Product Performance in Google Ads: What Sells, How Long

Only a small part of a catalogue makes money from Google Ads, and usually not for long. In our study of 1.4 million products over 13 months, 7.9% of products sold and took 78.1% of ad spend, 37.1% spent without a single sale, and 55.0% got no budget at all. A steady bestseller lasted a median 3 months. So judge each product by its behaviour over 12 months, sort the catalogue into groups by sales, spend and ROAS, and give each group its own budget rule.

What is product performance, and where does Google show it?

Product performance is how each product in your catalogue does in your ads: how often Google shows it, how many clicks it gets, what it costs and how much revenue it brings back. Google counts these numbers per product, so every SKU, with its own ID in the feed, has its own results. One model in seven sizes is seven sets of numbers.

In Google Ads, these numbers sit in the product reports. The same data is available in the Google Ads API as shopping_performance_view. According to Google’s guide to Shopping ads in the API (Reporting), product reporting covers every campaign type that uses products from a Merchant Center feed: Shopping, Performance Max, Demand Gen, Video, App and Display.

The product performance metrics look familiar, but Google counts some of them differently from campaign reports:

Divide a product’s revenue from ads (its conversion value) by its spend, and you get its ROAS. That one number answers most of the question “does this product pay?”, as long as you read it per product. The guide to product-level ROAS explains how a healthy account average can hide products that lose money.

Two caveats apply before you read a year of history:

1. Google widened product reporting in June 2026. Before that, product-level spend and conversions for Performance Max did not cover all of its networks. From 15 June 2026 they do, and Google warns of a one-time increase in reported metrics for Performance Max (Retail Campaign Performance). Compare months on either side of that date with care.
2. Product attributes are recorded as they were at the time. Brand, category, product type and custom labels reflect their state when each click or conversion happened. If you relabel products today, older rows keep the old labels.

7.9% of products sell, and they take 78.1% of the budget

In our study of 1.4 million products across 130+ stores over 13 months, the typical catalogue split into three unequal parts:

A small share of products also took most of the spend. In the median store, the top 1% of products took 27.7% of spend and the top 10% took 68.1%. That split is sharper than the familiar 80/20 rule, and the 80/20 rule on ad accounts shows by how much.

The part that spent without selling deserves the closest look. The median store put 40.5% of its budget into products that never sold in 13 months. Among stores in the top quarter by ROAS, the median was 39.8%, and in the bottom quarter 71.8%. Account ROAS hides this split: you see it only when you look product by product.

Every number from our data in this guide comes with two caveats. They are observations across stores, not an experiment, so they show what goes together, not what causes what. And our ROAS counts revenue, not profit, because product costs aren’t in our data.

Which products a store carries at all is an assortment decision, covered in the assortment planning guide. This guide stays with the products you already advertise: which of them earn, for how long, and what to do with each.

How long does a bestseller keep selling?

Not long: a steady one lasts about a quarter, and most winning products sell only once. In ad data, a bestseller is more often a short run than a permanent fixture. In our study of 1.4 million products:

• 64.7% of winning products were one-offs: one conversion in one month, and no more.
• A steady bestseller, a winning product with sales in at least 2 months, lasted a median 3 months.
• 71% of bestsellers stayed in the top for 3 months or less out of the 13.

The study also found no pronounced seasonality in which products were winning. The top changes all year, not only around the peak season.

So a product’s lifecycle in ads is measured in months. A list of “our best products” made once goes out of date within a quarter, and so does any campaign structure built around it. In ad data, the lifecycle has a recognisable shape:

The early stages are long. Of 1,196,357 products that started the 13 months with no spend, 74.1% never got a single impression and 2.59% made a sale. Those that sold took a median 4 months to their first sale, and more than half of them went through a “clicks but no sale” stage first. The study puts the minimum honest test at 15–20 clicks, about $3–4 at the sample’s average cost per click of $0.20.

Each stage has its own guide:

• How the three months were counted, and what they mean for planning: how long bestsellers last.
• How to tell a true bestseller from one good month: identify your real bestsellers.
• Which early signals point to the next one: find the next bestseller.
• How bids, budget and campaign placement should change from launch to decline: product life cycle marketing.

Products with one or two orders bring 57.0% of revenue

The long tail is the mass of products that each sell rarely but together add up. Chris Anderson set out the idea in his 2004 Wired article The Long Tail. He argued that with unlimited shelf space, enough niche products can add up to a market bigger than the hits.

Ad-driven stores show the same shape. Products with one or two orders in a month make up 64.9% of the products that sell and bring 57.0% of revenue (GetProfit data, 116 stores, June 2025 – June 2026, median across store-months). This is a different sample and window from our study of 1.4 million products, so read the two sets of numbers separately.

The tail is also the part that changes fastest:

Two practical points follow. First, the tail changes too fast to manage product by product: three in four of this month’s tail products were not selling a month earlier. Second, any rule that cuts the tail touches the products behind 57.0% of revenue in the median store-month. So judge the tail as a group, by the group’s spend and revenue.

The guide to long tail theory in e-commerce shows how to check whether tail products really add up in your store. A separate guide, long-tail products in Google Shopping, covers how to set up Shopping and Performance Max for a tail that keeps changing.

Which products lose money, and should you switch them off?

Two kinds of product lose you money: products that sell below your target, and products that spend without selling at all. Only products with spend can waste money, so start the review with them. The ones that spend and never sell are often called zombie products. In the portal they’re labelled Dormant, or To remove once they have burned a noticeable budget.

Switching them off is riskier than it looks. In our study, a rule like “no sale after N spent, switch it off” lost accuracy as N grew: it was right in 88.1% of cases at $5 or more, and in 66.2% at $500 or more. Of products that had spent at least the cost of one conversion without a sale, 35.4% converted in later months, after a median 2 months. Of products that had spent less than that, but more than zero, 10.6% recovered.

So check the reasons first (price, listing, category), then cut the product’s share of the budget rather than dropping it to zero. The campaign structure the portal recommends excludes a product after 50 or more clicks with no conversion over 360 days.

Products that sell but miss the target are a different case. Demand is there, but the return is too low, so the fix may be the price, the bid or the budget. When the spend is large enough for the conclusion to hold, the portal calls it spend with nothing back: products that spend noticeable money and never pay it back. The guide to borderline and loss-making products shows how to choose between a new price, a lower bid and an exclusion.

What happens when a bestseller stops selling?

Because so much rests on a few products, losing one hurts. In our study, the top 5 products brought 24.4% of revenue in the weakest quarter of stores by ROAS and 14.3% in the strongest quarter. The fewer products carry the store, the more one stockout or one lost supplier can cost in a month.

Google’s API guide to Shopping reporting keeps two product reports apart: one holds each product’s performance history, the other its current state, including any issues that stop a campaign from using it. A bestseller needs both checks: sort products by conversion value, then make sure the ones at the top can still appear in ads.

A product can also drop out quietly: it sells out, leaves the feed or fails review, and you notice the loss only later, in your sales. The portal’s Products section keeps a separate list of products that used to sell consistently or pay off and are not showing now. For each one it shows the reason, how many days it has been out and how much revenue that costs per month.

Three articles cover this stage:

• How to find out within days that a bestseller has lost impressions or sales: bestseller alerts.
• What a sold-out bestseller costs beyond the missed orders: the cost of a bestseller stockout.
• Whether to keep, pause or exclude products while they’re unavailable in Shopping and Performance Max: handling out-of-stock products.

Eight product groups and what to do with each

A label turns a product’s history into a decision. The portal gives every product one of eight labels, based on how it behaved in ads over 12 months, and recomputes them with every data refresh. The 12-month window matters: in a single month, most of the products that sell are tail products, and three in four of them were not selling the month before (GetProfit data, 116 stores, June 2025 – June 2026).

“Profitable” here means ROAS at or above your target, and ROAS counts revenue, not profit. A label describes a product’s role, not its quality: the same product can land in different groups in different seasons.

Labels are one way to group products. If you already use classic inventory methods, ABC analysis for an online store shows how to run them on sales and ad spend rather than stock value. Add how steadily each product sells, and you get ABC-XYZ analysis, which separates steady sellers from one-off spikes.

How to turn product groups into campaigns

Groups change results only when campaigns treat them differently. Merchant Center has five fields for this: custom labels 0 to 4. Google’s Merchant API reference (ProductAttributes) describes them as fields for custom grouping of items.

For Performance Max, Google’s API documentation (Listing Groups for Retail) advises targeting groups of products by dimensions such as custom labels or brand rather than single products. There, the groups decide which products an asset group includes. Unlike in Standard Shopping, they don’t set bids.

With the groups in the feed, each one can go to a campaign with its own budget and its own target ROAS. Split only where the data can carry it. The portal recommends splitting a campaign only when both parts get at least 30 conversions a month, so that each part has enough conversions to learn from. The product segmentation guide shows how to choose the segments and check that they really move budget.

What an account average hides: an example

Example store, not client data.

A tableware shop with 3,000 products spends 40,000 a month on ads and gets 180,000 in revenue from 300 orders, so its ROAS is 4.5. Only 1,200 products got an impression this month. Here is the same month, broken down by product:

Read product by product, the month shows three things. First, 35% of spend (14,000) brought no sale. Second, the 135 products that sold returned 180,000 on 26,000 of spend, a ROAS of 6.9. Third, the 105 tail products brought 40% of revenue.

This table hides two traps. One month is too short to judge the 1,065 products without a sale: in our study, 35.4% of products that had spent a conversion’s worth without selling converted later. And this month’s 35% measures something different from the 40.5% in our study, which counts products that never sold in 13 months. Compare only windows of the same length.

Product performance analysis: a monthly review in eight steps

1. Export 12 months of product data from the product reports in Google Ads, or from shopping_performance_view in the API. Mark 15 June 2026 on the timeline: from that date, product data for Performance Max covers all its networks. Check that the export covers every product you advertise, including the ones without sales.
2. Sort products into groups by behaviour: no impressions, spend without a sale, sells below target, sells rarely at target, sells steadily at target. Check that every product sits in exactly one group.
3. Work out each group’s share of spend and of revenue. Check that the shares add up to 100% in both columns.
4. Start with the bestsellers. Look for any that can no longer appear in ads, have lost impressions or have sold out since the last review, and fix those first.
5. Judge the long tail as a group, by the group’s spend and revenue. Product-by-product verdicts on the tail go stale within a month.
6. Check products that spend without selling for price, listing and category before you cut anything. Lower their share of the budget rather than dropping it to zero.
7. Give newcomers a real test. In our study, the minimum honest test was 15–20 clicks.
8. Write the groups into custom labels so campaigns can treat them differently, and repeat the review every month. With a steady bestseller lasting a median 3 months in our study, a quarterly review is already late.

If you manage the account for someone else, agree every change with the owner before you apply it.

Which products bring money in and which spend it. Every product gets its own label based on how it behaves in ads — from bestsellers to dormant ones that spend budget without a single sale. The portal changes nothing without your consent.

Label your catalogue →

Sources

• Reporting — campaign types covered by product reporting; how product impressions, clicks, cost and conversions are counted; the Product clicks column; separate reports for performance history and current product state. Checked 2 October 2026.
• shopping_performance_view — product-level statistics for Shopping and Performance Max in the API; attributes recorded as of the event date; totals differ from campaign reports. Checked 2 October 2026.
• Retail Campaign Performance — from 15 June 2026, product data covers all Performance Max networks, with a possible one-time increase in metrics. Checked 2 October 2026.
• ProductAttributes — five custom labels, 0 to 4, for custom grouping of items. Checked 2 October 2026.
• Listing Groups for Retail — in Performance Max, target product groups by custom labels or brand; listing groups include or exclude products and don’t set bids. Checked 2 October 2026.
• The Long Tail — Chris Anderson, Wired, October 2004: where the long tail idea was set out. Checked 2 October 2026.
• GetProfit data: study of 1,404,808 products, 130+ stores, 13 months — catalogue layers, spend concentration, bestseller lifespan, switch-off accuracy, products that started with no spend, top-5 revenue share, minimum test.
• GetProfit data: 116 stores, June 2025 – June 2026 — share of selling products and revenue from products with 1–2 orders a month; month-to-month renewal of the tail and the head.
• GetProfit portal methodology — eight labels from 12 months of behaviour; campaign split at 30+ conversions a month on each side; exclusion after 50+ clicks and no conversion in 360 days.