Retargeting isn't dead. Lazy retargeting is.
AI now runs most of the mechanics of retargeting: audience list-building, creative assembly, and bid optimization. But both **Meta** and **Google** are explicit that AI-driven audience signals are suggestions, not hard boundaries, which means your job shifts from manual targeting to setting tiers, feeding clean first-party data, and policing frequency so the algorithm doesn’t wear out the people most likely to buy.
Every year since about 2019, someone has told me retargeting is dying because of cookie deprecation. It hasn’t died. In April 2025, Google walked back its own plan to phase out third-party cookies in Chrome, announcing it would “maintain our current approach to offering users third-party cookie choice” instead of forcing a new opt-out prompt on everyone[6]. The Privacy Sandbox site that was supposed to replace cookie-based targeting now runs a banner admitting some of its own technologies are being phased out[6]. If you paused retargeting in 2023 waiting for the cookiepocalypse, you waited for something that didn’t arrive on the timeline anyone promised.
And the performance case for retargeting never really wobbled. Retargeted display ads still convert clicks at roughly ten times the rate of standard display, a 0.7% average click-through rate versus 0.07%, according to Invesp’s retargeting benchmark data[7]. That gap is exactly why AI vendors keep building more automation into this specific channel instead of quietly deprecating it.
What actually changed is which data drives the targeting, and how much of the decision-making AI now does instead of you. The old playbook, segment by URL, bid manually, rotate three banner sizes, is gone. What’s left is a mix of first-party pixel data, AI audience expansion, and creative systems that assemble ads on the fly.
If your retargeting strategy is still “show everyone who visited the site the same ad for two weeks,” the technology moved past you, not the cookie policy.
AI is genuinely good at building and refreshing audience lists from behavioral signals: people who viewed a product page, added to cart, watched most of a video, or bought once and are due for a refill. Google Ads calls these your “data segments” and explicitly recommends feeding in website visitor lists, app user lists, customer lists, and video viewer lists as remarketing inputs[3]. Meta’s equivalent is Custom Audiences built from pixel or Conversions API events.
Where AI gets it wrong is priority. Left alone, most platforms will happily spend your retargeting budget on someone who bounced off your homepage in two seconds, right alongside a cart abandoner who left product sitting in checkout. Those aren’t the same person and shouldn’t get the same bid or the same message. I still build tiered audiences by hand: hot (cart and checkout abandoners, past purchasers), warm (product viewers, video watchers past the halfway mark), and cool (general site visitors, blog readers). Then I let AI optimize delivery and bidding inside each tier, not across all of them at once.
Worth remembering: Baymard Institute’s running average across 50 separate studies puts cart abandonment at 70.22%[8]. Seven out of ten people who put something in a cart never buy it. That number is your checkout traffic, day in and day out, not some rare edge case. If your retargeting isn’t structured around that number specifically, you’re leaving the single warmest, most decision-stage pool of buyers you have on the table.
Meta’s default answer to almost everything now is Advantage+, and for retargeting that’s a mixed blessing. Advantage+ Shopping and Advantage+ Catalog campaigns combine prospecting and retargeting into a single campaign, with Meta’s system dynamically shifting budget toward whichever segment, new or returning, is converting best that day. That’s genuinely useful for accounts too small to run separate top and bottom-funnel campaigns.
But there’s a real tradeoff. Hand Meta a broad Advantage+ Audience and just suggest a website-visitor list, and the algorithm treats that list as a starting point, not a boundary. It’ll spend outside it the moment it thinks it’s found better performers elsewhere. For prospecting, I honestly don’t mind that. For genuine retargeting, where the entire point is reaching a specific warm person again, it becomes a problem fast. If you need certainty that your cart-abandoner ad actually reaches cart abandoners, build a Custom Audience directly, website events, customer list, or engagement audience, and target it deliberately rather than trusting Advantage+ to rediscover the same people on its own.
My rule after years of running both: use Advantage+ broad targeting for top-of-funnel prospecting, and keep dedicated Custom Audience ad sets for retargeting tiers. Blend the two into one campaign and you lose visibility into what’s actually driving your retargeting conversions.
Dynamic Creative is Meta’s system for auto-assembling ads from a pool of assets instead of running one fixed creative. You load a set of images or videos, headlines, body text, descriptions, and calls-to-action into what Meta calls an asset_feed_spec, set is_dynamic_creative to true at the ad set level, and Meta’s model tests combinations per impression, learning which pairing of image, headline, and CTA performs best for each person it shows the ad to[5]. Meta’s own developer documentation recommends it specifically for ongoing campaigns and anything running longer than five days, to automate the creative-testing workflow instead of manually split-testing variants[5].
For retargeting, this is where dynamic creative earns its keep: pairing the same product catalog images with different urgency-driven headlines for cart abandoners versus softer “still thinking about it?” copy for browsers, and letting the system find the winning combination per segment instead of you guessing which one wins.
Here’s my honest gripe. Dynamic Creative gets sold as “AI creative that writes itself,” and it doesn’t. It’s a testing and assembly engine, and it needs good raw ingredients. Feed it five mediocre headlines and ten near-identical product shots, and it will very efficiently find you the least-bad combination of mediocre assets. The lift comes from input quality, not algorithmic magic. Spend real time writing genuinely distinct headline variants (urgency, social proof, price, benefit) instead of dumping every asset you own into the feed and hoping the machine sorts it out.
On the Google side, remarketing lives mostly inside Performance Max now. PMax is a goal-based campaign type that runs across Search, Display, YouTube, Gmail, Discover, and Maps from a single campaign, using Google AI across “bidding, budget optimization, audiences, creatives, attribution, and more”[2]. You feed it audience signals, including your own customer lists, website visitor lists, and app user lists, at the asset group level[1]. Google is explicit that these are signals, not hard targeting: “Performance Max may show ads to relevant audiences outside of your signals if they have a strong likelihood of converting”[1].
That’s the honest, slightly uncomfortable truth about AI remarketing on Google: you’re steering, not driving. You can add a cart-abandoner list as a signal, layer in demographics and in-market segments, and Smart Bidding will use all of it to chase your CPA or ROAS target[2]. What you can’t do is guarantee your budget only reaches that list. If keeping strict control over exactly who sees a specific retargeting message matters more to you than automated reach, standalone Display remarketing campaigns built on manually curated “your data” audiences still exist and still give you that control[3].
For product-based retargeting specifically, dynamic remarketing, ads that automatically pull in the exact products or services someone viewed on your site or app via your Merchant Center feed, is still the mechanism doing the heavy lifting inside both PMax and standalone Display campaigns[3].
| Capability | Meta | Google Ads |
|---|---|---|
| Core retargeting audience | Custom Audiences (pixel, Conversions API, customer lists) | “Your data” segments: website visitors, customer lists, dynamic remarketing feed[3] |
| AI creative assembly | Dynamic Creative auto-tests image, headline, and CTA combinations[5] | Performance Max auto-generates text, image, and video assets by asset group[2] |
| Audience control level | Advantage+ treats uploaded lists as expandable suggestions | Audience signals guide bidding, but PMax “may show ads… outside of your signals”[1] |
| Frequency management | Ad set-level frequency controls | Manual or auto-optimized frequency capping (Display and Video campaigns only)[4] |
Source: Google Ads Help Center and Meta for Developers documentation, verified August 4, 2026.
AI copy tools are genuinely good at one thing in retargeting: matching message to funnel stage at scale, faster than a human writing team can turn around fifteen ad set variants. The trick is briefing it like you’d brief a junior copywriter, not asking it to “write ad copy for retargeting” and hoping for the best.
My working framework, split by audience tier: cart and checkout abandoners get urgency and friction removal (still-in-your-cart reminders, free shipping callouts, limited stock). Product viewers who never added to cart get social proof and objection handling (reviews, comparisons, guarantees). Past purchasers get replenishment or cross-sell timing tied to realistic repurchase cycles, not a generic “we miss you.” Blog readers or video viewers who never hit a product page get top-of-funnel education, not a hard sell, because pushing a purchase CTA at someone still researching is exactly how you burn frequency on people who aren’t ready to buy.
Where I push back on AI copy generation: it defaults to generic urgency language, “don’t miss out,” “limited time,” unless you feed it your actual product details, price points, and the specific reason this person didn’t convert the first time. Generic urgency copy is exactly what trained consumers to tune out retargeting ads in the first place. Give the model the specific SKU, the specific price, the specific objection (shipping cost, indecision, comparison shopping) and it writes something worth showing. Skip that step and you get filler that looks like every other retargeting ad in the feed.
Nobody wants to deal with this part. It’s not glamorous work, checking frequency caps instead of admiring a shiny new campaign, but I’ve watched it single-handedly decide whether a retargeting budget compounds or just quietly burns. Both major platforms hand you frequency controls precisely because AI bidding systems, left unchecked, will happily show the same person your ad far more times than makes sense if that’s what the optimization signal rewards.
In Google Ads, frequency capping is set manually per campaign, ad group, or ad in Display campaigns, or you can let Google Ads optimize frequency automatically[4]. Only viewable impressions count toward the cap, and it works off third-party cookies by default, falling back to first-party cookies when those aren’t available[4]. Video campaigns cap separately, by impressions, views, or both, set at the campaign level[4]. Frequency capping isn’t available in Demand Gen campaigns, which is worth knowing before you assume it’s universal[4].
My practical settings after a decade of running retargeting: cap hot-tier audiences (cart and checkout) around four to six impressions per week, go lower on cooler tiers, and refresh creative on an actual calendar instead of waiting for performance to visibly drop, because by the time frequency-driven fatigue shows up in your CTR, you’ve already burned the goodwill. Exclude converters immediately, same day if you can, not next week. Nothing tanks trust faster than retargeting someone for a product they already bought.
If you’re not watching frequency by audience tier, you’re not managing a retargeting campaign, you’re managing an annoyance machine that occasionally converts.
Strip away the platform jargon and every AI retargeting feature described above, Meta Custom Audiences, Advantage+ signals, Google’s audience signals, dynamic remarketing, runs on the same fuel: first-party data you collect directly, not third-party cookie data bought from a broker. Google’s own guidance is explicit that “your data” segments should be built from things you own: website visitor lists, app user lists, customer lists, and video viewer lists[3].
That’s actually good news, because it means the health of your AI retargeting doesn’t depend on how a cookie policy shakes out next quarter. It depends on whether your pixel and Conversions API (or server-side tagging on the Google side) are firing cleanly, whether your email list is current, and whether you’re segmenting purchase and cart events accurately. I’ve seen accounts with sloppy pixel setups blame “the AI” for weak retargeting performance when the actual problem was half their conversion events never firing in the first place.
The upside of doing this well shows up in the numbers. Criteo’s case study on Represent Clothing, an apparel brand that expanded AI-driven dynamic retargeting across Meta and the open web, reported a 56% increase in sales, a 153% increase in ROAS, and an 80% increase in conversion rate[9]. I take vendor case studies with a grain of salt as a rule, but that pattern matches what I’ve seen firsthand on smaller accounts: clean first-party signals feeding a system built to act on them, not a platform promise off a slide deck.
In practice, marketers use the terms interchangeably today. Historically, retargeting referred to ads shown to anonymous website visitors via a pixel, while remarketing referred to re-engaging known contacts, usually by email. Google Ads itself now groups its pixel-based retargeting features under “your data” and “dynamic remarketing,” blending both concepts into one workflow[3].
No. Advantage+ treats an uploaded audience list as a suggestion it can expand beyond, which helps prospecting but works against you when the goal is reaching a specific warm segment like cart abandoners. For precise retargeting, build and target Custom Audiences directly rather than relying on Advantage+ to rediscover the same people on its own.
There’s no single universal number, but a workable starting point is four to six impressions per week for high-intent audiences like cart abandoners, and fewer for cooler audiences like general site visitors. Both Google Ads and Meta let you cap frequency manually or let the algorithm optimize it[4]; watch CTR and conversion rate by frequency band and pull the cap down the moment engagement drops.
Less than the industry expected a few years ago. Google reversed its plan to phase out third-party cookies in Chrome in April 2025, choosing to maintain the existing cookie-choice setup instead of forcing deprecation[6]. That said, first-party data, your pixel, your Conversions API events, your customer lists, is still the more reliable and durable foundation for retargeting regardless of what any single browser does next.
Treating audience signals as a hard boundary instead of a suggestion, then not checking back in. Both Meta and Google state plainly that their AI systems can and will reach people outside the audiences you specify if the algorithm predicts they’ll convert[1]. Set your tiers, feed clean first-party data, then audit delivery and frequency regularly instead of assuming the AI is respecting the boundaries you set on day one.
This article was researched live on August 4, 2026, using official Meta for Developers documentation, official Google Ads Help Center pages, Google’s own Privacy Sandbox blog, and named industry benchmark sources (Invesp, Baymard Institute, Criteo). Every statistic and platform feature described was verified against its original source before publication.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.