Exact match keyword expansion

Exact Match was expected to produce higher CTR and CR from more precise intent, while Phrase Match could keep reach and cheaper clicks in broader segments.

Observational comparisonComparison: Exact and Phrase ran together for four weeks7 performance snapshots
Observationaltest design
ComparisonExact and Phrase ran together for four weeks
KPICTR / CR / CPA-CPO / ROAS-CAC
PeriodFour-week concurrent comparison

Experiment context

Hypothesis. Exact Match was expected to produce higher CTR and CR from more precise intent, while Phrase Match could keep reach and cheaper clicks in broader segments.

How the test ran. Both match types were active in the same campaigns; results were compared by region, project and direction.

Comparison design. Exact and Phrase ran together for four weeks; no traffic split was used.

Visual summary — Exact change vs Phrase

Each percentage below shows Exact Match performance relative to Phrase Match for the same geo and service. Negative CPA is good because Exact is cheaper per conversion.

Geo / serviceCTRCRCPAWinner
DE service3+25.7%+7.2%+17.6%Phrase
EU service3+25.5%+13.1%-21.2%Exact
WW service3-25.7%-37.8%+34.3%Phrase
US service3+30.0%+46.4%-23.6%Exact
EU service6+102.2%+2.1%-1.7%Exact
US service6+92.4%+3.6%-2.1%Exact
EU game-3 service11+116.6%-9.5%+19.2%Phrase
Performance results7 snapshots · Click to expand
DE — game-2 — service3
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact5,81026,43111,827.45$2.04$21.98%295.045.08%18,526.35$40.09$62.79$
Phrase1,99611,4093,226.03$1.62$17.49%94.674.74%7,560.92$34.08$79.87$

Result. DE: Phrase remains preferred because Exact improved CTR and CR but increased CPA 17%.

EU — game-2 — service3
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact9933,6691,378.43$1.39$27.06%47.174.75%3,455.20$29.22$73.25$
Phrase1,3576,2942,111.92$1.56$21.56%56.974.20%3,772.99$37.07$66.23$

Result. EU: Exact won across CTR, CPC, CR and CPA; roll out Exact.

WW — game-2 — service3
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact4062,666956.44$2.36$15.23%13.883.42%1,049.66$68.89$75.62$
Phrase5,60127,30615,807.37$2.82$20.51%308.055.50%19,600.29$51.31$63.63$

Result. WW: Phrase won; Exact had 26% lower CTR, 38% lower CR and 34% higher CPA.

US — game-2 — service3
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact4,16518,09417,592.44$4.22$23.02%477.3111.46%30,154.85$36.86$63.18$
Phrase4,80727,13618,151.73$3.78$17.71%376.167.83%31,886.05$48.26$84.77$

Result. US: Exact won; CTR increased 30%, CR increased 46% and CPA decreased 24%.

EU — game-2 — service6
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact1,88723,1082,349.57$1.25$8.17%27.421.45%2,432.22$85.69$88.70$
Phrase1,06026,2291,314.80$1.24$4.04%15.091.42%1,443.75$87.16$95.68$

Result. EU: Exact won on service6, with roughly 2× CTR and slightly lower CPA.

US — game-2 — service6
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact1,21717,1783,100.56$2.55$7.08%45.073.70%3,737.47$68.79$82.93$
Phrase58915,9931,476.93$2.51$3.68%21.013.57%2,201.35$70.30$104.78$

Result. US: Exact won on service6, with 92% higher CTR and slightly lower CPA.

EU — game-3 — service11
Match typeClicksImpr.CostAvg. CPCCTRConversionsConv. rateConv. valueCPAconv value / conv
Exact4062,666956.44$2.36$15.23%13.883.42%1,049.66$68.89$75.62$
Phrase2223,160484.52$2.18$7.03%8.383.78%536.85$57.79$64.06$

Result. EU: Phrase won; Exact doubled CTR but produced lower CR and 19% higher CPA.

Conclusions

Overall verdict: market- and direction-specific. Exact won 4 of the 7 retained geo/direction comparisons, while Phrase won 3. Exact was consistently stronger for game-2 service6 and won game-2 service3 in EU and US; Phrase remained preferable for service3 in DE and WW and for game-3 service11 in EU. Use regional match-type decisions rather than one account-wide rollout.

Limitations. Exact and Phrase ran simultaneously without a randomized traffic split. Traffic allocation, query overlap and cannibalization may affect the observed differences.