Google wants your keywords on broad match, and it has wanted that since 2020.
Open any Search campaign and the Recommendations tab will offer to switch your exact and phrase keywords to broad, usually with a bigger budget attached. The pitch is that broad match plus Smart Bidding finds conversions your keyword list cannot.
Sometimes that is true. In the independent tests covered below, broad match wins anywhere from a quarter of the time (Optmyzr, across 2,637 accounts) to 85% of the time (Brainlabs, across 54 of its own experiments). That spread is the whole story. Broad match is a bet, and the odds depend on the account.
This article sets out what Google claims, what the outside data shows, the conditions that decide the result, and a four-week test you can run in your own account before accepting the recommendation.
What Google claims broad match and Smart Bidding do together
Google's position is that broad match gives Smart Bidding more auctions to learn from, and Smart Bidding stops broad match from overspending on bad ones. The two are sold as a pair.
The keyword matching help page says broad match may take into account "the user's recent search activities", "the content of the landing pages and assets" and "other keywords in an ad group to better understand keyword intent". None of those are available to exact or phrase match. The same page says it is "critical" to use Smart Bidding with broad match, because each query carries different contextual signals and the bid should reflect them.
On the bidding side, the About Smart Bidding page lists what the system reads at auction time: device, physical location, location intent, weekday and time of day, remarketing list, browser, operating system, interface language and the actual search query. Broad match hands it a wider set of queries to apply those signals to.
The uplift figure Google publishes is on its Grow your Smart Bidding campaigns with broad match page: about 25% more conversions in Target CPA campaigns and about 12% more conversion value in Target ROAS campaigns, while meeting targets. The footnote reads "Google internal data, September 2020". The number has not been refreshed in six years, and it is an average across advertisers Google chose to include.
What the independent data says
Independent results are mixed, and the mix is informative. The agencies that ran controlled experiments report broad match winning often. The platforms that pulled data across thousands of accounts report exact match winning more often.
Brainlabs, a London agency, published the strongest pro-broad result. Its review of 54 tests across 29 advertisers and 12 verticals found a median uplift of 19%, and broad match was the winning variant in 46 of the 54 tests, an 85% success rate. Google's claim at the time was 30%.
"If there has ever been a silver bullet in paid advertising, broad match is it." Andy Goodwin, Brainlabs, writing in PPC Hero
The catch is who ran those tests. Brainlabs is a large agency with mature accounts, first-party conversion data and people watching the search terms daily. The same article says broad match needs real-time bidding and conversion goals that distinguish valuable customers, or "the algorithm may drive low-quality conversions".
Optmyzr looked at it the other way, across every account using its software. Its 2023 study compared broad and exact inside the same 2,637 accounts, all on Smart Bidding. Exact match had the better CPA in 70.79% of accounts, the better ROAS in 72.52% and the better conversion rate in 56.73%. Broad match won on CPA in 27.48%.
xychart-beta
title "Share of 2,637 accounts where exact match beat broad match (Optmyzr, 2023)"
x-axis ["CPA", "ROAS", "Conversion rate"]
y-axis "Percentage of accounts" 0 --> 100
bar [70.79, 72.52, 56.73]
Accounts spending under $10,000 a month did slightly better with broad than larger ones, with broad winning CPA in 30.04% of them. That is still fewer than one in three. Navah Hopkins, who wrote the study, put the advice plainly: "Only test Broad if you go in with protections in place and have budgeted for data acquisition."
Optmyzr went back to the question in May 2026 with 30,000 accounts. Frederick Vallaeys's summary was that phrase match is "the most consistent overperformer", exact leads on efficiency, and broad "is a bet on the algorithm" that works when Smart Bidding has strong conversion value or volume signals. In lead generation accounts, broad "loses efficiency more noticeably" than in ecommerce.
Adalysis reached a similar place with 16,825 campaigns. For lead gen on Maximise conversions, exact dominated and broad brought "cheap clicks but questionable lead quality". For lead gen on Target CPA, broad lagged unless the advertiser needed volume. The study's closing advice began with "Start with exact match."
The smallest and most recent dataset is the most useful for a UK SME, because the accounts look like yours. Ryan Bowman at Web Eminence ran 50/50 experiments in six accounts in April 2026: a photographer, an emergency clinic, a skip-hire firm, a floor cleaner, a bridal shop and a fertility clinic. Broad match cut cost per conversion by 29% and 23% in two of them, raised it by 21% in one, cut ROAS by 60% in the bridal shop, and in the fertility clinic tripled conversion value while nearly tripling cost per lead. He applied it in three of six.
So the honest reading is this. Broad match can beat exact match in a well-run account with enough data. Across the average account, it usually does not.
The four conditions that decide it
The outside tests and Google's own guidance point at the same four conditions. Miss one and broad match tends to spend before it learns.
- Conversion volume. Google's Smart Bidding page says to judge performance over periods with at least 30 conversions, and 50 for Target ROAS. Search Engine Land's Target CPA guide puts the floor at 30 conversions in 30 days. Michelle Morgan at WordStream will go as low as 15 a month, and only with everything else in place.
- A clean conversion signal. If the campaign counts form spam, job applicants and tyre-kickers as conversions, Smart Bidding will find more of them, because broad match gives it the queries to do so. The Brainlabs piece names first-party conversion data as the fix.
- Negatives before launch. Broad match does not stop matching to "jobs", "free", "course" or a competitor's name because you would prefer it did. A shared negative keywords list has to exist before the first click, with a review cadence after.
- Budget that is not capped. Smart Bidding on a limited budget bids for the cheapest conversions it can see. Morgan also notes Google can charge up to twice your daily budget on a day it expects conversions, so a £15 a day campaign on broad match can swing hard.
Where broad match burns money
Broad match burns money in three recognisable situations, and all three are common among UK small businesses.
The first is the campaign producing fewer than 30 conversions a month. There is not enough signal for Smart Bidding to tell a good query from a bad one, so the system spreads spend across whatever broad match brings in. The Optmyzr and Adalysis numbers above are averages that include plenty of accounts like this.
The second is lead generation with a junk conversion. A solicitor counting every contact form submission, including the ones asking for a job, is teaching the algorithm that job seekers are customers. Broad match then matches "solicitor jobs northampton" with a clear conscience. This is the "questionable lead quality" result in the Adalysis data.
The third is the tiny budget. At £10 to £20 a day on a competitive term, the campaign cannot buy enough clicks to exit the learning period, and broad match widens the funnel the budget cannot fill. Google's recommendation usually comes bundled with a higher budget for exactly this reason. The upsell is the budget, and broad match is the reason given for it, which is one more thing to weigh when choosing a digital marketing agency that profits from your spend.
A decision table
Use the table before you accept the recommendation. "Yes" on all four rows means test it. One "no" means fix that first.
| Condition | Test broad match | Hold off |
|---|---|---|
| Conversions in the last 30 days | 30 or more (50 for Target ROAS) | Under 30 |
| Conversion action being optimised | Qualified lead or sale, with spam and job enquiries excluded | Every form submit, every call, or a page view |
| Negative keyword list | Shared list built from the last 90 days of search terms, applied before launch | None, or built after the fact |
| Daily budget vs current spend | Campaign is not "Limited by budget" and can absorb a 2x day | Campaign is capped most days |
| Bidding | Target CPA, Maximise conversions, Target ROAS or Maximise conversion value | Manual CPC or Maximise clicks |
| Who is watching | Someone reviews search terms at least weekly for the first month | Nobody, or monthly |
Manual CPC is on the hold-off side because Google's own broad match page "strongly recommends" the switch only with Smart Bidding. Broad match on manual bids is the old, unfenced version that gave it its reputation.
A four-week test design
Run the test as an experiment with a control. Flipping the keywords and comparing this month to last month tells you about seasonality, not match type. Google now makes the experiment easy. Its single-campaign broad match experiment creates broad versions of your exact and phrase keywords, splits traffic 50/50 and keeps every campaign-level change applied to both arms. It needs Smart Bidding and does not support portfolio strategies.
Google suggests six to eight weeks. Four weeks is the shortest run that will tell you anything, and it only works if each arm clears about 30 conversions in that time. If it does not, extend to eight. Calling it early on 12 conversions a side is a coin toss with a spreadsheet attached.
The steps:
- Pull the last 90 days of search terms and build the negative list first. Add it to both arms, because the test is about matching, not about whether you have negatives.
- Set the experiment at 50/50 from the Experiments page, or from the Recommendations tab using "Apply as experiment". Do not apply the recommendation directly.
- Freeze everything else for four weeks: no bid target changes, no new ads, no budget edits. Any change hits both arms, but it also restarts learning.
- Review the search terms report in the broad arm twice a week for the first two weeks, then weekly. Add negatives to both arms as you go.
- Score at the end on cost per qualified lead or ROAS, not on conversions. If you import offline conversions, wait for the lag before scoring.
- Apply broad match only if the trial arm wins on the efficiency metric with the extra volume, or wins on volume at a cost you have agreed in advance. "Slightly better" with an inconclusive label is not a win.
The six-account Web Eminence test is a useful model. Three accounts passed that bar, one was paused after a 21% rise in cost per conversion despite more leads, and two were left alone. That is what a sensible outcome looks like.
The principle
Broad match is a tool for buying more of what the algorithm already understands. If your conversion data is thin or dirty, it has nothing to understand, and the extra reach is spend you cannot read. Earn the right to use it with 30 clean conversions a month and a negative list, then test it against a control. Last month is not a control. A UK Google Ads agency that proposes broad match without those protections is selling you the upsell, not the result.
If you would like a second pair of eyes on whether your account is ready, our paid media team can run the pre-test checks and the four-week experiment, or get in touch with the campaign in question.