Meta Extends Purchase Audiences to 730 Days: Bigger Remarketing Pools, but Mind the Automation

What changed on May 18

The purchase-audience lookback used to cap out at 180 days. As of May 18, 2026, it reaches back 730. Anyone who bought from you in the past two years can now seed a retargeting audience or feed a lookalike. For a lot of cross-border accounts that single config change roughly doubled the pool overnight.

Why it matters depends entirely on your sales cycle. If you sell something people buy on impulse and never again, a longer window mostly adds noise. If you sell something with a long deliberation period or an annual repurchase rhythm, six months was always too short a memory. You were dropping buyers from your seed lists right as they were getting ready to come back.

The mechanics are the same as before. You build a custom audience off the purchase event, or you hand that audience to Meta as a lookalike source. The only difference is how far back the source data goes. Bigger source, in theory better lookalikes, because the model has more examples of what your real buyer looks like.

One caveat up front. A 730-day window does not mean you should treat all 730 days as one audience. Intent decays. A person who bought in week one is a different prospect from one who bought 22 months ago, and pointing the same ad at both wastes budget. Segmentation is the whole game here, and most of this article is about that.

Which categories actually benefit

The longer window pays off when the gap between purchases is naturally long, or when the decision itself takes weeks. Here is the rough split.

For high-consideration goods the math is straightforward. Someone who bought a $1,200 sofa is not buying another next month, but they are a credible lookalike seed and a real referral target. Keeping them in your modeling set for two years gives Meta a cleaner picture of your customer than a six-month snapshot ever did.

Seasonal sellers get the most obvious win. If your peak was last November, those buyers used to age out of your purchase audience by spring. Now they sit there waiting for you to re-engage them ahead of the next season. That is a warm list you were previously rebuilding from scratch every year.

The categories where I would not bother: anything bought on impulse with no repurchase pattern. A two-year-old buyer of a novelty gift tells the algorithm almost nothing about who buys next. You will pay to reach a larger, colder pool for no lift. Keep those audiences short and spend the saved budget on prospecting.

Segmenting the 730-day pool

The mistake to avoid is one giant “purchasers, 730 days” audience getting the same creative and the same bid. Split it by recency and treat each band as a different relationship.

A practical breakdown: 0 to 30 days for post-purchase cross-sell and accessories, 31 to 180 days for replenishment and category expansion, 181 to 365 days for a “we miss you” win-back, and 366 to 730 days for reactivation with a stronger offer or a fresh product angle. Build these as separate custom audiences using the purchase date, and exclude newer bands from older ones so a recent buyer never lands in a stale reactivation campaign.

Match the message to the band. The 30-day group does not need a discount, they need a reason to add to what they already bought. The 18-month group has probably forgotten you, so lead with what is new since they last shopped, not with a generic “come back” line. Same audience source, four very different ads.

Lookalikes deserve their own thought. Seeding a lookalike off the full 730-day purchaser file gives you volume, but you can also seed off just your high-value or recent buyers to bias the model toward better prospects. Run both as separate ad sets and let the data tell you which lookalike converts. The longer window gives you enough seed volume to do this without the audience getting too small.

One housekeeping note. Longer audiences mean more overlap between ad sets if you are not careful. Use exclusions aggressively so your reactivation campaign and your prospecting lookalike are not bidding against each other for the same person.

The automation traps

Two defaults are working in the background and you should check both today. Advantage+ automation is now on by default for new campaigns, and Meta will auto-add products it finds on your website that are not in your catalog.

Auto-catalog-add is the one that bites cross-border sellers. Meta crawls your site, finds product pages, and pulls them into your advertised catalog without asking. If your site lists SKUs you do not ship to a given market, or old products, or regional variants at the wrong price, those can start showing in ads. Open Commerce Manager, review the catalog source settings, and confirm what is being auto-pulled. Turn off the auto-add or set up filters if you sell different ranges to different regions.

Advantage+ by default is not bad, but “default” is the operative word. It will make targeting, placement, and budget decisions you did not explicitly approve. Before you let it run a meaningful budget, check the audience controls to confirm it is not blowing past the geos you actually ship to, and check that the products it is promoting match the catalog you just audited. The two automations compound: Advantage+ can happily advertise a wrong SKU that auto-catalog-add quietly inserted.

There is also a creative-disclosure rule to plan around. Meta groups its AI-generated creative options into named buckets such as “Refined product look,” “Popular in your niche,” and “High return on ad spend.” If you use Meta’s generative tools to make or edit creative, an AI label is now required, and that label becomes part of the ad people see. Decide in advance whether labeled AI creative fits your brand before you build campaigns around it, because you cannot quietly strip the label later.

The audit cadence I would set: review auto-added catalog items weekly, and check Advantage+ geo and product scope every time you launch or scale a campaign. Five minutes each, and it saves you from spending two-year-window budget on the wrong markets.

Using the AI assistant and Perplexity connector

Meta opened an AI business assistant to all advertisers on April 24. It lives inside Ads Manager and Business Suite, takes natural-language questions about performance, surfaces benchmarks, and helps troubleshoot. For the 730-day rollout it is genuinely handy because you can ask plain questions about audience behavior instead of building pivot tables.

Useful prompts in practice: ask it which of your purchase-recency segments is converting best, ask how your reactivation campaign compares to category benchmarks, or ask why a particular ad set’s frequency is climbing. Treat the answers as a fast first read, not gospel. The assistant sees your account data, but it does not know your shipping constraints or which SKUs you actually want to push.

The same May batch added Perplexity to Meta’s official AI connectors, alongside ChatGPT and Claude, for conversational account analysis. The value of having external assistants connected is cross-checking. If the built-in assistant and Perplexity read your segment performance the same way, you can act with more confidence. If they disagree, that is your cue to look at the raw numbers yourself.

Where this lands for a cross-border seller: the assistant is good for the “what is happening” questions and for catching the automation problems above faster than scrolling through reports. It is not a substitute for deciding which two-year-old buyers are worth reaching. The window got longer and the tooling got chattier. The judgment about who to actually pay to reach is still yours.