Smart Bidding Exploration is a setting on target ROAS campaigns that lets Google bid on queries your ROAS target would normally exclude. You hand it a ROAS target tolerance, which is a percentage your return is permitted to drop by, and the bidder uses that slack to buy traffic it was previously priced out of. It does not change your keywords, your audiences, or your budget. It changes how far the bidder is allowed to miss your target while hunting for converting queries you've never seen. Google reports meaningfully more unique converting query categories from it. Whether that's good news for your account depends almost entirely on whether your conversion data is honest.
What the setting actually changes
A target ROAS strategy is a filter. You tell Google you want a 400% return, and the bidder declines auctions where its forecast says it can't hit that number. That's the point of it, and it works. The side effect is that everything the model is uncertain about gets filtered out too, because uncertainty reads as risk against the target. New query phrasings, new intent clusters, seasonal language nobody typed last quarter: all of it sits outside the filter.
Smart Bidding Exploration widens the filter without moving it. The target stays where you set it. The tolerance tells the system how much it can undershoot on a given slice of traffic in exchange for information. Google's framing, as covered when the feature rolled out broadly, is that you're buying query discovery with a bit of ROAS, and that's an honest description of the trade.
The numbers Google publishes are about discovery, not efficiency. Their reported results centre on roughly 18% more unique converting search query categories and around 19% more conversions, with Search campaigns seeing about 27% more unique converting users. Read those carefully. They're volume and diversity metrics. Google is not claiming your ROAS goes up, and you shouldn't expect it to.
The confusion that sends people to the wrong fix
Search "bidding exploration" and half the results describe something completely different: the recalibration period a campaign enters after you change a bid strategy, a target, a budget, or a conversion goal. That phase is automatic, it lasts roughly one to two weeks, performance wobbles, and the advice is to leave it alone.
Those are two different things with nearly identical names, and we've watched account managers conflate them into a real mistake. Someone sees volatility, decides "exploration" is to blame, hunts down the Smart Bidding Exploration checkbox, and unchecks it. That's a bid strategy change. It restarts the very recalibration they were trying to escape. Two weeks later they've got a second round of volatility and a conclusion built on nothing.
One is a phase you wait out. The other is a policy you opted into. If your campaign got weird within days of a change you made, you're in the learning period and the setting is irrelevant. If it's been stable for a month and you're deciding how aggressively to buy new queries, that's when the setting is the actual question.
Where to find it, and how to turn it off
The setting lives inside the campaign, not at the account level. Open the campaign, go to Settings, expand Budget and bidding optimization, find Smart Bidding Exploration, and you'll see the option to allow Smart Bidding to explore new traffic on this campaign. Uncheck it to opt out.
Two things to know before you do. First, it only appears on campaigns running target ROAS, since there's no target to loosen on Maximize Conversions or target CPA. If the setting isn't showing up, check the bid strategy before you file a bug. Second, unchecking it is a change like any other, with its own recalibration window attached. Flip it, then wait.
The prerequisite nobody puts in the release notes
Here's the part that decides the outcome, and it has nothing to do with Google Ads.
Exploration is an amplifier pointed at your conversion feed. The bidder expands into unfamiliar queries and the only thing telling it which expansions worked is the conversion data you send back. If that data is clean, you get genuine discovery. If it isn't, you get faster, more confident, more expensive wrongness.
We audit the same three failures on nearly every account that comes to us with bidding problems:
Soft conversions counted as hard ones. Form starts, newsletter signups, and page views on a pricing page all firing as primary conversions. The bidder optimises toward whatever's cheapest to produce, and soft events are always cheapest. Turn on exploration over that and you'll discover an impressive volume of people who start forms.
Duplicate firing. A tag on the thank-you page plus a server-side event plus a container-level trigger, all reporting the same purchase. Now the model believes certain query types convert at double the real rate, and exploration will chase them hard. If your platform numbers and your analytics disagree, start with our breakdown of why Google Ads and GA4 report different conversion counts before you change a single bid setting.
Unqualified lead volume with no value attached. Every lead reported as the same conversion value, so a tire-kicker and a signed deal look identical to the bidder. This is the one that quietly ruins lead gen accounts. If you can send value back, send it. That's the entire premise behind offline conversion imports, Google's Data Manager migration, and server-side setups like Meta's Conversions API.
Clean your signal first. It's not a nice-to-have prerequisite, it's the whole experiment. Exploration with good data is query discovery. Exploration with bad data is an accelerant.
Why this setting matters more in late 2026
Context helps here. Through September 2026 Google is auto-upgrading campaigns that use automatically created assets or the campaign-level broad match setting into AI Max for Search, and per Google's own announcement the Dynamic Search Ads sunset has been pushed out to February 2027. So a lot of accounts are simultaneously getting a broader matching surface whether they asked for it or not.
That matters because AI Max and Smart Bidding Exploration widen different valves on the same pipe. AI Max broadens what you're eligible to match on. Exploration broadens what you're willing to pay for. Turn both on in the same week and you will not be able to say which one moved your numbers. We stagger them, deliberately, and we wrote up our full position on the matching side in the AI Max playbook.
The skepticism is also worth taking seriously. Independent testing by Monks published in August 2025 found that in the accounts they examined, 99% of AI Max impressions produced zero conversions across roughly 30,000 search terms. Google's own published figure for the full AI Max feature suite is about 7% more conversions or conversion value at a similar CPA or ROAS versus search term matching alone. Both of those can be true at once, and the gap between them is mostly a story about account quality and measurement. Broad automation rewards accounts that already feed it good data and punishes ones that don't.
How we actually read the result
Most people test this wrong by looking at the wrong number. They enable exploration, watch ROAS, see it dip three percent, and switch it off. But a small ROAS dip is the price on the label. It's supposed to happen. You agreed to it when you set a tolerance.
The metric that tells you whether it worked is the count of distinct converting search terms. That's what the feature is for. Pull your converting query list before you enable it, pull it again after a full window, and compare the two sets, not just the totals. If you're converting on phrasings and intent clusters that weren't in the first list, exploration bought you something durable: those queries are now known-good traffic your bidder can pursue at full target forever after. A three percent ROAS dip for a permanently wider converting query set is a good trade. A three percent dip with an identical query set means you just paid a premium for the same traffic, and you should turn it off.
We hold everything else still during that window. No target changes, no budget moves, no creative refreshes, one campaign at a time with a comparable campaign left alone as a control. That's slower than most people want. It's also the only version that produces an answer instead of a vibe. The same discipline applies to reading automated campaign types generally, which is why we build separate channel-level Performance Max reporting rather than trusting a blended number.
So should you leave it on?
Our default on client accounts: yes on Search target ROAS campaigns with clean conversion data and enough volume to read, no on anything else.
Leave it on when the conversion feed is genuinely trustworthy, the campaign has enough conversion volume that a two-week window means something, and you have room to absorb a modest efficiency hit in exchange for query discovery. Under those conditions it's one of the better automation defaults Google has shipped in a while, because it's honest about the trade instead of hiding it.
Turn it off when you're running thin volume, when you're in a period where hitting the ROAS number matters more than growing, when you haven't verified your conversion actions in the last six months, or when you've got another major change in flight. And turn it off if you can't say out loud what your conversion values represent. That's not a bidding problem, and no setting is going to fix it.
Most of the accounts we take over don't have a bidding problem at all. They have a measurement problem wearing a bidding problem's clothes. Once the data underneath is right, decisions like this one get easy, because you can finally tell what happened. That's the same reason we're opinionated about attribution infrastructure before anyone touches a bid strategy.
If you're not sure your conversion data is good enough to explore on, that's the thing to find out first. Get a free automation audit and we'll tell you what your account is actually reporting, and whether it's safe to hand Google more rope.