If you’ve run Google Ads for a while, you’ve probably noticed your keywords don’t mean quite what they used to. In our accounts, only 35% of the searches exact match keywords triggered were the keyword word for word. The rest was Google’s interpretation.
Match types are broadening, and that part isn’t news. What we really wanted to know was whether it’s helping or hurting.
So we decided to take matters into our own hands. We pulled nearly 50,000 exact and phrase keyword-and-search-term pairs from three of our managed Google Ads lead-gen accounts, then zeroed in on the new searches Google started matching to our keywords, ones that weren’t being matched three years ago.
Turns out, it all depends on the match type. Leads from brand new exact match searches cost 7% less, while leads from brand new phrase match searches cost 28% more.
Here’s what we found, and what it means for your account.
Key takeaways
- Exact match is semantic, but new queries cost less. In New Path Digital’s study of three lead-gen Google Ads accounts, 35% of exact-match query volume was identical to the keyword in 2025–26. Conversions from exact-match queries first matched this year cost 7% less than those from existing queries ($109 vs. $116), driven by cheaper clicks.
- Exact match is BMM-like, not BMM. 75% of exact-match query volume contained every word of the keyword (the old broad match modifier rule), compared with 41% of phrase-match volume.
- Phrase match drifted, and it costs more. Phrase-match queries sharing no words with the keyword grew from 2% to 9% of volume, and conversions from new phrase-match queries cost 28% more ($209 vs. $163) because they converted less often.
- What advertisers should do: launch new lead-gen accounts on exact match, audit new phrase-match queries before scaling, steer Google with negatives and pinned ad copy, and import qualified-lead outcomes instead of raw form fills.
What broadening match types cost us: existing vs. new queries
We call queries “existing” or “new” throughout. An existing query showed up in our search term reports for the exact same keyword in both 2022–23 and 2025–26. A new query triggered that keyword only in 2025–26. Our leads are form fills from non-brand keywords, and every campaign ran on Maximize Conversions.
Here’s how the numbers compare:
| Match type | Query volume identical to the keyword, 2022–23 | Query volume identical to the keyword, 2025–26 | Existing query CPL | New query CPL |
|---|---|---|---|---|
| Exact | 29% | 35% | $116 | $109 |
| Phrase | 7% | 5% | $163 | $209 |
Total volume from 2025-2026 keywords, for reference:
| Match type | Query type | Keyword-and-search-term pairs | Pairs that spent money | Conversions |
|---|---|---|---|---|
| Exact | Existing | 2,013 | 588 | 344 |
| Exact | New | 6,129 | 1,055 | 189 |
| Phrase | Existing | 3,430 | 649 | 190 |
| Phrase | New | 38,332 | 3,234 | 375 |
Across our client datasets, exact match was already mostly non-literal in 2022–23, while phrase match expanded its semantic matching. For our clients, new exact-match queries were strong additions:
Conversions from them cost 7% less than conversions from existing queries
Cheaper clicks ($10 vs. $11) more than covered a slightly lower conversion rate (9% vs. 10%).
Our read: when Google stretches an exact keyword to a query with the same intent, you may get a similar searcher at a lower price.
For example, “military disability lawyer” matched the new, longer query “va lawyer for disability.” Those clicks cost about $13 each, ~40% cheaper than the keyword’s average. 
On the other hand, conversions from new phrase match queries cost 28% more and ate up ~75% of phrase match cost. For our clients, new phrase-match queries were more expensive:
The extra cost came from a dip in conversion rate (8% vs. 10%), not click price ($16 vs. $16)
This suggests semantic phrase matching may be a rougher fit for lead generation businesses. Take the phrase keyword “veterans legal services,” which matched the new query “jag fort bragg nc.” They share no words, and JAG offices typically provide free legal help to service members, so that search likely comes from someone looking for a free option, not a paying client. If you audit one thing, audit new phrase-match queries.
How far exact and phrase match stray
Those conversion results raise a bigger question: how far does Google actually stretch each match type? Google says an exact-match keyword can trigger ads on searches with the same meaning or intent, even when the wording differs, and there’s no opt-out of close variants.
To see how far Google stretches its semantics, we sorted every impression in both periods (2022–23: 75,698 exact and 119,808 phrase; 2025–26: 74,786 exact and 145,453 phrase) into five buckets by comparing its query with the closest keyword of the same match type in its campaign. This only accounts for visible queries in Google’s search term report.
Exact match
| How the search query relates to its keyword | 2022–23: share of query volume | 2025–26: share of query volume |
|---|---|---|
| Identical to the keyword | 29% | 35% |
| Identical to the keyword, with extra words | 12% | 9% |
| Identical keyword structure, with a one- or two-word synonym | 10% | 10% |
| Shares some words | 49% | 46% |
| No words in common | <1% | <1% |
Surprisingly, exact match didn’t loosen across our client accounts. Queries identical to the keyword rose from 29% to 35%. Almost half of exact-match volume shares only some of the keyword’s words, but 54% is the keyword itself, the keyword plus extra words, or a one- or two-word synonym swap.
If that pattern sounds familiar, it should. It’s a lot like the retired broad match modifier, a match type plenty of advertisers still miss. Notably, 75% of exact-match query volume contained every word of an exact keyword, compared to just 41% of phrase-match query volume. So if you miss BMM, take some solace. Exact match is doing a decent impression.
Phrase match
| How the search query relates to its keyword | 2022–23: share of query volume | 2025–26: share of query volume |
|---|---|---|
| Identical to the keyword | 7% | 5% |
| Identical to the keyword, with extra words | 18% | 10% |
| Identical keyword structure, with a one- or two-word synonym | 6% | 5% |
| Shares some words | 68% | 70% |
| No words in common | 2% | 9% |
Phrase match is where the drift shows. Queries sharing some or no words with the keyword grew from 70% to 79% of volume, while queries that were the keyword with extra words or synonyms fell from 24% to 15%.
So exact match tightened while phrase match loosened, wandering further into the vaguely related corners Google decided were relevant.
One caveat is that we could only analyze the search terms visible in the report. Nearly 43% of eligible search terms were hidden by Google, mostly on phrase match. So the “shares some words” and “no words in common” shares could be even larger if we could see the hidden terms.
Is exact match basically phrase match now?
Not really. To see why, think about what each match type was originally built to do:
Exact match: show your ad for your keyword and its near twins.
Phrase match: show your ad when your keyword appears inside a longer search, like “disability lawyer” inside “disability lawyer near me.”
Here’s what today’s exact match actually brings in:
Your keyword or a near twin (about 45%): the same words, or a swap like “attorney” for “lawyer.” That’s the exact match job.
Your keyword inside a longer search (about 9%): that’s the phrase match job, and it’s a small slice.
Everything else (about 46%): searches that share some of your keyword’s words, but not the same shape.
So where does exact match land? In a middle ground, and the closest match is the retired broad match modifier. BMM required every keyword word (or a close variant) to appear in the query, in any order, with extra words allowed.
Run that definition against our query volume and 75% of exact-match queries pass, against 41% of phrase-match queries. It’s not a perfect match, though. The other 25% of exact-match volume didn’t contain every keyword word, which is Google reading meaning where BMM would have needed the words.
So think of exact match as looser than literal exact, tighter than phrase, and BMM-like with some semantic overflow.
What this means for advertisers
If match types are about meaning, managing an account is about communicating:
Treat keywords as intent signals. A keyword marks the center of an intent, and the match type sets how far Google can wander. Choose the match type for how much interpretation you want.
Steer with negatives and ad copy. Read the search terms and add negatives when the intent is clearly wrong. Then let your ad copy do some of the targeting: write headlines (and pin the ones that matter) so every version of the ad says plainly what your business does and who it’s for. A loosely matched query that lands on a responsive search ad stitched from the wrong headlines can pull in clicks from people the ad was never meant for, and you may never know. Creative is targeting in 2026, and paid search advertisers can borrow the mindset Meta advertisers have had for years.
Feed the system the outcome you value. A semantic system will happily find more of whatever you reward, so reward qualified leads, not just form fills. Use conversion adjustments in Google Ads to retract form fills or calls that counted as conversions but missed the mark.
Consider starting with exact match only. For smaller accounts, or larger ones just getting started, see how much qualified volume you can wring from semantic exact match before adding phrase match. Exact can help you build the conversion signals phrase match needs to thrive.
To run a new-query audit, export search terms with keyword, match type, cost, and conversions. Label each pair existing or new, then compare cost per lead within each match type before you look at totals. Add CRM outcomes, review the highest-spend new queries for intent, and repeat every six months.
The catch: a cheap form fill isn’t always a good lead
Every figure here is cost per form fill, not cost per qualified lead, enrollment, sale, or signed deal. When Google reads intent for you, the check on whether it read it well is what happens after the form fill. Import qualified-lead and revenue outcomes from your CRM through offline conversion import, add qualifying questions to your forms, and compare match types on qualified outcomes, not just conversion cost.
Bottom line: match types are intent settings now
Google has spent more than a decade moving match types from words to meaning. In our data, conversions from new exact-match queries cost 7% less while conversions from new phrase-match queries cost 28% more, so “more meaning” isn’t automatically good or bad. It depends on how well Google understands your business, and how well you understand your business decides how successful you’ll be as a paid search advertiser in 2026 and beyond.
About New Path Digital
New Path Digital is a full-service advertising agency located in High Point, North Carolina. We’ve been around since 2015 and advertising with Google Ads for longer, and we spend a lot of our days inside accounts like the ones in this article. We like digging into the numbers, finding what’s really going on, and sharing it, because the best advice comes from what’s actually happening in the accounts, not from what the platform says should happen.
About the data
We pulled Google Ads search-term data from three New Path Digital lead-gen accounts: a law firm, a higher-education advertiser, and a local B2B service business. We combined them, so every figure blends all three and is split only by match type. Performance figures come from April 2025 through March 2026.
A keyword qualified only if it had impressions in both periods (same account, campaign, match type, and keyword). An existing query triggered the exact same keyword in both 2022–23 and 2025–26. A new query first appeared in 2025–26 for a keyword that ran in both periods. We counted each keyword-and-search-term pair separately, so one search term can appear more than once.
For the word-overlap table, we compared each query with the closest keyword of the same match type in its campaign and weighted by impressions, so shares describe total query volume, not distinct queries.
We used the closest keyword because Google credits each query to one keyword even when several could match. Using only the credited keyword, 31% of exact-match volume was identical. The 2022–23 shares use the same method.
These are correlations. Search behavior, negatives, budgets, bidding, and Google’s own matching changes could all have shifted between periods, and three accounts make this a field report rather than a benchmark.
Results by account. The phrase result wasn’t uniform. New phrase-match queries cost about 75% more per lead in one account and 29% more in another, but only about 3% more in the third. Exact match was cheaper for new queries in both accounts that ran it (about 7% to 8% lower in each). The phrase gap is concentrated in two of the three accounts, and the smallest account has only about 100 conversions behind its figures.
What “like for like” means. The three accounts have very different costs per lead, and new queries made up a different share of each. Pooling everything lets that mix bend the answer: conversions from new exact-match queries were cheaper in both accounts that ran exact match, yet the pooled number looked more expensive, because the pricier account supplied 24% of new exact conversions but only 6% of existing ones. So for each match type we weight each account’s new-query cost per lead by its share of existing-query conversions. In plain terms: what would new queries have cost if they’d come from the same account mix as existing ones? Click price and conversion rate are adjusted the same way.
FAQs
Q1: How did new exact-match queries perform in this study?
Ans: Across the three lead-gen accounts, conversions from new exact-match queries cost 7% less than those from existing queries.
Q2: How did new phrase-match queries perform?
Ans: Conversions from new phrase-match queries cost 28% more in the studied accounts, mainly because they converted less often.
Q3: Are these findings a Google Ads benchmark?
Ans: No. The figures come from three lead-gen accounts and are correlations, not a universal benchmark.
About Author
