusion: an advertiser who did not want budget flowing into conversational AI results could stay on exact and phrase and stay out. The experiment removes the second job. Whether an ad appears in AI Mode now turns on how Google reads the query rather than on how the campaign is built, and Google has published neither a definition of the test it applies nor a setting that declines it.

What did Google confirm about AI Mode ad eligibility?

Ginny Marvin, Google’s Ads Liaison, told Search Engine Roundtable on 4 September that the company had begun “a small experiment that recently kicked off”, and set out the rule in a comment on LinkedIn: exact and phrase match keywords are eligible to serve text ads in AI Mode, limited to cases carrying “explicit and direct user intent”. Asked a second time, she confirmed again that this was a new limited experiment run by the Google Ads team.

The behaviour was spotted first by Anthony Higman, founder of the agency AdSquire, who posted that Google was serving ordinary Search campaigns with restrictive match types in AI Mode and said he had confirmed it across multiple campaigns and tests. Higman disputed the framing of the change as small, on the grounds that the three routes previously available all shared one property: each moved query matching off the advertiser’s keyword list and into Google’s automation.

Marvin drew a boundary around what the test does not cover. AI Max for Search and Performance Max remain the products Google positions for complex conversational intent and for its newer AI Mode ad formats, including Highlighted Answers. Standard Search campaigns are being tested for text ads tied to clearer intent, which is a narrower slot than the one the automated campaign types occupy.

What Google has not supplied is the part that decides whether any given account is affected. There is no help page defining explicit and direct user intent, no list of qualifying query types, no named geography or vertical, no share of accounts, and no end date. The condition that governs eligibility exists, at the time of writing, in a comment thread.

Why has exact match stopped working as a control?

Exact match had already loosened well before this experiment reached AI Mode. An analysis of 383 million impressions across EMEA ecommerce Search campaigns, published in late July 2026 by Mike Ryan, head of ecommerce insights at Smarter Ecommerce, found that the share of exact match keyword impressions that were genuinely exact had fallen to roughly 71% by July 2026, from at or near 100% through the first half of 2025. Ryan described the search term matching technology responsible as broad match in substance.

That figure measures AI Max-enabled ecommerce campaigns rather than every account, and it is one practitioner’s dataset rather than a Google disclosure. It still puts a number on a shift advertisers had been describing anecdotally: roughly one impression in four attributed to an exact match keyword now arrives through expansion rather than a literal match.

The timing matters more than the percentage. Google auto-upgraded broad match campaigns to AI Max on 1 September, a migration confirmed to advertisers by email on 5 August. Three days later, the match types that advertisers had kept as the alternative to that automation became eligible for the surface the automation was built to reach. Taken together, the two changes narrow the gap between a tightly controlled campaign and an automated one from both directions within a single week.

There is a further wrinkle in how Google treats its two AI surfaces. Marvin told the PPCChat community in December 2025 that ads served within an AI Overview are matched against both the query and the Overview’s content, so only broad match or keywordless targeting can trigger them, even where exact match keywords sit in the account. AI Mode now works differently from its sibling surface, which means a campaign’s exposure varies by which AI result a searcher lands in.

What can a marketing team see in its reporting?

Google has not said whether reporting will separate AI Mode delivery, and at present it does not. Impressions and clicks from AI surfaces are folded into Search campaign totals, and AI Max traffic is credited against the exact and phrase keyword rows it expanded from, so a keyword row showing healthy volume may be reporting queries the advertiser never listed. For a B2B team running a narrow, high-value keyword set, that removes the audit trail between what was targeted and what was bought.

The practical work available is unglamorous and sits in tools that already exist. The search terms report remains the only place expansion becomes visible, and it rewards being read weekly rather than monthly while the experiment runs. Negative keyword lists and brand exclusions still apply. Teams comparing performance across the first week of September and the weeks before it are comparing two different eligibility regimes, and should label the break rather than read it as a campaign result.

The performance evidence on Google’s automated matching is mixed and mostly independent. An analysis of more than 250 retail campaigns published in November 2025 found return on ad spend roughly 35% lower with AI Max than with classic match types, a finding from agency testing rather than from Google, and one drawn from retail accounts rather than B2B. Google’s own claim for AI Max is a conversion lift of about 7% when the full feature set is used, which is not a head-to-head against the campaign structures it replaces.

Google has not said how long the experiment will run, how many accounts are in it, or whether an advertiser can tell from the interface that their account is one of them.