Lookalikes Built From Buyers and Lookalikes Built From Browsers Are Different Animals

A seed audience of purchasers teaches an ad platform who buys. A seed of site visitors teaches it who clicks. Confusing the two is how stores scale the wrong crowd very efficiently.

Lookalike audiences inherit the character of their seed, and seeds get chosen carelessly because bigger feels safer. Feed a platform your purchasers and the model finds people resembling the ones who completed the whole journey: intent, trust, price tolerance, delivery expectations, all of it baked in. Feed it all site visitors and the model finds people resembling your traffic, which includes researchers, accidental clicks, comparison shoppers who bought elsewhere, and the simply curious, a far larger and far less committed population. Both audiences can be useful. The visitor lookalike is a prospecting net for reach. The purchaser lookalike is a harvest tool. The failure is expecting harvest results from a net: judging a visitor seeded audience by return on ad spend, watching it fail, and concluding lookalikes do not work, when what failed was the seed choice. Worse is the blended seed, purchasers and visitors stirred together for size, which produces an audience resembling neither, optimized for a customer who browses like a fan and buys like a ghost.

Seed quality beats seed size past a low floor

Platforms ask for volume because their models need signal, and there is a floor below which a seed is noise. Above that floor, purity wins. Two thousand genuine purchasers of the relevant product line outseed twenty thousand mixed visitors for a buying campaign, because the model's job is pattern matching and every non buyer in the seed is a pattern pointing somewhere you do not sell. Segment the purchaser seed by product line where volumes allow, since the buyer of your premium range resembles a different prospect than the buyer of your entry range, and a single store wide purchaser seed averages the two into a person who buys neither.

Recency matters in seeds the way it does everywhere. Purchasers from the current season's range resemble next month's buyers more than purchasers from three years ago resemble anyone still shopping. Refresh seeds on the same cycle as the catalogue turns, and drop cohorts whose products you no longer stock, because their lookalikes will arrive looking for a store that no longer exists.

Related reading on this site: SegmentMuse and Test Creative Per Segment Without Shattering the Budget Into Confetti cover the neighbouring problems in more depth.

Match the audience to the job of the campaign

Prospecting campaigns meant to fill the top of the funnel can honestly use visitor based reach, judged by cost per engaged visit and assisted outcomes over a long window. Buying campaigns, the ones carrying revenue targets, should run on purchaser seeds and be judged like it. The discipline is in the labeling: name audiences by seed and purpose inside the account, buyer seed harvest, visitor seed prospecting, so a future media buyer does not inherit an unlabeled audience and a wrong expectation in the same click.

Exclusions are the other half of seed craft. A purchaser lookalike campaign that does not exclude existing customers spends its budget finding people who resemble your buyers, starting with your buyers, who were free. It sounds too basic to check and it is the first thing to check when a harvest campaign reports suspiciously good returns at suspiciously low new customer rates.

Read results by new customer share, not applause metrics

The tell for a seed problem is new customer share. A purchaser seeded campaign with a low share of first time buyers is rediscovering your base. A visitor seeded campaign with strong click rates and no downstream purchases is scaling curiosity. Both can look fine on the platform's preferred metrics. Set the success metric per audience before launch, in writing, because after results arrive every audience looks successful by some number, and the meeting will find that number. Decide in advance which number is allowed to declare victory, and the seed conversation next quarter starts from evidence instead of from whichever chart somebody screenshotted.

Frequently asked questions

What is the difference between purchaser and visitor lookalike audiences?

A purchaser seeded lookalike finds people resembling those who completed a purchase, carrying intent and price tolerance patterns. A visitor seeded lookalike finds people resembling site traffic, which includes researchers and accidental visitors, making it a reach tool rather than a buying audience.

How large should a lookalike seed be?

Large enough to clear the platform's minimum signal threshold, after which purity matters more than size. A smaller seed of genuine purchasers of the relevant product line usually outperforms a much larger blended seed of purchasers and general visitors for revenue campaigns.

Why exclude existing customers from lookalike campaigns?

Because without exclusions the campaign spends budget reaching people who resemble your buyers, starting with your actual buyers, inflating reported returns while producing few new customers. New customer share is the metric that exposes the problem.

How often should lookalike seeds be refreshed?

On the cycle the catalogue and customer base turn, typically seasonally. Seeds built from discontinued product lines or purchasers from several years ago produce audiences looking for a store that no longer exists.