For as long as ads have run on Google's AI surfaces, exact and phrase match were the way you stayed off them. If you did not opt into AI Max, Performance Max for Search, or broad match with Smart Bidding, your restrictive match types kept your budget on the classic results page where you could read the search terms report and reason about every query. As of a test Google confirmed on 4 September 2026, that is no longer strictly true: standard Search campaigns using exact and phrase match keywords can now serve inside AI Mode, and there is no report segment that tells you when it happened.
This post is about what that change does to your reporting and what you can actually do about it while it is still a test. The short version: an exact or phrase match keyword row in your account may now blend three different kinds of traffic — literal matches, close variants, and AI Mode impressions — and standard reporting cannot pull them apart. That contaminates the clean match-type reads you rely on to make negation and bidding decisions. The right response is not to panic and pause; it is to baseline your affected campaigns now, learn the indirect signals that reveal AI Mode traffic, and keep the one control that still works — a tight negative keyword list — sharp.
What Google actually changed
Google has begun serving text ads from standard Search campaigns using exact and phrase match keywords inside AI Mode — a departure from the eligibility rule that had governed the surface since ads arrived on it. PPC Land, reporting the confirmation, framed it as exact and phrase match keywords gaining AI Mode ads in a Google test, and Search Engine Land described it as Google testing traditional Search campaigns in AI Mode. Before this, the only routes into those AI Mode placements were AI Max, Performance Max for Search, or broad match paired with Smart Bidding. The restrictive match types were, in effect, the opt-out.
The scope is deliberately narrow and deliberately vague. Google's Ginny Marvin said eligibility is limited to cases with "explicit and direct user intent", but Google published no definition of that phrase, named no countries, gave no end date, and released no performance data. That matters for how you weigh it: this is a small experiment with unknown account coverage, not an announced product rollout with a migration deadline. It could expand, it could change shape, or it could quietly end. Everything below is written for that reality — enough to act on if it reaches your account, calibrated so you do not over-correct for a test whose size nobody outside Google can see.
Why this contaminates your exact match rows
The practical damage is to what a keyword row means. Historically an exact match keyword row already blended two populations you could not fully separate: literal matches and close variants, the latter being the source of the close-variant wasted spend that plagues tightly-built accounts. With AI Mode now eligible on those same keywords, a single exact or phrase match row can aggregate a third population: AI Mode impressions. Literal matches, close variants, and AI Mode traffic all land in the same row, all counted together, with no flag on any of them. The number you read as "exact match performance" is now a weighted average of three surfaces that behave differently.
That is worse than close-variant creep because AI Mode is a genuinely different context. A user in a conversational AI surface arrives with different intent framing and a different path to your ad than someone scanning a classic results page, so blending them dilutes every derived metric you use to make decisions — CTR, conversion rate, cost per conversion, the lot. If AI Mode traffic converts worse (or better) than your classic Search traffic, the blended row hides it, and you tune bids and negatives against an average that describes no real audience. This is the same class of problem as the exact-match degradation into unrelated terms, one layer up: not just looser query matching, but looser surface eligibility on top of it.
The reporting blind spot: there is no AI Mode segment
The reason you cannot simply subtract AI Mode from the row is that no standard report isolates it. There is currently no column, filter, or segment that separates AI Mode impressions from ordinary Search impressions at the keyword level. You can still open the search terms report and read the actual queries — that discipline does not go away — but you cannot attribute a given query, click, or conversion to the AI Mode surface versus the classic results page. The query is visible; the surface it showed on is not. That is the blind spot: the data you would need to quarantine AI Mode traffic and judge it on its own terms is simply not exposed.
This compounds a trend the search terms report has been on for a while. Between redaction and AI-intent rewriting, the report already shows you less of the literal truth than it used to, which is why the visibility-ratio audit exists as a standing check. The AI Mode test adds a new dimension of opacity: not which queries are hidden, but which surface the visible ones ran on. Until Google ships a segment — and there is no commitment that it will — you are reasoning about match-type performance through a report that cannot tell you where a third of a row's impressions physically appeared. The correct posture is to stop treating the blended row as ground truth and start triangulating from indirect signals instead.
How to detect AI Mode traffic in your account this week
Because there is no direct readout, detection is a matter of watching for a cluster of indirect signals on campaigns you never opted into any AI surface. Start by baselining: for each exact and phrase match campaign, record its current impression volume, CTR, average CPC, and conversion rate over a stable recent window. You cannot measure a delta you never captured, and the single most useful thing you can do this week is write down what "normal" looked like before the test could touch your account. Do this per campaign, not just at account level, so you can localise any change later.
Then watch three things. First, unexplained impression growth on exact and phrase match keywords with no change in bids, budgets, or seasonality — new eligibility showing up as volume. Second, a CTR or CPC shift on long-established exact match keywords whose behaviour had been flat for months. Third, new entries in the search terms report that read like paraphrased, conversational, or intent-inferred queries rather than the literal terms you targeted — the fingerprint of AI-interpreted search terms rather than clean exact matching. None of these is proof on its own; seasonality and competition move all of them. But several of them arriving together, on campaigns you kept deliberately restrictive, is the strongest tell available while the surface stays unsegmented.
What still gives you control, and what doesn't
The instinct is to reach for match types to fix a match-type problem, but that is exactly the lever this test weakens. Exact match still governs which queries you match — the query must still clear Google's relevance bar — but it no longer governs the surface, so tightening match type does not reliably keep you out of AI Mode the way it used to. If the test holds, appearing in AI Mode stops being a setting you toggle and becomes a property of the query. That is why the broad-versus-phrase-versus-exact decision is increasingly about query control rather than placement control: the match type still shapes what you match, but the platform increasingly decides where the match is shown.
The control that does still bite is the negative keyword list. A negative blocks an eligible query regardless of which surface it would have shown on, so a tight, well-structured negative list is the one instrument that behaves the same whether the impression would have landed on the classic page or in AI Mode. That makes now a good moment to revisit your account, campaign, and shared-list structure and confirm your shared negatives are applied everywhere they should be. The caveat is the same one that dogs every AI surface: negatives match strings, and AI-inferred queries can arrive as paraphrases your list never anticipated, so lean on pattern-level negatives over one-off strings and keep reviewing what actually shows up. Match type is the weakened control here; negatives are the durable one.
How to respond while it's still a live test
The proportionate response is to instrument, not to retreat. Pausing your exact and phrase match campaigns to dodge AI Mode would throw away your best-controlled traffic to avoid a change whose magnitude Google has not disclosed — a large, certain cost to avoid a small, unquantified one. Keep your structure steady, hold the baseline you captured, and let the indirect signals tell you whether the test has actually reached your account and whether it is moving your numbers enough to matter. Most accounts, most of the time, will see nothing measurable from a small experiment; a few will see real dilution, and only the baseline lets you tell which camp you are in.
If the signals do fire and your affected campaigns show meaningful CPA or intent dilution, isolate before you amputate. Segment by campaign to find where the change concentrates, tighten negatives around the paraphrased queries surfacing in the report, and only then consider structural moves like splitting a campaign to quarantine the affected keywords. Re-verify the test's status as you go, because a small experiment can expand into a default without much warning — and if it becomes permanent, this stops being a monitoring exercise and becomes a live constraint on how much placement control the restrictive match types can still promise. Until then, baseline, watch, and keep your negatives sharp.
Related reading
For the underlying query-matching problem this sits on top of, see exact-match degradation into unrelated search terms. For how to choose match types when placement control is eroding, see the broad, phrase, and exact match-type decision, and for reading a report that increasingly shows AI-inferred rather than literal queries, see negating AI-interpreted search terms and the visibility-ratio audit.