Optimized Targeting in Google Ads 2026: Controlling Audience Expansion
You uploaded a customer list, built a custom intent segment, carefully excluded the audiences you did not want — and then the audience report shows rows you never added, plus 40% of your spend sitting inside them. That is optimized targeting at work. It is on by default in Demand Gen, Performance Max and most Display campaigns, and in 2026 the job is not to toggle it off in irritation but to understand what signal you are feeding it.
By the end of this article you will know how optimized targeting differs from an audience signal, exactly where it can and cannot be switched off, how to see its contribution in reporting, how to build a signal that steers expansion toward profitable users, and the specific cases where turning it off is the honest answer.
Optimized targeting versus audience signals
Three things look similar in the interface and behave very differently.
- Targeting is a hard constraint. Show only to these people. The system stays inside the list.
- Observation adds an audience without narrowing reach: you get a reporting slice and the option to adjust bids. The full mechanic is covered in our piece on audience observation in Google Ads.
- Optimized targeting is the opposite of a constraint. Your audiences become a starting point, and the system deliberately goes beyond them when it finds users whose behaviour suggests they are more likely to convert.
The important difference from the retired similar audiences: those were built once and lived as a static list. Optimized targeting re-evaluates similarity continuously using page signals, the user’s recent search behaviour and the conversion data your account reports. The practical consequence is blunt — the quality of the expansion depends on the quality of your conversion tracking, not on how elegant your audience lists look.
Optimized targeting is not another audience. It is a system that learns from your conversions. Feed it a junk conversion and it will scale junk with impressive efficiency.
Where it is on, and where you cannot turn it off
As of 2026 the picture looks like this. Always verify in your own account, since Google keeps shifting which controls appear where.
| Campaign type | Status | What you can do |
|---|---|---|
| Display | On by default | Uncheck at ad group level |
| Demand Gen | On by default | Can be disabled per ad group; audiences then act as a signal |
| Video with a conversion goal | On by default | Can be disabled, usually at the cost of volume |
| Performance Max | Built in, no separate toggle | Influenced only through audience signals and exclusions |
| Search | Not applicable in this form | Expansion happens through match types and AI Max |
Performance Max deserves a separate note: the audience signal there is not targeting at all — it is a hint about where the system should begin looking. If you still treat that signal as a boundary, revisit our breakdown of Demand Gen and how audience-first campaigns actually behave; half of the complaints about automated campaigns start with this misunderstanding.
Finding the control in 30 seconds
- Open the campaign, then Ad groups, then the ad group in question.
- Go to Audiences, or Ad group settings depending on your interface version.
- Look for the Optimized targeting or Audience expansion block.
- Unchecked means the system stays inside your lists. Checked means your lists are a launch pad.
Seeing its contribution in reporting
The most common complaint about optimized targeting is “I cannot tell where the money went”. You can, but the report is not where most people look.
- Ad group level audience report. Rows you never added are attributed to optimized targeting. Segment by targeting type to split “your audiences” from “expansion”.
- Audience insights. Shows which segments the system found on its own. That is effectively a shortlist of audiences worth adding manually if they convert.
- Placement report. On Display and video, expansion frequently drifts into apps and low-value placements. Review it alongside brand hygiene — the mechanics are in our guide to brand lists and brand exclusions.
- New versus returning segment. If expansion mostly brings back people who already visited, you are paying new-reach prices for remarketing.
A rough interpretation benchmark, to sanity-check against your own data rather than adopt as a rule: if expansion consumes more than 50–60% of ad group spend while running a CPA more than 1.5× worse than your core audiences, either the signal is weak or the model is learning from a low-value conversion event.
Building a signal that steers expansion the right way
Optimized targeting orbits whatever you gave it. Here are signal sources ranked from strongest to weakest.
- CRM customer lists. The cleanest signal available: real people who really paid. Upload profitable customers, not “all leads”. Data requirements and matching mechanics are covered in our guide to Customer Match and first-party data.
- High-intent site audiences. Not “all visitors” but: viewed a product page for 60+ seconds, added to cart, began checkout, returned twice within seven days.
- Custom segments. Built from search terms and competitor URLs, they describe your buyer in behavioural language instead of demographics. See our walkthrough of custom segments and intent audiences.
- GA4 predictive audiences. Purchase probability and churn probability hand the model information it otherwise lacks: which of your users are close to money. Details in our piece on GA4 predictive audiences in Google Ads.
- Broad affinity and in-market interests. The weakest option. As a steering signal they are close to useless — too wide to point anywhere.
The one-signal rule
A frequent mistake is dumping ten lists into a single signal. The model averages them and ends up with a blurred portrait. It works better to give each ad group one strong signal — or two or three closely related audiences — and split competing hypotheses across ad groups so the report stays readable.
When it earns its keep and when it burns budget
| Situation | Call | Why |
|---|---|---|
| 30–50+ monthly conversions, event = purchase or qualified lead | Leave it on | The model has enough data and a meaningful target |
| Conversion = button click or contact page view | Turn off until the goal is fixed | Expansion will learn to find clickers, not buyers |
| Narrow B2B niche, total market in the hundreds of companies | Turn off | There is nowhere to expand; “similar” means “irrelevant” |
| Abandoned-cart remarketing | Turn off | The entire point is specific people; expansion dilutes it |
| Launching in a new geo with no data | On, with a tight signal and a capped budget | Finds the first pockets of demand faster |
| High-ticket product with a long sales cycle | On only after offline conversion import | Otherwise the model optimizes for the top of the funnel |
Notice the pattern: nearly every “turn it off” row is really about conversion quality, not about the feature. Before arguing with expansion, check what you are counting as a conversion — our guide to enhanced conversions from first-party data is the place to start.
A 40-minute audit
- 0–5 min. List every campaign with optimized targeting enabled. It is usually running in campaigns you had forgotten about.
- 5–12 min. For each, open the audience report and segment by targeting type. Record expansion’s share of spend and its CPA versus core audiences.
- 12–20 min. Open audience insights. Write down 5–10 segments the system discovered that are actually converting.
- 20–28 min. Check the campaign’s conversion goal. Purchase or qualified lead, or something earlier? Earlier means the model is learning from a weak event.
- 28–35 min. Review exclusions: brand, existing customers, irrelevant placements, dead geos.
- 35–40 min. Decide per ad group: keep, keep with a new signal, or disable. Do not change everything at once or you will learn nothing.
Mistakes and what they cost
- Disabling expansion and changing the signal in the same edit. Two changes, zero conclusions. Move one lever at a time.
- Judging after three days. Any edit sends the campaign back through learning. Wait at least one or two full conversion windows.
- Not excluding existing customers. Expansion will happily find people who already bought — they look exactly like your buyers. Upload purchasers as an exclusion list.
- Using “all site visitors” as the signal. That describes anyone who has seen you, not who your customer is.
- Assuming off equals cheaper. Disabling often just reduces volume at the same CPA. That is a scale decision, not an efficiency one.
- Ignoring the placement report. On Display, expansion leaks into mobile games fast, and exclusions there pay off quickly.
Measuring the effect honestly
The audience report shows attribution, not causation: a conversion credited to expansion might have happened anyway. For a real answer you need an experiment.
- Create a 50/50 split experiment in Experiment Center.
- Make optimized targeting on/off the only difference between arms.
- Keep budget and bid strategy identical.
- Let it accumulate at least 100 conversions in total — a working benchmark, not a law — and compare CPA and volume, not CTR.
The full procedure and its pitfalls are in our article on running experiments in Google Ads. The same discipline settles every other automation argument: what gets compared is two identical campaigns, not two opinions.
How this connects to the other two levers
Optimized targeting decides who sees the ad. What they see, and whether it worked, are separate questions: creative quality and measurement. For the first, see our breakdown of Ad Strength and RSA asset performance ratings. For the second, see how Brand Lift and Search Lift measure video that produces no clicks. Audience, message, measurement — one loop.
What to do in a brand-new account
With no history, expansion is close to guessing. A sane sequence:
- First two to three weeks: hard targeting, no expansion, collect the first conversions.
- Set up high-quality conversion events — payments, not form fills — and import offline conversions if the cycle is long.
- Upload Customer Match even if the list is small.
- Enable optimized targeting on one campaign, not all of them.
- Compare against a control campaign after two conversion windows.
If you are building an account from scratch and the question is broader than one checkbox, start with the fundamentals: PPC Rebels runs a Google PPC agency service for high-volume advertisers, and the English blog archive covers the setup steps most accounts skip.
Optimized targeting and Smart Bidding: which one affects which
A fair question: if an algorithm sets the bids, why think about targeting at all? Because these are two separate decision layers that fire in sequence.
- Targeting defines the candidate pool — who is eligible to see the ad at all.
- Smart Bidding decides what to pay for a specific impression inside that pool, scoring conversion probability in the moment.
Two consequences follow. First, audience expansion does not “trick” your bidding. The bid algorithm evaluates a newly reached user exactly as it evaluates anyone else; if that person is unlikely to convert, the bid is low and nothing catastrophic happens. Second, and less comfortable: when the conversion signal is dirty, both layers err in the same direction. Expansion finds people who resemble your bad conversions, and Smart Bidding pays confidently for them because the model considers them valuable.
Which is why the repair order never changes: conversions first, bidding second, audiences third. Rearranging audiences while measurement is broken is rearranging furniture in a burning room.
What happens when you enable expansion on a live campaign
- The campaign re-enters a learning period, typically several days depending on conversion volume.
- CPA is almost always worse in the first days — the system is exploring a new pool.
- Then one of three outcomes: CPA returns to baseline at higher volume (a win), stays worse at the same volume (weak signal), or volume does not grow at all (the market is tapped out).
Judge in week three, not on day three. A decision made on learning-period data is nearly always the wrong one.
A campaign structure that keeps expansion controllable
The main reason optimized targeting feels uncontrollable is organisational rather than technical: everything sits in one campaign, so separating expansion from the core is impossible even in principle.
A structure that works:
| Campaign | Audiences | Expansion | Job to be done |
|---|---|---|---|
| Core / remarketing | Cart, product view, active CRM segments | Off | Maximum efficiency, predictable CPA |
| Similar-to-buyers | Customer Match, purchasers in last 180 days | On | Scale on a high-quality signal |
| Demand discovery | Custom intent segments | On | Find new pockets of audience |
| Exclusion list (shared) | Buyers, job seekers, current clients | — | Stop paying for people you do not want |
Three campaigns instead of one give you exactly what was missing: separate budgets, separate reporting, and the ability to switch off what fails without touching what works. The side benefit is that you stop arguing with the interface and start arguing with numbers.
One caveat: splitting only makes sense if each campaign has enough volume. A working benchmark is 15–30 conversions per campaign per month. Below that, splitting hurts, because every campaign lives permanently in learning.
A worked example
An illustrative case assembled from a typical pattern. The figures are an example, not a benchmark.
Demand Gen campaign, $8,000 monthly spend, goal is paid orders. Audience report segmented by targeting type:
| Source | Spend | Conversions | CPA |
|---|---|---|---|
| Your audiences (CRM + cart) | $3,100 | 62 | $50 |
| Expansion | $4,900 | 49 | $100 |
The instinct is to kill expansion: twice the CPA. Ask three questions first.
- What CPA does your unit economics allow? If the ceiling is $120, expansion is profitable, just less efficient. Turning it off cuts volume by 44% to improve an average.
- What happens to core audiences without it? They are often already at the ceiling of available demand. The freed $4,900 has nowhere to go, so spend simply falls instead of redistributing.
- What is the lead quality from expansion? If lead-to-sale is 12% there versus 30% in the core, the real gap is not 2× but roughly 5× — and that is a genuine reason to switch it off.
The correct sequence: check lead quality in the CRM, compare against your maximum allowable CPA, and only then decide — usually by moving expansion into its own campaign with a tighter target CPA rather than disabling it outright. How to calculate that ceiling is covered in our guide to unit economics and allowable CPA.
FAQ
Is optimized targeting the same as similar audiences?
No. Similar audiences were a static list derived from a seed audience. Optimized targeting selects users dynamically using live signals and the campaign’s own conversion data.
Can I disable it in Performance Max?
There is no separate toggle; expansion is part of how PMax works. You influence it indirectly through audience signal quality, exclusions and asset group structure.
Why did impressions jump but conversions fall after enabling it?
Usually because the model is trained on a weak conversion event. It faithfully finds people similar to those who triggered that event — and if the event was a phone-number click, you get clickers.
Does expansion steal budget from my own audiences?
The budget is shared, so yes, it competes with your core segments. If core audiences deliver a better CPA and you need more volume, raise their priority and move expansion into a separate campaign.
How long before I judge the result?
At least two full conversion windows after the change, and no fewer than 30–50 conversions per arm. Smaller samples produce decisions made from noise.
Which exclusions are mandatory?
Existing customers when the goal is acquisition, irrelevant geos, mobile apps on Display if you have no app product, and brand queries if brand is handled by a separate campaign.
Does it help in a small geographic area?
In a small city it hits the ceiling of available demand quickly and starts serving progressively less relevant users. Controlled geo expansion usually beats audience expansion there.
How do I know the signal is bad?
Warning signs: audience insights full of segments unrelated to your product, expansion delivering mostly returning users, or a sharp rise in junk leads while CPA looks stable.
Should I disable it during a sale period?
Usually not — peak demand is when expansion performs best, because the addressable market is temporarily larger. Do re-check customer exclusions and budget caps for that window.
Does it affect Search Quality Score?
Not directly. Optimized targeting applies to Display, video, Demand Gen and PMax. Search quality is driven by ad relevance, expected CTR and landing page experience.
What if expansion spend keeps growing and the data stays opaque?
Split hypotheses across ad groups — one signal per group — and settle the disputed question with a 50/50 experiment. Transparency comes from structure, not from arguing with the interface.
Do I need an agency account for this?
The feature is available in any account. Agency accounts solve a different problem — stability and higher limits at scale. If that is your constraint, PPC Rebels provides agency ad accounts for scaling advertisers.
Bottom line: optimized targeting is neither an enemy nor a magic button. It amplifies the quality of your conversion signal. Improve the event the model learns from and expansion starts working for you; leave a junk goal in place and it will scale the junk faithfully.
