Remarketing in Google Ads still works in 2026, but the mechanism underneath it has changed: browser-based audience lists have become smaller and shorter-lived, while first-party Customer Match lists have become the only audience asset an advertiser genuinely controls. Across the accounts we audit, the pattern is consistent. Display remarketing lists built from site tags have lost a substantial share of their reachable membership over the past two years, particularly on Safari, while accounts that uploaded a customer file kept a stable, targetable audience the whole time. The strategic conclusion is simple: stop treating remarketing as a tag you install, and start treating your customer database as the audience.
Quick reference:
- Customer Match needs at least 1,000 active members in a list before it will serve on any surface
- Typical match rates run 30% to 70% depending on data quality and how many identifiers you upload
- Maximum membership duration is 540 days for Search and Display remarketing lists
- Safari caps script-set first-party cookies at 7 days, which quietly truncates most site-tag lists
- Similar Audiences were removed in 2023 and are not coming back; optimised targeting replaced them
- Remarketing is the easiest place in an account to pay for conversions you would have had anyway
Why have remarketing lists shrunk?
Three separate forces compressed browser-based audiences, and they compound.
Browser restrictions. Safari’s Intelligent Tracking Prevention limits client-side cookies set by JavaScript to seven days, and to 24 hours in some navigation contexts. Firefox blocks known trackers by default. Chrome retained third-party cookies rather than removing them outright, but it added user-facing controls and the practical direction of travel has not reversed. A visitor who returns after two weeks on Safari is, for list purposes, a new visitor.
Consent requirements. Any user who declines advertising cookies is excluded from your lists entirely. In Australia the current framework is less restrictive than the EU’s, but the 2024 Privacy Act amendments moved consent and transparency obligations in a stricter direction, and the direct-marketing provisions apply directly to audience uploads.
Tag reliability. Ad blockers, script errors and slow pages mean a meaningful proportion of qualifying visitors never make it onto the list in the first place. This is the same failure mode that undermines conversion data, and the fix is the same: check that your conversion tracking actually fires before you trust anything built on top of it.
The net effect is not that remarketing stopped working. It is that the audience you think you are targeting is materially smaller than the number in the interface suggests, and skewed toward Chrome users on desktop.
What makes Customer Match different?
Customer Match matches hashed customer identifiers you upload, email addresses, phone numbers, names and addresses, against signed-in Google accounts. Because the match happens on identity rather than on a cookie, it survives browser restrictions, device changes and time. A customer who bought from you eighteen months ago is still matchable; a cookie from that visit is long gone.
Three practical requirements govern whether it works:
List size. Google requires at least 1,000 active members in a list before it will serve. This is the single most common reason a Customer Match list sits unused: the business uploaded 900 customers, saw no delivery, and concluded the feature was broken. Note that the threshold applies to matched, active members, not to rows in your CSV.
Match rate. Uploading email addresses alone typically produces a lower match rate than uploading email, phone and address together, because each additional identifier gives Google another chance to resolve the same person. Across the e-commerce accounts we work with, single-identifier uploads commonly land between 30% and 50%, while multi-identifier uploads reach the higher end of the range. Google’s Customer Match policy documentation sets out the accepted formats and hashing requirements.
Data hygiene. Lists decay. An upload from two years ago that has never been refreshed will lose members to changed email addresses and closed accounts. Automate the refresh, weekly for active e-commerce, monthly at minimum for everyone else.
Which segments are actually worth building?
Most accounts run one list called “All visitors, 540 days” and target it uniformly. That is barely a strategy. The segments that earn their place are the ones where your message or your bid should genuinely differ.
Cart and checkout abandoners, split by recency. Someone who abandoned four hours ago and someone who abandoned three weeks ago are different people commercially. Split at roughly 3 days, 14 days and 30 days, and expect the short window to carry most of the value.
Purchasers, used as an exclusion. This is the highest-value list in most accounts and it is used least. Excluding recent purchasers from prospecting campaigns stops you paying to reach people who just bought, which on a typical e-commerce account is a meaningful share of wasted remarketing spend.
Purchasers, used for a genuine second sale. The same list becomes a targeting list when you have a real cross-sell, a consumable due for replenishment, or a renewal date. Time it to the product cycle rather than to a generic 30-day window.
Lapsed customers. Customers who bought once, twelve to twenty-four months ago, and never returned. Browser lists cannot reach these people at all. Customer Match can, and the economics are usually excellent because acquisition is already paid for.
High-value customers, for value-based bidding. Uploading your best customers as a distinct segment gives Smart Bidding a signal about who is worth more, which matters when your conversion values do not differ at the point of sale but the lifetime values do.
Existing customers, excluded from brand campaigns. Contentious, and worth testing rather than assuming. If people who already know you are searching your name and clicking a paid ad above your own organic listing, you are paying for a click you would have had for free.
When does remarketing destroy margin?
Remarketing reports beautifully and audits badly. It targets people who have already shown intent, so it converts at a high rate and a low cost per acquisition, and the reported numbers look like the best-performing part of the account. Much of that performance is attribution rather than incrementality.
Four failure modes recur across audits:
Paying for conversions that were already coming. If a visitor added to cart, got distracted, and would have returned the next day regardless, the remarketing click did not cause the sale. It intercepted it, and took credit. The larger your remarketing budget relative to your prospecting budget, the more of this you are buying.
No frequency capping. Display and Video campaigns without a cap will serve the same person dozens of impressions a week. Beyond a low single-digit weekly frequency the additional impressions rarely add measurable lift, and they generate brand irritation you never see in the reporting. Note that Performance Max gives you no frequency control at all, one of several reasons we are cautious about PMax for lead generation.
Discount conditioning. Remarketing a discount code to cart abandoners teaches your best customers to abandon carts. The revenue shows up this quarter; the margin damage shows up over the following year and never gets attributed back.
Blending remarketing into a single blended ROAS. Remarketing ROAS and prospecting ROAS should never be evaluated in the same number. Blending them lets high-attribution remarketing conceal prospecting that is losing money. Work out the return each part of the account actually needs with our break-even ROAS calculator and hold them to separate targets.
The honest test is a holdout. Suppress remarketing to a randomised share of your audience for four weeks and compare total conversions, not remarketing conversions. Most accounts that run this test find real incremental value, and find it is smaller than the platform reports.
How should RLSA be used on Search?
Remarketing Lists for Search Ads applies your audience lists to Search campaigns, either as a bid adjustment on existing targeting or as a targeting condition in its own right. It remains one of the most reliably profitable uses of audience data, and it is underused.
The two patterns that work:
Bid adjustments in observation mode. Layer your audience lists onto existing Search campaigns without narrowing targeting, then adjust bids upward for past visitors and purchasers. Low risk, immediately measurable.
Broad match, restricted to a known audience. Run a campaign on broad match keywords you would never bid on cold, targeting only people who have already visited or bought. The audience restriction contains the risk that makes broad match dangerous in an open campaign, while letting you capture query variations you could never enumerate. This works considerably better than broad match with Smart Bidding alone on a cold audience.
Add generic, high-volume, high-cost head terms to the second pattern rather than the first. They are usually unaffordable cold and perfectly affordable to someone already in your funnel.
What has to be in place before any of this works?
Audience strategy sits on top of data infrastructure, and it fails silently when the infrastructure is weak.
You need reliable conversion tracking, ideally with enhanced conversions configured, because the same hashed first-party identifiers that improve conversion matching also improve Customer Match. You need a consent mechanism that records and respects user choices, and passes that state to Google’s tags. You need an automated way to keep customer lists fresh rather than a quarterly manual CSV upload. And you need your list membership durations set deliberately: 540 days is the maximum for Search and Display, but a 540-day cart-abandoner list is not a cart-abandoner list, it is an all-visitors list with extra steps.
One deprecation worth stating plainly, because it still appears in agency proposals: Similar Audiences were removed in 2023. Nothing replaced them as a discrete audience type. Optimised targeting and audience expansion now use your lists as a seed signal rather than generating a separate targetable segment. If a proposal offers to build you similar audiences, the person writing it has not run an account since 2023.
Frequently asked questions
How many customers do I need for Customer Match to work? At least 1,000 active matched members in the list before it will serve on any surface. Because match rates typically run between 30% and 70%, a customer file of roughly 2,000 to 3,000 records is a realistic practical minimum. Uploading email, phone and address together rather than email alone materially improves the match rate.
Is Customer Match legal in Australia? Yes, when you have a lawful basis for using the customer data for marketing and your privacy policy discloses it. The 2024 Privacy Act amendments tightened transparency and direct-marketing obligations, and the Office of the Australian Information Commissioner publishes current guidance on the Privacy Act. Data is hashed before upload, so you are not sending raw customer details to Google, but hashing is not a substitute for a lawful basis.
Do third-party cookies still matter for remarketing? Less than they did, and less than most advertisers assume. Chrome retained third-party cookies rather than removing them, but Safari and Firefox restrictions already truncate a large share of list membership, and consent declines remove more. Treat browser-based lists as a decaying asset and first-party lists as the durable one.
How long should a remarketing list keep members? Match the window to the buying cycle, not to the maximum. Cart abandoners are worth targeting for days, not months. Considered purchases with long research phases justify 90 to 180 days. The 540-day maximum is appropriate for lapsed-customer reactivation and almost nothing else.
Should I remarket to people who already bought? Exclude them from prospecting always. Target them only when there is a genuine second purchase available: a consumable due for replenishment, a real cross-sell, or a renewal. Generic remarketing to recent purchasers is one of the most common sources of wasted spend we find in account audits.
Why does my remarketing ROAS look so good? Because remarketing targets people who already demonstrated intent, so it wins the attribution contest for conversions that were likely to happen anyway. The number is real; the incrementality behind it usually is not as large. Run a four-week holdout against a randomised share of your audience and compare total account conversions to find out what the campaign is genuinely adding.
The accounts that will still have working audiences in two years are the ones building them from customer data now, not from site tags. The tag-based lists will keep shrinking, and no amount of campaign optimisation reverses that. Your customer database is the part nobody can deprecate.
If your remarketing looks like your best-performing campaign and you have never tested whether it is, get in touch and we will design a holdout that answers the question properly.