Mike Ryan, Head of Ecommerce Insights at Smarter Ecommerce (smec), published an analysis of 383 million impressions finding that roughly 29% of exact match keyword impressions were expanded by AI Max search term matching by July 2026. PPC Land wrote the analysis up on 9 August 2026, including Ryan's description of that matching technology as "basically broad match".
The same campaigns showed close to 100% true exact match in early 2025. Square brackets still work as a keyword syntax. They have stopped working as a guarantee.
smec measured 383 million impressions from January 2025 to July 2026
Ryan's dataset counts how many impressions attributed to exact match keywords were in fact generated through AI Max search term matching rather than literal keyword matching. By July 2026 the answer was about 29%, up from effectively zero at the start of 2025. Ryan confirmed in the comment thread that the data covers EMEA only, and the campaigns are ecommerce Search campaigns with AI Max activated.
Most of the movement is recent. Roughly two thirds of the total expansion visible across the eighteen months of the series happened in its final four months.
The expanded share of exact match climbed in stages, not in one jump
Ryan published a two-series chart tracking genuine exact matches against impressions broadened by AI Max. This is how the expanded share moved across the period smec measured.
| Period in the smec series | Share of exact match impressions expanded by AI Max |
|---|---|
| First half of 2025 | Effectively 0%, with exact match at or near 100% |
| Around November 2025 | Roughly 10% |
| January 2026 | A brief retreat before climbing again |
| July 2026 | Close to 29%, with exact match at roughly 71% |
Read from the other side, genuine exact match starts the series near 100% and ends it at about 71%, with the steep part of the decline running from April 2026 onward.
Alphabet's Q2 2026 call put a 15% conversion uplift against the same feature
On Alphabet's Q2 2026 earnings call, Chief Business Officer Philipp Schindler said AI-powered campaigns delivered "an average of 15 percent more conversions or value on Search at a similar return on ad spend". Ryan set the smec expansion data beside that claim and asked of the newly matched searches: "But these new searches... is there any meat on the bone?" His objection was that the next 15% of conversion volume in an account already run tightly will not arrive as efficiently as the previous 15% did.
An earlier smec analysis of 250+ retail campaigns found roughly 35% lower ROAS
PPC Land also cited an independent smec analysis from November 2025, covering more than 250 retail campaigns, which put AI Max conversions at approximately 35% lower return on ad spend than traditional match types inside the same campaigns. Together the two smec numbers describe a trade: wider coverage at a worse rate per conversion. Whether it is worth taking depends on how much budget in a given account sits behind exact match keywords.
Why the 29% figure matters before 1 September 2026
Google has confirmed 1 September 2026 as the date when automatically created assets and campaign-level broad match auto-upgrade to AI Max. Any Google Ads account that leans on match type alone to keep budget away from loose queries is leaning on something that is already about a quarter porous in the campaigns smec looked at. The practical work is unglamorous: read the search terms report on a weekly cadence rather than a monthly one, and keep negative keyword lists as a living document instead of a quarterly cleanup.
What this means for Thai marketers
The smec sample does not include Thailand. Ryan confirmed the data is EMEA, and the source reports no Thai or APAC figures at all, so nobody should quote 29% as a Thai number. The mechanic is what travels. An account running exact match in Thai script will see expansion into Thai queries it never chose, and mixed Thai and English commercial vocabulary makes those expansions harder to spot by eye in a search terms report. For ecommerce accounts, which is where the smec sample sits, expansion tends to surface as traffic on adjacent product categories the store does not stock. Deciding whether those extra impressions are worth paying for is a question that only an account's own conversion data can answer, which is a reason to confirm that GA4 tracking is reporting cleanly before the auto-upgrade date rather than after it.
Frequently asked questions about AI Max and exact match
Does exact match still exist?
Yes, exact match still exists as a match type, but in the campaigns smec measured about 29% of its impressions by July 2026 came from AI Max search term matching rather than literal matching. The keyword still sets the starting point for the auction. It no longer defines the full set of queries that can trigger the ad.
Is this happening in Thailand too?
The source does not say. Ryan confirmed the 383 million impression dataset is EMEA only, and he declined to draw a regional comparison from the North American data he holds. No Thai or APAC breakdown appears anywhere in PPC Land's report.
Do I have to do anything before 1 September 2026?
Nothing is technically required of an advertiser, since the auto-upgrade happens on Google's side. What is worth doing is capturing the before state while it still exists: export search terms and current match type performance now, so that any change after 1 September can be measured against a real baseline rather than a memory.
Does the 15% uplift claim contradict the 35% lower ROAS finding?
Not directly, because the two figures measure different things. Schindler's 15% refers to additional conversions or value at a similar return on ad spend across AI-powered campaigns, while the November 2025 smec figure compares the return on ad spend of AI Max conversions against traditional match types within the same retail campaigns. More volume at a worse marginal rate can both be true at once.
What to do with this
If exact match has been the containment layer in an account, the honest reading of the smec data is that the layer is thinner than the interface suggests. Pull a search terms export, tag which queries the keyword would never have matched literally, and check what those queries actually converted. If the answer is not much, that is a negative list, not a debate. Anyone who wants a second pair of eyes on an account before the September change is welcome to get in touch.
