Exact and Phrase Match Ads Are Serving in AI Mode (September 2026): What Solo Founders Should Know Before Accepting the AI Max Upgrade

On September 4, 2026, Google’s Ads Liaison Ginny Marvin confirmed to Search Engine Roundtable that exact and phrase match keywords are eligible to serve text ads in AI Mode. Until that confirmation, the only documented routes into that inventory were AI Max, Performance Max for Search, or broad match paired with Smart Bidding.

What exactly did Google confirm on September 4, 2026?

A live, limited test — not a launch. Marvin’s words to Barry Schwartz: “This is a small experiment that recently kicked off.”

The sighting came first from Anthony Higman, who posted on LinkedIn that Google was serving “regular pure search campaigns with restrictive match types in AI mode,” verified across multiple campaigns on his side. Marvin added the qualifier that matters most: eligibility is “limited to cases where there is explicit and direct user intent,” while AI Max and PMax for Search remain the products “designed to capture complex conversational intent and access our latest ad formats (like Highlighted Answers).”

Note what Google did not publish: no countries, no end date, no traffic share, no performance data, and — the one that decides everything below — no reporting segment separating AI Mode impressions from the rest of your Search numbers.

Why does match type decide whether your ad shows in AI Mode?

Because match type is a proxy for how finished a query is when it arrives.

At one end of AI Mode sits a fully specified query: “emergency plumber Tucson.” Nothing needs resolving before an ad can be matched — the intent is explicit and directional, precisely the condition Marvin named. At the other end sits “my water heater is leaking and I’m renting, who pays for this,” which has to be decomposed into several sub-questions before any commercial intent surfaces. An exact match keyword can only bind to the first kind. There is no keyword to match against a query that has not finished becoming a query yet.

That gives you a rule you can extrapolate with: the more decomposition a query requires before it yields a purchase-shaped sub-question, the less of your AI Mode exposure this experiment touches — regardless of how many keywords you add. A same-day locksmith sits almost entirely inside the eligible slice. A B2B SaaS company whose buyers spend six weeks comparing options sits almost entirely outside it. Same experiment, opposite value, and the difference is query shape rather than budget.

Does this mean a small advertiser can skip the AI Max upgrade?

For now, largely yes — and mostly because you could never have measured the upgrade in the first place.

Read this through direct response — the school that refuses to buy reach it cannot tie back to a sale — and the decision resolves quickly. It is the right lens here because the missing ingredient is attribution, not reach. A brand-building lens would shrug: mental availability is bought in aggregate, so an unattributable impression in a new surface is still worth having. Direct response will not accept that, because the advertiser here runs $1,500 to $3,000 a month and cannot absorb a mistake it discovers two quarters later.

Through that lens, the September 4 news is straightforwardly good. Google handed you inventory at no structural cost: no restructure, no surrendered match-type control, no automated asset generation. Any AI Mode impressions your exact and phrase campaigns pick up land inside a campaign whose CPA you already track weekly. New reach on an existing measurable line is the direct-response ideal.

The AI Max path is the opposite trade. It reaches the conversational middle of AI Mode and unlocks formats like Highlighted Answers, but it takes your match-type control with it and reports the result as a blended campaign number you cannot decompose. Put real figures on that: a local service business spending $2,000 a month at a $45 CPA books about 44 jobs. If the upgrade drifts CPA to $54 — a 20% move, unremarkable for a campaign entering a learning phase — that same $2,000 books 37. Seven jobs a month, 21 across a quarter, and no way to prove the upgrade caused it rather than Q4 auction pressure.

Exact and phrase match or AI Max — which route fits a three-person team?

Run exact and phrase now; treat AI Max as a test you schedule deliberately, not a default you accept from a notice.

• If your best converting queries run three to five words and end in a transactional noun — quote, near me, pricing, emergency — your campaigns already sit on the eligible slice. Change nothing structural.
• If your revenue comes from queries that read like full sentences, AI Max remains the only documented route to that inventory. Run it on one campaign with a fixed end date, not account-wide.
• If you cannot tell which you are, your search terms report settles it in ten minutes. That is step one below.

How should you respond in your account this week?

Seven steps, all finishable in an afternoon.

1. Count your short queries. Pull 90 days of search terms, sort by conversions, and count how many of your top 20 converting queries are five words or fewer. That ratio is your exposure to the eligible slice.
2. Baseline before anything moves. Export weekly impressions, clicks, CTR and CPA for your exact and phrase campaigns today. If the experiment widens or ends, this export is your only evidence anything changed.
3. Log the upgrade notices you decline. Google began surfacing in-account notices in September 2026 encouraging voluntary AI Max upgrades. Screenshot each one with its date so a future CPA argument has a timeline.
4. Do not add keywords because of this news. Eligibility is a serving change, not a coverage gap. A longer list cannot make an unfinished query matchable, and you pay for the bloat in negative-keyword maintenance.
5. Rewrite headline 1 to be answer-shaped. A text ad beside a generated answer competes with a paragraph, not with nine other ads. Lead with the concrete thing — price, response time, service area — not the brand name. If four headline variants per ad group is your throughput wall, creative tooling solves the volume; which promise leads is still your call.
6. Diarize a late-October re-pull. Compare against step 2. Rising impressions with flat clicks and falling CTR is the fingerprint of a surface where fewer people click — not a sign your copy got worse.
7. Leave broad match alone. Broad plus Smart Bidding campaigns were already eligible for AI Mode. Keeping their numbers separate from your exact and phrase lines is what makes step 6 readable.

Which dates belong on a small advertiser’s calendar right now?

Four dates, only one of which is this experiment.

• September 1–30, 2026 — the AI Max auto-upgrade window, which Search Engine Land reports covers campaign-level broad match and Automatically Created Assets campaigns. Dynamic Search Ads upgrades moved to February 2027, so a DSA-heavy account has more runway than the notices imply.
• October 1, 2026 — the practical start of Q4 auction inflation. Any CPA movement after this date has two candidate causes, which is why step 2 has to happen in September.
• Late October 2026 — your re-pull date. Long enough for a serving change to show in weekly averages, early enough to act before November spend.
• No published end date — Google gave none for the experiment. Unannounced withdrawal is a real possibility, so treat any lift you attribute to it as borrowed rather than banked.

Is Google’s “small experiment” framing worth believing?

Probably accurate about volume, and misleading about significance. Those are different claims, and Google only made the first one.

Higman pushed back publicly on this point, arguing the change matters because “before this the ONLY way to show in those places was through ai max, p-max or broad and smart bidding.” He is right about the structure even if Google is right about the traffic. “Small” is doing quiet work in that sentence — it describes the test’s footprint while implying the news is minor, and an experiment that removes a product’s exclusivity is not minor.

The timing is worth naming as a reading rather than a claim: the experiment surfaced in the same month Google is running in-account notices asking advertisers to upgrade to AI Max voluntarily. Whether that sequencing is deliberate is unknowable from outside. What a small advertiser can act on is duller — an undated experiment is a setting that can be withdrawn without notice, so build no plan that requires it to persist. When you next sit down with your campaign planning tools, the question to answer is whether the account survives losing this surface entirely, not how much you can extract from it this month.

Frequently asked questions

Will my exact match keywords suddenly get more impressions?

Possibly, but not dramatically. Google called the test small and limited it to queries showing explicit and direct intent — a narrow band of AI Mode traffic. If you see a large impression jump in September, check seasonality, a competitor pausing, or a budget change before crediting this experiment.

Can I see AI Mode traffic separately in Google Ads reports?

No. Google has published no reporting segment isolating AI Mode impressions from other Search inventory. That absence is the most important fact in this story: AI Mode performance arrives blended into your existing Search numbers, which is why a September baseline export is worth more than any analysis you could run in November.

Should I accept the AI Max upgrade notice in my account?

Not reflexively, and not account-wide. If your converting queries are short and transactional, this experiment already gives you exposure without the trade. If your buyers search conversationally, test AI Max on one campaign with a defined end date and a pre-recorded CPA baseline, so the result is readable either way.

Does this change anything for Performance Max?

Nothing directly. PMax for Search was already eligible for AI Mode and keeps access to formats like Highlighted Answers that exact and phrase campaigns do not get. What changes is the argument for it: AI Mode exposure alone is no longer a reason to adopt PMax, so it now has to win on performance you can measure.