Hypothesis
If the trial arm runs the same keywords on broad match with Smart Bidding, it will find additional converting searches within the target cost per conversion, because broad match reaches related queries the exact and phrase keywords miss. The decision behind it is concrete: move a mature exact- and phrase-match campaign to broad match under Smart Bidding, or keep the tighter match types.
The hypothesis carries its own condition. Google's documentation on keyword matching options describes broad match as reaching searches related to the keyword, and its Smart Bidding documentation describes bids set for each auction from signals available at that moment. Put together, broad match should find more searches and Smart Bidding should pay only for the promising ones. Whether that happens in a given account depends on its conversion data, its query space and its tolerance for waste, which is why it is a test and not a setting. The keyword match types guide explains how Keyword Match TypesSettings that decide how closely a search must match a keyword for the ad to enter the auction. and bidding interact; this page assumes that background and designs the comparison.
Metrics and guardrails
The primary metric is conversion value at the target cost per conversion, or conversions where the account does not track value. Read it at the target, not in absolute terms: an arm that reaches more searches will usually spend more and convert more, and the question is whether the extra conversions come at a price the account accepts. Use the same conversion actions in both arms, the ones the bid strategy optimises for, so each arm is judged on the outcome it was given.
Three guardrails protect the account. Cost per conversion shows whether the trial arm holds the target or drifts above it; the CPA calculator puts both arms on the same footing. The share of spend on irrelevant search terms, read from each arm's Search Terms ReportThe Google Ads report of the actual searches that triggered an account's ads., shows how much budget broad match spends on queries the business would never want; agree before launch what counts as irrelevant and how much of it is tolerable. The third is brand-term overlap: broad keywords can match brand searches and take them from the brand campaign, flattering the broad arm with conversions that were already won. Check the brand campaign's traffic by arm, and if the overlap appears, the brand exclusions and Negative KeywordsKeywords that stop an ad from showing for searches containing them. covered in the negative keywords guide should be in place in both arms before the read.
Design
Use the platform's campaign experiment. Duplicate the campaign as a trial arm, change every keyword's match type to broad in that arm and nothing else, and split traffic between the original and the trial. The split is search-based or cookie-based, as the experiment setup offers. Record which was chosen and why, because in accounts where people search several times before converting, how repeat searches are assigned affects how cleanly the arms are kept apart.
| Design element | Choice | Reason |
|---|---|---|
| Unit of randomisation | Search traffic, split by the campaign experiment | Both arms face the same auctions on the same days |
| Arms | Original exact and phrase versus the same keywords on broad | One change: match type |
| Bid strategy | The same Smart Bidding strategy and target in both | Isolates match type from bidding |
| Negatives, ads, budgets | Identical, with any mid-test negative added to both | Prevents management effort becoming the variable |
| Primary metric | Conversion value at the target cost per conversion | Rewards efficient reach, not extra spend |
| Learning period | Run but excluded from the read | Early delivery after the change is not representative |
The bid strategy stays fixed throughout. Changing the target or the strategy mid-test restarts learning and mixes two questions; the Smart Bidding strategies guide explains which edits do that. Where the question is the strategy rather than the match type, the bidding strategy geo split test is the matching design. The academy exercise on testing what platform automation adds for a model railway retailer practises writing this plan with the target held in both arms and late conversions waited for, then placing it beside a Meta audience test and an optimisation event test in a calendar that keeps each result separate.
Sample size reasoning
The arms are read on conversions, which are far rarer than clicks, so the test needs enough conversions per arm to separate the smallest gain worth the extra query risk from week-to-week noise. Decide that smallest gain before launch: broad match brings management effort and some irrelevant spend, so a gain too small to pay for that is not worth detecting. The sample size and significance guide shows how the baseline conversion volume and that minimum effect set the run length, and why value is noisier than a count.
If the campaign produces too few conversions to separate the arms in a tolerable period, the honest answer is that the account is not ready for this decision. Let conversions accumulate on the current structure, or run the test on a larger campaign whose result would guide the smaller ones.
Duration
Set aside the bid strategy's learning period after launch, then run at least two full weekly cycles of stable delivery. Extend the read for accounts with a long conversion delayThe time between an ad interaction and the conversion it leads to, which leaves the most recent reporting days incomplete., where many conversions arrive days or weeks after the click; reading before they land counts the trial arm's newest clicks as failures. Fix the end date before launch and avoid sales, seasonal peaks or site changes that would hit the arms on different days.
Reading the results and the decision rule
Read the test once, at the planned end and after the conversion delay has passed. Compare conversion value at target between arms with the interval, not just the point estimate, then check each guardrail. Write the decision rule into the plan before launch:
| Result | Decision |
|---|---|
| Trial arm adds conversion value at or below the target cost, irrelevant spend inside the agreed limit | Apply broad match to the campaign; keep the search terms review routine |
| Trial arm adds value but irrelevant spend breaks the limit | Keep exact and phrase; add the trial's converting queries as keywords and its waste as negatives |
| Trial arm adds conversions only above the target cost | Keep exact and phrase; retest when conversion volume is higher |
| No reliable difference | Keep the current structure, which is cheaper to control |
| Gain comes mostly from brand queries | Fix brand exclusions in both arms and rerun; the result as read is overlap, not reach |
Whatever the outcome, the trial arm's search terms are a by-product worth keeping. The irrelevant search terms troubleshooting guide shows how to turn them into negatives without cutting the converting queries the test found.
A plumbing supplier tests broad match on its core campaign
Dunmarrow Plumbing Supplies, an invented trade and retail plumbing merchant, runs its largest search campaign on exact and phrase match with a target cost per conversion of £22. It creates a campaign experiment with a cookie-based even split and moves the trial arm to broad match. After a two-week learning period set aside, the read runs for four weeks. The original arm records 410 conversions at £21.40; the trial arm records 468 at £21.90, a difference in conversions of about 14% with an interval that stays above zero. Irrelevant terms take 6% of trial spend against a limit of 10%, and brand overlap is negligible. Dunmarrow applies broad match and schedules a weekly search terms review.
Pitfalls
Reading the learning period as a result is the commonest error: the first days after a match-type change often look worse, and a team that judges them abandons a structure that might have worked. Brand queries leaking into the broad arm are the second, and they flatter it with conversions the brand campaign would have won. The third is conversion delay. An arm that buys new, less familiar searches may attract people who take longer to convert, so an early read understates it; wait for the delay to pass.
Two smaller traps sit in the setup. Changing ads or landing pages in one arm, even to fix a typo, adds a second variable. And the platform may show recommendations to apply the trial early; apply nothing until the planned read.
Common mistakes
Match-type tests go wrong on their setup more often than on their arithmetic. The search ad asset test changes ad text with targeting fixed and is the natural partner to this one, and the experiment library lists every design.
- Changing the bid strategy with the match type. Two changes give an answer to neither question.
- Adding negatives to one arm only. Management effort becomes the variable.
- Judging conversions without the target. More spend buys more conversions; read value at target.
- Reading inside the learning period. Calibration costs are temporary.
- Ignoring brand overlap. The broad arm can win on conversions the brand campaign already earned.
- Throwing the trial away after a loss. Its search terms are new keywords and new negatives.
