Observation vs Targeting Audiences: The Setting People Get Backwards
The difference is who can see your ad, not who gets a bid adjustment, and getting it wrong either tanks your reach or wastes your budget.
The short version
- Observation changes bids and reporting; Targeting changes eligibility. They are not two flavors of the same control.
- Default Search audiences to Observation. The 'better' CPA under Targeting is usually survivorship from a smaller, biased pool.
- Targeting on Search excludes every keyword-matched user Google cannot confirm is in the audience, which is often most of them.
- To test Targeting, duplicate the campaign and run it in parallel instead of flipping in place.
- On Display and Video, the rule inverts: Targeting is often the correct default because there is no query intent to fall back on.
Observation adds an audience to a campaign for reporting and optional bid adjustments without changing who sees your ads. Targeting restricts the campaign so only people in that audience can see your ads. The setting people get backwards is flipping to Targeting on a Search campaign that was running fine, which collapses eligible impressions to a sliver of the audience Google has actually tagged. For most Search campaigns you want Observation. Reserve Targeting for when narrow reach is the entire point.
What each setting actually does to delivery
Observation is a lens, not a gate. When you add an audience in Observation mode, every query that matches your keywords is still eligible exactly as before. The audience membership gets layered on top as a reporting dimension and, if you set one, a bid adjustment. Someone outside the audience still sees your ad. Someone inside it sees the same ad, just at a bid that may be nudged up or down.
Targeting is a gate. It adds a hard condition: the user must be in the audience (or audiences) you attached, in addition to matching your keywords. If Google cannot confirm the user is in that audience at auction time, your ad is simply not eligible. There is no fallback to the broader keyword-matched pool.
The reason this trips people up is that the two settings live in the same UI toggle and look like variations on a theme. They are not. One changes bids; the other changes eligibility. A bid adjustment of -100 percent in Observation still is not the same as Targeting, because Observation never removes you from auctions you would otherwise enter outside that audience.
Why Targeting quietly destroys Search reach
Audience membership is not a clean, fully-populated flag. Google can only place a user in an in-market or affinity segment when it has enough signal: browsing history, logged-in behavior, YouTube activity, and so on. A large share of the people typing your exact keyword have no confirmed membership in the audience you attached. In Observation, they still match on the keyword and still see your ad. In Targeting, they are excluded.
So the failure mode is this: you switch a healthy Search campaign to Targeting because the audience 'converts better,' and impressions fall off a cliff overnight. The campaign did not get more efficient. It got blind to every keyword-matched searcher Google could not confirm belonged to the segment. On Display and Video this matters less, because those networks are audience-first to begin with. On Search, where intent lives in the query, gating on audience throws away your strongest signal.
The trap is that the remaining traffic often does look better on a CPA report, which feels like confirmation that Targeting was right. It is survivorship. You kept the subset Google could identify, and that subset skews toward heavy, logged-in users. You are not reading a better campaign; you are reading a smaller, biased one.
The decision rule: default to Observation
For Search campaigns, start every audience in Observation. Let it accumulate data for at least a few weeks or until each audience has enough clicks to read, then act on bid adjustments rather than eligibility. Observation gives you the performance breakdown without the reach penalty, so you keep optionality.
Use a positive bid adjustment when an audience's conversion rate or value runs clearly ahead of the campaign average across a meaningful sample. A practical starting point is a +15 to +20 percent adjustment on an audience that is beating the account consistently, then watch for two weeks before pushing further. Keep adjustments modest early: a large bump on a thin sample just overpays for noise.
Only move to Targeting when restricted reach is the actual goal, not a performance hope. Legitimate cases: a remarketing-only campaign where you deliberately want to speak to past visitors, a competitor-conquesting campaign aimed at a specific in-market segment, or a tightly budgeted push where you cannot afford to serve outside a known group. In every one of those, you are accepting the reach loss on purpose.
How to flip it without blowing up the account
The safest path from Observation to Targeting is to not do it in place. Duplicate the campaign (or ad group), set the copy to Targeting with the audience, and run it alongside the Observation version with its own budget. Now you can compare like for like: same keywords, different eligibility rule. If the Targeting version delivers the efficiency you hoped for without starving on volume, shift budget toward it. If it just goes quiet, you have lost nothing.
You can check which mode an audience is in under the campaign or ad group's Audiences section: the setting shows as 'Observation' or 'Targeting' at the ad group level, and Smart Bidding ignores your manual bid adjustments once it takes over, which is another reason to lean on Observation plus automated bidding rather than hard Targeting. With a value-based or conversion-based Smart Bidding strategy, the algorithm already uses audience signals to bid more aggressively on high-value users, so Targeting often adds a reach constraint for a job the bidder is already doing.
If you do find a campaign stuck in Targeting that should not be, do not assume the historical conversion data proves the audience's worth. Switch it back to Observation, let it rebuild a full-funnel picture, and only then decide on bid adjustments from the broader sample.
Honest limits and the one case where the obvious call is wrong
You cannot see, per impression, why a given user did or did not qualify for an audience. Google does not expose the match rate of its segments, so you cannot measure exactly how much reach Targeting is costing you except by observing the before-and-after impression drop. Treat that drop as your evidence.
The counterexample: on Display and Video campaigns, defaulting to Observation can be the mistake. Those networks have no query intent to fall back on, so a broad Display campaign in Observation mode can serve to almost anyone your placements and topics allow. There, Targeting is frequently the correct default, because the audience is the only intent signal you have. The Observation-first rule is a Search rule.
Finally, remember that bid adjustments in Observation are relative, not absolute controls. A +20 percent on an audience and a -10 percent on a device stack multiplicatively, and Smart Bidding overrides them entirely. If you are on a fully automated strategy, your Observation audiences are feeding the model, not driving the bids yourself. That is fine. It is still the right place for them.
Questions people also ask
Does a -100 percent bid adjustment in Observation do the same thing as Targeting?
No. A -100 percent adjustment stops you bidding on users inside that audience, but you still remain eligible for all the keyword-matched traffic outside it. Targeting removes eligibility for everyone not in the audience. The two settings touch different parts of the auction.
Why did my impressions crash after switching to Targeting?
Because Targeting makes audience membership a requirement, and Google can only confirm membership for users it has enough signal on. Every keyword-matched searcher it cannot place in the segment is now excluded. The traffic that remains often looks efficient, but that is a smaller, skewed sample, not a better campaign.
Should I use Targeting for a remarketing campaign?
Yes, that is a legitimate Targeting use. If the entire point is to reach only past visitors, you want the gate. The reach restriction is the goal, not a side effect, so the usual Observation-first rule does not apply.
How big should my audience bid adjustment be?
Start modest. A +15 to +20 percent bump on an audience that consistently beats your account average is a reasonable opening move, then watch two weeks before increasing it. Large adjustments on thin data just overpay for noise. If you run Smart Bidding, your manual adjustments are ignored anyway and the audience serves as a signal instead.
Does the Observation-first rule apply to Display and Video too?
No. On Display and Video there is no query intent to fall back on, so a broad campaign in Observation can serve to almost anyone your placements allow. On those networks Targeting is frequently the correct default. The Observation-first advice is specific to Search.