The bidding setting that buys queries you would never pick

Smart Bidding Exploration is a tolerance band you set underneath your target return, and its whole purpose is to buy traffic your target would otherwise refuse. Google reports 18% more unique search query categories producing conversions and 19% more conversions in its own Search testing, later restated as 27% more unique converting users. It became available as an opt-in through Google Ads Editor 2.11 from 6 November 2025, and expanded in May 2026 to Performance Max, PMax with product feeds, and Shopping.

Most accounts have not touched it, because on the surface it reads as accepting a worse return, and nobody wants to walk into a board meeting having chosen that.

That framing is wrong, and the reason is worth understanding.

Why a strict target eventually stalls

Target ROAS bidding is greedy. It bids toward whatever the model already believes converts, which means it re-buys the same query neighbourhoods, the same audiences and the same moments, day after day. Each purchase reinforces the belief. The model gets more confident about a shrinking slice of the market.

That is efficient and it is also a trap. The queries you have never bought have no conversion data, so the model predicts them poorly, so it bids low, so it never buys them, so they never get data. Nothing in a strict target setup breaks that loop.

Exploration breaks it deliberately. The tolerance is a licence to place bids the model is unsure about, priced in efficiency rather than in dollars. You are not buying worse traffic. You are buying information about traffic, and paying for it in blended return.

How to set it without wrecking the quarter

Four rules.

Set the tolerance where you can afford it. A tolerance of a few percent under target does very little. A wide one moves real money. Start narrow on one campaign and widen it once you have seen the shape of the spend, rather than starting wide and reacting.

Do not stack it with a budget squeeze. Exploration needs headroom. If the campaign is already budget-constrained, the exploratory bids compete with your proven ones for the same limited spend, and you get the cost of exploration without the volume.

Judge it over a full purchase cycle, not a week. If your average time from click to sale is 21 days, a fortnightly read is measuring the cost and none of the return. This is where most tests get killed prematurely.

Watch the query category count, not just the return. The metric Google itself reports on is unique converting query categories. If that number is not moving, exploration is not finding anything and you should turn it off. If it is moving and blended return has softened slightly, the trade is working as designed.

Where it earns its keep, and where it does not

It earns its keep in accounts with a broad addressable market, a healthy conversion volume, and a product where demand shows up in a hundred different phrasings. Manufacturers, suppliers, multi-service businesses, anything with a long tail of use cases the marketing team has never thought to write a keyword for.

It does not earn its keep in three situations, and it is worth being blunt about them.

Thin conversion volume. If the campaign generates a handful of conversions a week, the model cannot distinguish an exploratory win from random variance, and neither can you. Exploration in a low-volume account is just noise you paid for.

Sharp seasonal spikes. Exploring during your peak is expensive, because every exploratory impression displaces a proven one at the moment proven ones are worth the most. Explore in the shoulder, harvest in the peak.

Lead generation with weak lead quality feedback. This is the dangerous one. If your conversion action is a raw form fill, exploration will find you the cheapest possible form fills, and they will be worse than your current ones by exactly the margin the tolerance allowed. The model is doing its job. You gave it the wrong target.

The fix for that last case is not to avoid exploration. It is to feed qualified leads back into Google Ads via offline conversion import, so the thing being explored toward is a real business outcome and not a page view on a thank-you screen.

The PMax and Shopping expansion

The May 2026 expansion matters more than the original Search release for retail and ecommerce accounts, because PMax and Shopping are exactly where the greedy-bidding trap bites hardest. A product feed with 4,000 SKUs typically has a few hundred doing all the work, and the rest sitting in a permanent cold start.

If you run a large feed, this is the first mechanism Google has offered that specifically funds discovery across the tail rather than concentrating on the head. Set it against a control campaign, split by product type or custom label so the comparison is clean, and give it a full season.

Reading the results honestly

Google's 18%, 19% and 27% figures come from Google, drawn from aggregate internal testing over a one-month window in early 2025. They are a reason to test, not a forecast for your account.

The read that will actually convince a finance team is simpler than an attribution debate: total conversions and total contribution over a full quarter, exploration on versus a matched control, with product mix and seasonality noted. If contribution is up and blended return is a little down, you have bought growth at a known price. That is a decision anyone can defend.

Set the tolerance where you can explain it out loud, and the rest of the argument takes care of itself.