Google retargeting (remarketing) ads: bring people back without being creepy

Retargeting shows ads to people who already interacted with you. It needs a relevant follow up, allowed data use and a campaign that respects the audience you set. Exclude people who already bought, and match the message and the window to the buying decision. Then judge it on qualified outcomes, not attributed ROAS, because people coming back anyway flatter any campaign.

What retargeting does and when it is worth running

Retargeting and remarketing mean the same thing. Google now calls the audiences "your data". Each one is a segment built from an earlier interaction:

• a website visit through the Google tag

• an app event

• a video view

• a customer record you uploaded with permission

You work with segments and counts, not a list of names. That does not make it anonymous. The segment is built from personal information. And the person on the other end can tell they are being followed.

It is worth running when three things are true at once. A follow up would help the person. A relevant audience exists and can be reached. The tracking, creative and destination are ready.

It is not worth running yet when the campaigns feeding it are starved. Or when the decision is urgent and over within hours. Or when there is no creative worth showing twice. Or when the service sits in a category the platform or the law restricts.

Whether paid advertising is viable at all is the earlier question. Our guide to whether Google Ads is worth it for a small business answers it.

Run that test inside a properly built account. Our guide to setting up Google Ads for a small business describes how to build one.

Where Google retargeting ads can appear

Retargeting is an audience strategy, not a campaign type. Which campaign type you run it in decides how much control you keep. Our guide to Google Ads campaign types compares them.

On Search, the setting matters more than the audience. Targeting restricts the ads to people in your segments. Observation doesn't change who can see them. Google recommends Observation for all Search campaigns. Google also uses the first party segments you observe as Smart Bidding signals.

So a Search campaign is rarely a retargeting campaign in the strict sense. It is a Search campaign that knows who came before.

On Display, the ground is moving. Display campaigns are becoming Google Display Network inventory inside Demand Gen. The key dates and limits:

• The migration tool began a phased rollout in June 2026.

• Existing campaigns stay editable until they are migrated. Migration cannot be reversed.

Not every account has migrated yet. And not every Display feature survives the move. Check what your own account offers before you copy last year's setup.

Demand Gen can serve on YouTube, Discover, Gmail, Maps and the Display Network. In the ad group settings, its channel controls let you pick which ones it uses. Maps only serves when a non affiliated location asset is enabled on the account.

Performance Max, the campaign most often sold as retargeting, isn't one. Its audience signals are suggestions that help Google's AI optimise. Your past visitors and customers can be one of them. Performance Max may still show ads to relevant audiences outside the signals. Our review of Performance Max for small businesses covers what you can and cannot restrict.

Dynamic product ads fill the ad with items from a feed. The feed, the tag and the campaign type have to agree. The ecommerce examples below cover how.

Audience minimums and who you can reach

Four numbers get confused with each other:

• the rows in your spreadsheet

• the people Google could match

• the active eligible users in the segment for one network

• the people reached in a period

Only the last two decide whether ads serve. GA4's user counts are not Google Ads serving counts.

We ran 180 Google Ads account audits from August 2021 to September 2026. Across them, customer lists lost people at every one of those steps. The funnel follows a 5,000 row list through them.

Plan from the bottom of that funnel, not the top. The upload is the biggest number you'll ever see. It decides nothing.

The floor comes from Google's guide to how your data segments work:

• A segment needs at least 100 active visitors or users in the last 30 days to serve. That goes for the Display Network, Search and YouTube.

• Customer lists uploaded before 1 February 2024 need 1,000 matched users instead.

• The default membership duration is 30 days and the maximum is 540.

• A return visit resets the clock. Shortening a duration removes earlier members.

• Visitors join within seconds of a tagged page view. A new list can take 48 to 72 hours to fill.

Then there's the catch: Google's page on your data segments for Search ads. It still says a segment needs 1,000 cookies before it can tailor Search ads. The two pages disagree for Search.

Treat 100 active users in 30 days as the documented floor for Display and YouTube. For Search, read the segment's eligible size and status in your own account. Trust that over either page.

Eligibility is a technical floor, not a budget plan. If a segment shows as too small, check four things:

• the tag is collecting

• consent and data sharing settings allow it

• exclusions have not emptied it

• it has had time to fill

If the segment is still too small, combine similar behaviours or wait for more suitable traffic. Do not grow it with pages that should not be tagged. Do not stretch the window past the real buying decision just to reach a number. Geography, exclusions, identity matching and privacy choices will all shrink the reachable audience again after launch.

Google tag, GA4 audiences or Customer Match?

GA4 audiences need Google Signals or user provided data collection switched on for the property. They also need a link to Google Ads with personalised advertising enabled. Google's guide to remarketing with Google Analytics data sets out both.

Audiences then export to the linked account automatically. That can take up to two days. Our guide to linking GA4 and Google Ads covers connecting the two properly.

Customer Match needs more care than the tag. Google's Customer Match policy sets the rules:

• Every policy compliant advertiser can use it for Observation and Exclusions.

• Targeting and manual bid adjustments need more. They need enough account history and lifetime spend, plus a good record of policy compliance and payment.

• Only customer information collected in the first party context may be uploaded.

• The privacy policy must disclose that customer data is shared with third parties. Consent is also needed where the law requires it.

• Memberships expire. A list needs members added or updated regularly to stay eligible.

• The data may not be used to target sensitive interest categories, or to identify them among your customers.

Two of those numbers get misread. The 100 member rule is list maintenance, not a serving threshold. And USD $50,000 is about $71,830 at the RBA rate of 13 October 2026. But Google converts your spend to US dollars at each month's average rate. So the Australian figure moves with the exchange rate.

In practice Customer Match works on Search, the Shopping tab, YouTube, Gmail and Display. Google recommends uploading through Data Manager or its API.

Campaigns using Smart Bidding and optimised targeting automatically include every Customer Match list in the account. You have to opt out or remove specific lists to stop that. So an uploaded list can steer automation across the account without anyone choosing to target it. Review that setting. Don't discover it by accident.

Hashing a list before upload is a transport detail. It does not make an unauthorised list authorised.

Build audiences around the buying decision

Windows and overlap

A membership window is how long a person stays in a segment after the interaction. It is not an ad sequence.

Nested seven, 30 and 90 day lists overlap completely. Everyone in the seven day list is also in the other two. So a person can get the "late stage" message on day two.

To turn those lists into real stages, exclude the shorter segment from the longer one. The 8 to 30 day stage excludes the seven day segment. The 31 to 90 day stage excludes the 30 day segment. That works where the source supports it.

Even then, a repeat visit resets membership and sends the person back to the start. Google promises no delivery order.

Those windows suit a long decision. They are not defaults. An emergency repair is decided within hours. A 90 day window on that audience mostly reaches people whose problem was fixed weeks ago.

Set the window from how long the decision takes. Shorten it when the data says the decision is faster than you thought.

Service and ecommerce examples

Warm does not mean commercially useful. Staff, job applicants, suppliers and existing customers all visit service pages. So do people who were never going to qualify. A pricing page view is a hypothesis about buying readiness, not proof of it.

For a store, dynamic ads add a feed. Dynamic remarketing shows previous visitors more personalised ads based on the products or services they viewed. The site has to be tagged for it. Dynamic remarketing works with Display, Performance Max and App campaigns. It's fed from Merchant Center or a dynamic ads feed.

Demand Gen has its own route through product feeds. They need a Merchant Center account with Shopping ads enabled, an approved feed and a linked account. Approval for a new feed can take up to three days. All products are included by default unless filters narrow them. Google's machine learning picks which relevant products to show.

Neither route promises that a viewer sees the exact item they looked at. Product identifiers, prices and availability on the landing page must agree with the feed.

Converter exclusions

A person who enquired by form, then by phone, then became a customer is one person. The campaign should stop treating them as three prospects. Build exclusions from the website conversion and from phone and booking outcomes where the platform supports them. Add exclusions from the CRM through a customer list. Keep them current.

Audience exclusions work in Search, Display, Demand Gen, Standard Shopping and video campaigns. You can add them at campaign or ad group level. They apply to first party data. But they may not apply to people who opted out of ads personalisation.

Add matching gaps, consent choices and update delays to that. So no exclusion removes every converter instantly. A customer list only excludes the customers Google could match.

A new enquiry is a different case again. It often needs a different message before the sale, not permanent exclusion. That is a separate audience with its own purpose.

In the same audits, about 1 in 2 accounts running remarketing excluded neither converters nor existing customers. No list or membership rule kept them out. Those campaigns were paying to advertise to their own customers.

Set up a campaign with deliberate targeting

After the eligibility checks, choose one campaign purpose and one tested commercial conversion. Set the geography, the audience source and the spend the business authorises. Add the intended segment and the converter exclusions. Choose the channels. Then deal with expansion, because a campaign meant for past visitors only has to say so.

Optimised targeting is the reason:

• It looks for extra conversions among people likely to convert beyond the segments you selected.

• It is turned on automatically in Display, Demand Gen and some Video campaigns.

• You can turn it off in the ad group settings.

• Your selected segments become hints for it, not limits.

• Its traffic shows in the "Total: Expansion and optimised targeting" row.

If the campaign is meant to reach only your segments, turn it off. Then read that row to confirm it is empty. Lookalike and custom prospecting segments are not lists of previous visitors, whatever they are called.

In the same audits, we looked at Display and Demand Gen remarketing. About 2 in 3 accounts running either had an ad group the owner described as remarketing only. That ad group had optimised targeting on. Look at yours before you judge the results.

Check the content suitability settings and the actual placements too. A relevant audience does not make every site, app or video a suitable place to appear. Perfect placement safety does not exist. A blanket rule against all apps is not a substitute for reading the placement report.

After a Display migration, go through every one of these controls again. That's because only the settings Demand Gen supports carry over. The rest of the campaign's controls are in our guide to Google Ads settings that waste budget.

Write a useful follow up ad and landing page

The test for a retargeting ad is simple. It either continues the person's decision or it announces that you watched them make it.

The right column only works if the proof behind it is real. Actual completed projects. An actual quoting process. Never claim review counts, qualifications, warranties, fixed prices or scarcity you can't back up.

Each ad goes to a destination that continues the same thought. For a renovator that is the project gallery and the quoting page, not the homepage. Our guide to Google Ads landing page vs homepage covers that choice.

Test the cart and product links. Keep personal data out of the URLs. If the headlines need work, use the rules in how to write Google Ads. They apply here as much as on Search.

Control frequency and watch for saturation

Most retargeting can no longer be capped by hand. Frequency capping is only available in Display and video campaigns. In Display, a manual cap goes on the campaign, ad group or ad. It can only be set after the campaign is created. Only viewable impressions count.

Video caps are set per campaign and apply to in stream and bumper ads in the auction. Demand Gen doesn't support frequency capping at all. So as Display moves into Demand Gen, the manual cap moves out of reach. A cap on one campaign is not a cap across the account. And it does nothing on other platforms.

When no cap exists, the controls are indirect:

• shorter windows

• tighter exclusions

• a smaller budget against a small audience

• more creative variety

• fewer channels

Then read what happened.

Google's reach and frequency metrics for Display, Demand Gen, Video and App campaigns show what actually happened:

• Google estimates unique users, not cookies, with statistical models across browsers and devices.

• It reports average impression frequency per user over 7 and 30 days.

• It reports a frequency distribution of 1+, 2+, 3+, 4+ and 5+. The top band is 10+.

• A minimum number of users has to be reached before any data shows. The data takes up to three days to arrive.

• The distribution table only works for date ranges of 31 days or less.

The same audits included uncapped Display and Demand Gen retargeting campaigns. The middle half averaged 8 to 12 impressions per person over 30 days. For scale, Google's Target frequency setting lets YouTube awareness campaigns aim for 4 to 12 a month.

An average still hides the tail. Ten a month can mean a tenth of the audience seeing the ad forty times. Only the 10+ share of the distribution shows it. The distribution is the number that matters. Do not swap the list size in for measured reach and call the result frequency.

Set a budget the audience can support

The budget follows the reachable audience and the job the campaign has. The arithmetic below shows how to think about it.

A thousand people reached twelve times each is 12,000 impressions. At an assumed $12 per thousand, that is $144 of media for the month.

Twelve impressions is not a recommendation. Demand Gen won't enforce it as a cap, either. Reach is modelled. Prices move with competition and placement. And none of it says what the 1,000 people were worth.

A small warm audience cannot absorb a large budget without the frequency rising or the targeting widening. There are only two honest ways to spend more. Reach more eligible people, or accept more exposure per person.

The opportunity cost sits on the other side. Starve the acquisition campaigns that feed the audience and the follow up has nothing to follow.

Bidding should match the campaign type, the goal and the reliability of the conversion data. A click strategy does not optimise for qualified customers. And Manual CPC doesn't survive the move to Demand Gen. Choosing a strategy is the job of our guide to Google Ads Smart Bidding.

Measure qualified outcomes and extra sales

Track these:

• spend

• eligible reach

• exposure

• qualified leads or orders

• revenue quality

• contribution

Separate three kinds of credit: click through, engaged view and view through. Each has its own window.

Google reports view through conversions in their own column and in All conversions. But it keeps them out of the main Conversions column for most campaigns. The exceptions are App campaigns and Performance Max for store goals. The same goes for Demand Gen campaigns opted into view through optimisation. So check which column your report reads before you believe it.

A person who saw an ad and later bought is one observation. The impression may have changed nothing, because the audience is, by definition, people who already knew the business. The platform's attributed total is not a count of extra customers. Lifecycle stages are not extra customers. Two platforms' totals do not add.

In the same audits, we checked the accounts whose reports counted view through credit. Each check covered the 90 days before the audit. For the middle half of those accounts, view through credit was 40% to 60% of the reported conversions. That is the gap between a campaign that looks brilliant and one that is proven to work.

Only a controlled lift or holdout test answers the incremental question. Use one where volume and access allow it. In smaller accounts, triangulate the CRM results with business trends. Then say plainly that the reading is observational.

Before and after comparisons move with the seasons and with every other channel. A rise in branded searches or returning visitors is not proof of incrementality.

The measurement has to capture the qualified outcome first. Our guide to Google Ads conversion tracking setup shows how. The all in return and the test methods belong to our guide to Google Ads ROI.

Privacy and sensitive service restrictions

Two separate rulebooks apply. A permission under one is not a permission under the other.

The Privacy Act and tracking pixels

On 24 June 2026 the OAIC announced two privacy determinations, both dated 11 June 2026. One is against Medmate Australia, a telehealth provider. The other is against Monash IVF, a fertility services provider. Both cases involve third party tracking pixels. The Commissioner found each interfered with the privacy of people whose sensitive information was collected through those pixels.

The announcement states the principle plainly. Say a health related website uses pixels to track visitors, then targets them with social media ads. That amounts to collecting sensitive information. The website provider must obtain consent for that.

This is existing law applied to a common setup. It is not a new law banning pixels. And the cases concern health services rather than finance. Who the Privacy Act covers is in our guide to Australian privacy law changes for websites. So is what's changing.

The OAIC's guidance on tracking pixels and privacy obligations asks APP entities to assess:

• collection

• consent where required

• disclosure

• direct marketing obligations

• data minimisation

Sensitive information should only be collected through a pixel with express consent. And a visit to a mental health or counselling site can reveal sensitive information on its own. A privacy policy is not a consent process. A banner is a mechanism, not permission.

Keep email addresses, phone numbers, health details and sensitive page context out of URLs and tags. Don't fire the tag on every page. Tag the pages you may tag. What your privacy policy should then say is covered in our guide to privacy policies and website terms.

Google's sensitive interest categories

Google's rules sit on top of the law. Under its restricted targeting policy, promoting something in a sensitive interest category rules out advertiser curated audiences. Those are Customer Match, your data segments, audience expansion and lookalike segments. So a clinic cannot retarget its treatment page visitors, whatever its consent banner says.

The service you promote decides, not your industry. Google's category pages give examples and say the lists are not exhaustive. Four of them touch legal, finance and health services:

A mortgage broker, a conveyancer or a commercial lawyer isn't on those lists. That's why the policy doesn't stop them building segments from their own visitors and customers. Keep pages about debt help, divorce or criminal charges out of every segment. Check any service the lists don't settle against the category pages.

The policy's consumer finance, employment and housing rules apply only in the United States and Canada. Google's predefined audiences include affinity, life events, location and in market segments. They stay open to every advertiser because Google builds them without sensitive signals. And nobody under 18 is eligible for personalised ads of any kind.

Review the first month and decide what comes next

Review in that order, not weekly by habit. The first days are for delivery, tracking and placements. Outcomes need the conversion lag plus enough volume to mean anything. For a small audience that can be the whole month. The wider first month routine is in our guide to the first 30 days of Google Ads.

What we recommend at Elev8d

Next steps

Written by Ajay K.Co-founder, Elev8d

Ajay is the co-founder of Elev8d and writes most of what you will read here. Psychology grad turned marketer, which is a fancy way of saying he cares about why people click, not just where things rank. Every guide is written in plain English for Australian businesses, tested on real campaigns before it earns a spot on this site.