Do Meta ads drive Amazon sales? How to measure it

Meta ads can drive a lot of Amazon sales, but Meta’s own report will say they drove almost none. Meta’s pixel is code on your website, so it never sees a checkout that happens on Amazon. To know what those ads sell, you have to measure on Amazon’s side and bring the spend over from Meta.

There are three ways to do that. Each answers a different question, and each gives a different number for the same campaign. Below: why the pixel reads zero, the three lenses, a worked example where they disagree, the traps, and a setup you can start this week.

Why Meta shows near-zero purchases for Amazon-bound ads

Meta counts a purchase when its tracking fires. The Meta Pixel is “a snippet of JavaScript code” that tracks visitor activity on your website. The Conversions API does the same job from your own server. Both only report what happens on pages and systems you control.

When an ad sends a shopper to an Amazon product page, the add to cart and the purchase happen on Amazon. Your code is not there. So Meta records a click and then nothing.

That is why an Amazon-bound campaign can show three purchases on $10,000 of spend. The pixel is not broken. It is measuring the wrong store. If an agency tells you to “fix the pixel,” that will not help. A cost per purchase built from those numbers is not cautious. It is simply wrong.

Lens 1: Amazon Attribution tags on the ad link

Amazon Attribution is Amazon’s free tool for measuring traffic that starts outside Amazon. You create a tag, which is a tracked link to your product page, and use it as the ad’s destination. Amazon then reports, for shoppers who arrived through that link:

• clicks and detail page views (views of the product page)
• add to carts
• purchases and sales
• new-to-brand purchases (buyers who had not bought from your brand in the past year)

Amazon’s guide says it uses a 14-day, last-touch model. A purchase counts if it happens within 14 days of the click, and the most recent tagged click gets the credit. As of September 2026, Amazon’s product page lists it for professional sellers in Brand Registry (Amazon’s program for trademark owners), vendors and agencies. It runs in the US, Canada, the UK, much of Europe, and parts of Latin America, the Middle East and Asia.

What it can tell you: how many people your ads sent to Amazon, and what those people bought. Amazon splits sales into the promoted product and “brand halo,” meaning other products from your brand bought by the same shoppers.

What it cannot tell you:

• It has no cost side. There is no spend column. You must join spend from Meta yourself, ad by ad.
• It only sees clicks. A shopper who saw the ad, did not click, and bought later is invisible to it.
• It credits sales that would have happened anyway. A loyal customer who clicks an ad and buys their usual reorder still counts.

The Brand Referral Bonus

If you are eligible, Amazon pays you back on this traffic. The Brand Referral Bonus averages 10% of the sale price and arrives as a credit against your referral fees. Amazon’s page lists it for Brand Registry sellers on a professional plan in the US store. Only traffic that carries an Amazon Attribution tag qualifies, and Amazon’s own ads do not. Amazon’s 2023 announcement said other products from your brand, bought by that shopper within two weeks, also count. Terms change, so confirm them in Seller Central (Amazon’s seller dashboard).

The rate varies by category, so look up your own rate there. Then count the bonus as a share of attributed sales, not as a discount on your ad budget:

• net cost = Meta spend minus the bonus on attributed sales
• net return = attributed sales divided by net cost

Lens 2: the halo nobody clicked through

Many people see an ad on Instagram and do not tap it. Some of them open the Amazon app later and search your brand name. Some buy a different product from you. None of that carries a tag.

You can look for this halo in two places:

• Branded search on Amazon. Brand Analytics (Amazon’s reports for registered brands) includes the Search Query Performance report. It shows impressions, clicks, cart adds and purchases by search term, by week, month or quarter. A rise in searches for your brand name while a campaign runs is a sign of halo.
• Sales of products you did not advertise. If an ad for one product lifts the rest of the catalog, total Amazon sales move more than attributed sales do.

What this lens can tell you: that something moved, and roughly when. What it cannot tell you: why. Season, a price cut, a new product or a competitor running out of stock can all lift branded searches in the same week. A halo read is a clue, not proof. Before counting searches, fix the list of spellings that are your brand’s name, and keep it the same for every week you compare.

Lens 3: holdout and geo tests

A test is the only lens that measures cause. You take ads away from some people or places and compare their sales with everyone else’s.

• A holdout test keeps a random slice of your audience from seeing the ads. Meta’s Conversion Lift compares a test group that sees ads with a control group that does not. The catch for Amazon brands: a lift study compares the conversions you report to Meta. Your Amazon checkouts are not in your pixel data, so ask how the study would see Amazon sales before you pay for one.
• A geo test turns ads off, or down, in some regions and leaves them on in similar ones. You then compare Amazon sales by region. Amazon’s all orders report carries ship-to state and postal code, so you can build the regional totals yourself. Meta publishes an open-source tool, GeoLift, for designing and reading this kind of test.

What a test costs:

• Lost sales. The regions or people without ads buy less, if the ads work. That is the point, and it is real money.
• Time. A test needs several weeks, plus two more to cover Amazon’s 14-day credit window.
• Volume. A small brand may not sell enough per region for the difference to rise above normal noise.
• Clean conditions. A Prime Day, a price change or a stockout in the middle ruins the read.

That is why a test is a quarterly habit, not a weekly report.

A worked example: one campaign, three answers

Illustrative numbers only. One Meta campaign for a body lotion sends shoppers to its Amazon page. Take one closed week, read about three weeks later so Amazon sales have mostly settled.

Now the return on that $10,000, by lens:

Same week, same ads, three answers from 0.01x to 1.60x. None of them is a mistake. They measure different things.

Here is one way the test and the tags could fit together. The tags credited $13,000. Say $3,000 of that came from regular buyers who would have ordered anyway. The test also caught $6,000 of sales from people who never clicked, through branded searches and other products. $13,000 minus $3,000 plus $6,000 is $16,000.

The test could just as easily land below the tags. That happens when many attributed buyers were going to buy anyway and the no-click halo is small. You do not know which way it goes until you test.

Whether 1.49x is good depends on your margin. Compare it with your break-even ROAS (the sales-to-spend ratio where an ad just pays for itself), not with a number from someone else’s category.

Where it goes wrong

These are the traps we see in real multi-channel data.

Judging Amazon-bound ads on pixel purchases

The most common and most expensive one. Ads Manager ranks every Amazon-bound ad near the bottom, so the budget moves to ads that point at your own store. The Amazon program gets cut for looking like a total loss when it may be your best spend. Split spend by destination before you read any return.

One ad account, several destinations

One Meta account often runs campaigns for Amazon, your own store and a second retailer at once. A blended account ROAS then scores nothing correctly. Pixel purchases undercount the Amazon ads. Attributed sales divided by all account spend undercount them too, because store and retailer spend sits on the bottom. Scope Amazon spend to the ads that carry an Amazon tag. Score the retailer campaign on its own, or say plainly that nothing scores it.

Reading a fresh week as a drop

Amazon Attribution keeps crediting sales for 14 days after each click. So a click on the last day of a week can still earn a sale two weeks later. In the data we see, sales can also dip a little as orders are cancelled, while clicks barely move once the week closes. That means a fresh week can show a sales drop that is only timing. Compare clicks for recent weeks, and sales only once both weeks are settled. We cover this pattern in more depth in why Amazon ad sales numbers are not final.

Spend and sales at different levels

Meta reports spend per ad. Amazon Attribution reports sales per tag, grouped under campaign and ad group names that someone typed by hand. If five ads share one tag, their sales cannot be split between them. If you match Meta ad names to Attribution names, a renamed ad or a typo makes the match fail. The spend is still there, but its sales vanish, and nothing warns you. The result is a return that looks worse than the truth.

A quick check catches it: divide Amazon Attribution clicks by Meta link clicks for the same ads and days. Most of the account should sit near one rate. An ad far below that rate is sending its shoppers through some other link, so its own sales read near zero while it is working. Fix its link before you cut it.

Dividing a full week of spend by a partial week of sales

If the Amazon data for the week is only half loaded, the return looks like a disaster. Check that spend and sales cover exactly the same days before you publish a number.

How to set it up

1. Tag every Amazon-bound ad. Create one Amazon Attribution tag per ad, not one per product. Shared tags can never be split later.
2. Name ad sets (the level between campaign and ad) so a person can read them. Put the product and the destination in the name, for example Body lotion 8oz | Amazon | Prospecting. Campaign names usually carry strategy, so the product has to live one level down.
3. Keep a tracker of ad to tag. One row per Meta ad, with its tag link. This sheet, not the ad name, is how spend joins to sales.
4. Join spend to attributed sales by ad. Take Meta spend per ad per day, join each ad to its tag, and sum over the same days. Keep shared-link sales as their own row instead of dropping them.
5. Net the bonus, if you are enrolled. Use your own category rate.
6. Compare only settled weeks. Wait at least 14 days after a week closes before you judge its sales, and expect small changes after that. Use clicks for anything newer.
7. Watch the click rate. Amazon clicks divided by Meta clicks, per ad, every week. A sudden drop is a link problem, not a creative one.
8. Run one holdout per quarter. Pick a quiet month with no big deals. Turn off Amazon-bound ads in a set of matched regions for several weeks. Compare Amazon sales by region, then compare that answer with what the tags said.

Synthesis records whether each ad sends people to Amazon or to the store, as a definition every answer shares. Meta numbers carry a label saying the pixel cannot see Amazon checkouts, so the caveat travels with the figure.