Remarketing is the most profitable Display campaign there is, and the one most often done badly: an "all visitors, 30 days" list, a generic banner, no exclusions and no frequency cap. It works… and it could work twice as well. This lesson builds it the way it should be built.
Lists that mean something
"All visitors" mixes the person who bounced in 3 seconds with the person who left a €300 basket behind. Lists are built by behaviour, from the Google tag or GA4 (rules by URL, event, duration, pages viewed):
| List | Rule | Intent | Use |
|---|---|---|---|
| Abandoned basket | Reached basket/checkout and did not buy | Very high | Its own campaign, priority budget, closing message |
| Viewed product/service | Product or category URL, no basket | High | Dynamic remarketing (ecommerce) or by category |
| Engaged visitors | ≥ 2 pages or ≥ 60 s | Medium | General reminder campaign |
| Bounce | 1 page, < 10 s | Low | Exclude or don't use |
| Buyers / customers | Thank-you page or customer list | Already converted | Exclude from acquisition; repeat-purchase or cross-sell campaign |
| Open leads | Form submitted, no sale (CRM list) | High for closing | Trust/proof message |
| Customers (Customer Match) | Uploaded emails | Existing | Loyalty, exclusion, or seed for similar audiences/signals |
Minimum size for Display: 100 active users in 30 days; for video/Demand Gen, more. Lists that are too narrow are useless in small accounts: combine them.
Membership duration: the buying cycle decides
How many days someone stays in the list after the visit. Rule: equal to or slightly longer than the decision cycle:
- Impulse purchase / consumer ecommerce: 7-14 days.
- Services people compare (home renovation, insurance, training): 30-60 days.
- Long decisions (property, B2B): 90-180 days (540 maximum).
The same source can feed several lists with different durations (basket 7 days; visitors 30; product 60) so you can use different messages and bids by recency: someone who was there yesterday is worth more than someone who was there a month ago.
Exclusions: half of remarketing
- Recent buyers excluded from the acquisition campaign (nobody wants to see an ad for the tap they bought yesterday for a month).
- Bounces excluded (there is no intent to recover).
- Employees and offices (a list by IP or by email domain) excluded from everything.
- Leads already closed (CRM list) excluded from the closing campaign.
Without exclusions, remarketing spends on the people who need it least.
Frequency: remind, don't stalk
Display lets you cap frequency (impressions per user per day/week/month, at campaign level). With no cap, a visitor can see your banner 40 times in a week: annoyance, wear-out and accidental clicks. Benchmark: 3-5 impressions a day on short-duration remarketing; fewer on long lists. And rotate your creative: the same image for 30 days becomes invisible (fatigue, Demand Gen module in Advanced).
Dynamic remarketing
With a feed (Merchant Center or a business feed), the ad shows the
exact products each person viewed, with price and photo, plus related
products. It is the version that converts best in ecommerce (and in
travel, property and education with their own feeds). It needs the tag
with product parameters (ecomm_prodid, ecomm_pagetype) or GA4 with
ecommerce events.
Judging remarketing honestly
Remarketing claims credit for conversions that might have happened anyway: the person already knew you. Signs of cannibalisation:
- An "incredible" conversion rate (10-20%) with view-through conversions dominating.
- Remarketing conversions that grow by exactly what brand Search conversions fall.
- The account total does not go up when you turn it on.
How to find out the truth: judge by click with short windows; run a holdout experiment (a group from the list with no ads, via a random or geographic exclusion) and compare purchase rates; and look at the total lift for the account, not the campaign on its own. Honest remarketing usually contributes less than its own column claims and more than the sceptic claims: the truth is in the experiment.
💡 Ninja trick: remarketing lists are also the best signal for PMax and Demand Gen and the seed for similar audiences. An account with well-built lists (basket, product, customers, open leads) has Google learning from facts; an account with "all visitors" has Google learning from noise.
What you should remember
- Lists by behaviour (basket, product, engaged, buyers, leads), not "all visitors".
- Duration = buying cycle; several lists by recency.
- Exclusions (buyers, bounces, employees) are half the work.
- Capped frequency and rotated creative: remind, don't stalk.
- Judge it by click, by total account lift and, if you can, with a holdout experiment.