Google is serving exact and phrase match ads in AI Mode. Here is what actually changes

For most of 2026 the price of appearing in Google's AI Mode was giving up keyword control. Performance Max, AI Max or broad match with Smart Bidding were the ways in, and a tightly run exact-match account was, by design, outside. A small experiment confirmed on 4 September quietly removes that condition — and the most useful thing about it has nothing to do with joining the test.

What Google confirmed, and what it did not

The sequence is short. An advertiser, Anthony Higman, reported seeing exact and phrase match keywords producing impressions in AI Mode and said he had confirmed it across multiple campaigns. On 4 September 2026 Ginny Marvin, Google's Ads Liaison, confirmed it was real, describing it as "a small experiment that recently kicked off" and saying eligibility is limited to cases with explicit and direct user intent.

That is the entire disclosure. Google named no countries, published no definition of "explicit and direct intent", set no end date, released no performance data and offered no opt-in or opt-out. There is no setting in your account that has changed and nothing you can toggle.

Which makes the obvious question the wrong one. The useful question is not "how do I get into this test" — you cannot. It is "what does this tell me about where Google is taking match types", and that is worth more than the test itself.

The eligibility picture, before and after

Until this experiment, reaching an AI Mode ad placement required handing Google some degree of keyword control. That was the trade, and it was rarely stated as plainly as it deserved. Here is the full picture as it stands after 4 September.

The bottom three rows are the ones to sit with. Whatever else changes, you still cannot choose this placement, refuse it, or measure it separately.

Why a small test matters more than usual here

Through 2026 the consistent message from Google has been that AI surfaces are where query volume is going, and that reaching them means loosening keyword control. Advertisers who declined — and there are many, running tight exact-match accounts with years of negative keyword work behind them — were being told, implicitly, that they were opting out of a growing share of commercial search.

This experiment decouples those two things. If it graduates, the sentence "you need AI Max or Performance Max to appear in AI Mode" stops being true, and the case for the AI Max migration running through September gets narrower: it becomes a bidding and matching argument on its own merits, rather than an access-to-surface argument. For an account that migrated reluctantly, that is worth knowing before the next quarterly review.

The caution is equally important. "Small experiment" is a real description, not a hedge. Google runs many tests that never ship, and the eligibility condition — explicit and direct intent, undefined — is exactly the sort of qualifier that can be narrowed to near-nothing before a wider rollout. Nothing in this justifies restructuring an account.

Where AI Mode ads actually appear, by keyword value

The most useful published data on AI Mode ad load is not about match types at all. A 2026 study of AI Mode results found ads appearing in about 25.5% of responses overall, and nearly 30% of ad-eligible queries. But the distribution underneath that average is steep, and it is the part worth acting on.

Where an ad does appear, two ads appear together 71.1% of the time and a single ad 28.9% of the time. AI Mode itself is reported at roughly 75 million daily users.

Read the table as a filter rather than a benchmark. The honest answer to "should I care about AI Mode ads?" is not universal — it depends almost entirely on what a click costs in your category. A business paying $1.20 a click is looking at a surface that shows an ad on a quarter of responses and that it cannot yet measure. A business paying $35 a click is looking at a surface that shows an ad on more than half of them. Those two businesses should not be given the same advice, and most coverage of AI Mode gives them exactly that.

You still cannot see it, so here is what you can see

Google Ads offers no AI Mode segment, no filter and no impression-share breakout. Ads that serve in AI surfaces are reported as ordinary Search impressions. Google has said it is thinking about what reporting should look like and will announce more when there is something to share, which is a genuine answer and also not one you can build a plan on.

So the measurement problem is real, and the choice is between guessing and using proxies. Four proxies are worth the effort, in this order:

1. Annotate 4 September 2026. Everything below depends on having a before and an after. Put the marker in your analytics and your reporting deck now, while you still remember why.
2. Watch impressions on exact-match ad groups against flat query volume. If an exact-match ad group starts gaining impressions without a corresponding move in the keyword's search volume, a new surface is the most likely explanation. Not proof — seasonality and competitor exits look similar — but it is the cleanest signal available.
3. Read your search terms report differently. AI Mode queries are longer and more conversational than classic search queries. A negative keyword list built for three-word queries frequently fails to fire on fifteen-word ones, because the offending term sits inside a phrase your negatives never anticipated. This is worth an audit regardless of what the experiment does.
4. Compare on-site behaviour by landing page, not by campaign. Visitors arriving after an AI-assisted research process behave differently: fewer pages per session, shorter consideration on site, a higher rate of arriving with a specific question already formed. Tracking assisted conversions and time-to-conversion at the landing page level will show a shift before your campaign metrics do.

None of these is conclusive on its own. Together they are enough to notice a change, which is all you actually need, because the response to noticing is the same either way.

What to actually do this month

In priority order, and deliberately short.

1. Change nothing structural. No match-type migration, no new campaign, no budget reallocation on the strength of an unscoped experiment. This is the single most common expensive mistake made after a Google announcement.
2. Audit negatives against long conversational queries. Take your twenty most expensive search terms from the last ninety days, write out how someone would ask the same thing conversationally, and check whether your negatives still catch the ones you do not want.
3. Fix the landing page for answer-engine traffic. A visitor who arrives from an AI answer has already had part of the conversation. The page that meets them should confirm and advance rather than re-explain, which is a different brief from a classic paid search landing page. We covered the specific differences in landing pages for AI-referred traffic.
4. Revisit your AI Max position with the new information. If you migrated primarily to get AI surface access, the justification is weaker than it was on 1 September. If you migrated for matching and bidding reasons, nothing has changed.
5. Check brand terms specifically. Brand queries are the clearest case of explicit, direct intent, which makes them the most likely to be in scope for this experiment. They are also where an unexpected impression increase is easiest to mistake for a competitor incursion.

What not to do

Do not read this as permission to abandon broad match. The experiment says exact and phrase can reach AI Mode in some circumstances; it says nothing about them reaching it as often, as cheaply or as profitably as broad match with Smart Bidding does. Those are different claims and Google published data on none of them.

Do not build a reporting process around a placement Google has not defined. Several tools now market AI Mode visibility metrics for paid search. Google does not expose the underlying data, so anything presented as an AI Mode impression count is a model, and should be labelled as one in any report that reaches a client or a board.

And do not let the novelty pull attention away from the thing that still decides most accounts' results. The gap between a good and a poor campaign structure, or between a converting and a non-converting landing page, remains larger than the difference this experiment will make to almost any account this year.

Frequently asked questions

Can exact match keywords show ads in AI Mode now?

Inside a limited experiment, yes. Google's Ads Liaison Ginny Marvin confirmed on 4 September 2026 that exact and phrase match keywords are eligible to serve text ads in AI Mode, describing it as a small experiment that recently kicked off, with eligibility limited to queries carrying explicit and direct user intent. Google has not defined that phrase, named the countries involved, given an end date or published any performance data.

Do I still need AI Max or Performance Max to appear in AI Mode?

Before this experiment, effectively yes: the routes into AI Mode ad placements were Performance Max, AI Max for Search, Shopping campaigns, or broad match with Smart Bidding. The experiment shows Google is willing to serve restrictive match types there too. Until it graduates beyond a test, treat the requirement as unchanged for planning purposes, but stop treating AI surface access as the sole reason to adopt AI Max.

How do I see how many of my clicks came from AI Mode?

You cannot. Google Ads provides no segment, no filter and no impression share breakout for AI Mode or AI Overviews, and ads serving in those surfaces are reported as ordinary Search impressions. Google has said it is still deciding what that reporting should look like. The practical workaround is proxy signals: annotate 4 September 2026, watch exact-match impression volume against flat query volume, and compare on-site behaviour by landing page.

Should I restructure my campaigns because of this?

No. This is an unscoped experiment with no published performance data, and restructuring an account on the strength of one is the most common expensive mistake made after a Google announcement. The proportionate responses are to audit negative keywords against long conversational queries, fix landing pages for answer-engine traffic, and revisit whether your reason for adopting AI Max still holds.

How often do ads actually appear in AI Mode?

A 2026 study found ads in about 25.5% of AI Mode responses, and close to 30% of ad-eligible queries. The distribution is steep by keyword value: 24.33% of responses for keywords with CPCs below $2, 32.45% between $2 and $10, and 53.56% at $10 or more. Where ads appear at all, two ads show together 71.1% of the time.

Can I opt out of showing ads in AI Mode?

No. There is no setting to target AI Mode specifically and no setting to exclude it, either before or after this experiment. If your campaign type and match type are eligible, the placement is available to Google's systems. This is the same position that has applied to AI Overviews throughout 2026.

The takeaway

One small experiment does not change an account, but it does change an argument. The claim that reaching AI Mode requires surrendering keyword control is no longer reliably true, which narrows the case for AI Max to its actual merits in matching and bidding. Meanwhile the things you can act on today are unglamorous: annotate 4 September, audit your negatives against conversational queries, fix the landing page that meets AI-referred visitors, and check your average CPC to decide whether AI Mode is a live channel for you or a rounding error. It is one or the other, and the answer is vertical-specific.