Recommendation Share: How Much of the AI Shortlist Is Your Brand Winning?
Recommendation Share: How Much of the AI Shortlist Is Your Brand Winning?
Learn how recommendation share measures your brand’s presence in AI shopping shortlists, how to calculate it, and how ecommerce teams can track performance.
Recommendation share measures how much of a defined AI product shortlist is occupied by your products across the shopping questions, engines, markets and runs you have chosen to monitor. It answers a narrower and more useful question than general AI visibility: when a shopper asks for products that your catalog can genuinely compete for, how much of the recommendation space are you actually winning?
The formula is simple. The measurement design is not.
If the shopper-question set changes, branded questions are added, shortlist depth changes, variants are counted differently or one AI engine receives more weight than another, recommendation share can move even when the brand's underlying competitive position has not.
That makes the denominator the most important part of the metric.
What does recommendation share actually measure?
For a controlled measurement set:
Recommendation share = brand recommendation occurrences ÷ all qualifying recommendation occurrences
A recommendation occurrence is one distinct product from the brand appearing inside the qualifying shortlist for one tracked shopper question, on one AI engine, during one run.
Consider an illustrative measurement period. Across the monitored questions, engines and runs, the qualifying shortlists contain 420 product recommendation occurrences. Products from the brand occupy 63 of them.
Recommendation share = 63 ÷ 420 = 15%
That 15% does not mean the brand has 15% market share, 15% of category demand or 15% of AI-driven revenue. It means the brand occupied 15% of the product recommendation positions captured under that measurement definition.
It is also different from Google's AI share of voice metric. Google Merchant Center defines AI share of voice using your brand or product impressions relative to the impressions generated by you and your defined competitors. Recommendation share, as defined here, is narrower: it measures qualifying product recommendations inside a controlled shortlist rather than all AI impressions.
That distinction matters because appearing somewhere in an AI shopping experience and earning a recommendation position are not the same commercial event.
How should the denominator be constructed?
The denominator should represent the market you actually intend to measure.
Before calculating the first percentage, freeze the category, market, shopper-question universe, AI engines, run cadence, shortlist depth and product-identity rules. The shopper-question map should already establish the situations where the brand has at least one product with a legitimate right to compete.
If a running-shoe brand only sells performance footwear, a question about formal leather shoes should not enter its recommendation-share universe. If the catalog has no product below $100, a strict "best running shoe under $80" question should not quietly become a recommendation failure.
Branded and unbranded questions should also remain separate. A company can look extremely strong when the shopper has already named the brand and considerably weaker when the AI system has to decide which brands deserve to enter the shortlist.
Mixing the two can make the headline percentage improve without making the brand more competitive in discovery.
Engines should also remain identifiable underneath the roll-up. OpenAI says ChatGPT selects shopping products based on the shopper's query and context, alongside product metadata and other available information. Microsoft similarly says Copilot product options can depend on the shopper's prompt, likely engagement, available merchant data and other factors. Different AI environments can therefore produce different competitive sets.
The answer is not to invent one universal denominator. It is to define the denominator clearly enough that the percentage can be reproduced and compared.
Which recommendation occurrences should count?
A recommendation-share system needs a clear annotation rule before monitoring begins.
Count products that the AI experience actually presents as suitable options for the shopper's request. Do not count a product simply because its brand appears in a citation, a background explanation, a review summary or an incidental comparison.
Shortlist depth should also be fixed in advance. If the method measures the first five qualifying product recommendations, keep that rule consistent across reporting periods. If an answer contains fewer qualifying products, record what actually appears rather than creating artificial recommendations to fill the shortlist.
Product identity needs the same discipline.
A black shoe in size 9 and the same shoe in size 10 should not automatically create two competitive wins if the business reports recommendation performance at parent-product level. If variants are commercially distinct enough to count separately, lock that rule before measurement and apply it consistently.
This matters because AI shopping systems can operate on detailed product and variant records. OpenAI's commerce specification supports stable product and variant identifiers alongside fields such as price, availability, seller and variant options. Its file-feed specification likewise asks merchants to provide separate purchasable items or variants with stable IDs, current availability, price and seller information.
Google's Merchant Center specification makes the same need for product-level accuracy visible: incorrect or missing product information can limit product eligibility or cause incorrect displays.
The reporting unit should therefore match the product level at which the ecommerce team can actually interpret and act on the result.
How should recommendation share be segmented?
One company-wide percentage is rarely enough.
At minimum, recommendation share should remain segmentable by market, AI engine, category or product family, branded versus unbranded question, shopper stage and priority product group.
Imagine a brand with an overall recommendation share of 18%.
Segment | Recommendation share | What it suggests |
US unbranded discovery | 7% | Weak competitive entry into new consideration |
US branded evaluation | 62% | Strong once shoppers already know the brand |
UK unbranded discovery | 3% | Larger market-specific weakness |
The overall result is mathematically correct. It is also operationally incomplete.
The segments reveal whether the problem is entering new consideration, defending an existing brand preference or competing effectively in a particular market.
Google's current AI performance reporting applies a similar principle to its own metric. Merchant Center allows performance to be sliced by product category, country and time period, while also separating discovery, evaluation and ready-to-buy shopping stages.
Recommendation share should also be read alongside SKU coverage, but the two measures should remain separate.
Recommendation share tells you how much shortlist space the brand occupies. SKU coverage tells you how much of the relevant catalog participates at all.
A brand can gain recommendation share because three hero products win more often while most of the priority catalog remains absent. Another brand can increase SKU coverage without taking much additional shortlist space.
Those are different improvements and should trigger different investigations.
How should teams interpret changes over time?
A movement in recommendation share is meaningful only when the test remains comparable.
Suppose a brand moves from 12% to 18%. That looks like a six-point improvement.
But if the second reporting period contains more branded questions, fewer competitors, a different engine mix, a different product-identity rule or a shallower shortlist, part of that movement may be methodological rather than competitive.
Version the measurement whenever a material input changes. Keep the previous series intact and record the methodology change instead of joining two incompatible tests into one performance line.
Platform change adds another reason for that discipline. OpenAI says its merchant-ranking approach is expected to evolve as its shopping experience develops, while Google's AI performance report is updated regularly and provides historical data with a short reporting lag.
Repeated runs can help reduce the risk of overreacting to one AI response, but repeated runs cannot repair a weak denominator.
A useful review therefore asks two questions together:
Did recommendation share move?
Did the measurement definition remain materially the same?
If the answer to the second question is no, the change should not be interpreted as pure performance movement.
What can recommendation share tell you, and what can it not tell you?
Recommendation share tells you whether your products are occupying more or less of a defined AI recommendation space. It is particularly useful because it observes competitive product consideration before the shopper necessarily becomes website traffic.
It does not explain why a product won or lost. That requires product, attribute, source, trust and competitor diagnostics underneath the metric.
It does not tell you whether the shopper was routed to your own store. OpenAI says that after a product is selected, ChatGPT can show multiple merchants and rank them using factors including availability, price, merchant quality and whether the seller is the maker or primary seller. Merchant routing is therefore a separate commercial outcome.
It does not prove market demand. A strong recommendation share on a commercially unimportant shopper need may matter less than a weaker share on a much larger opportunity.
And it does not prove traffic, revenue or causal business impact.
Those boundaries make the metric stronger, not weaker. Recommendation share should answer one question well:
When AI systems assemble relevant product shortlists, how much of that recommendation space are our products winning?
At Gravton, that measurement sits inside a broader commerce view. The platform is designed to show which SKUs win across AI shopping journeys, where the purchase is routed and what changed over time. The recommendation-share number is the competitive headline; the product-level evidence underneath it is what makes the movement actionable.
What else should ecommerce teams know?
Is recommendation share the same as share of voice?
No. The names sound similar, but the measurement jobs differ. Google Merchant Center's AI share of voice is based on AI impressions relative to a defined competitor set. Recommendation share in this methodology counts qualifying product recommendation positions inside a controlled monitoring universe.
Should branded questions count?
Yes, when branded evaluation is commercially relevant, but report branded and unbranded performance separately. Otherwise the headline metric can improve simply because more of the measurement set starts with shoppers already naming the brand.
Should every AI engine be weighted equally?
Not automatically. Equal weighting can be useful for a controlled cross-engine view. If the business has a defensible reason to weight engines differently, lock that rule before the reporting period begins. Do not change the weights after seeing the result.
How often should recommendation share be reviewed?
Use a cadence that matches the stability of the monitored set and how quickly the business can act. Monthly is often useful for leadership, while ecommerce and category teams may inspect product or segment movement more frequently.
What should you read next?
Prompt Market Analysis: What Buyers Actually Ask AI explains how to define the shopper-question universe before recommendation measurement begins.
How to Measure ROI from GEO covers the downstream measurement boundary once AI activity becomes observable traffic and revenue.
The Hidden Revenue Loss from Missing the AI Shortlist explores why missing recommendation space can matter commercially without treating every absence as lost revenue.
Which sources support this guide?
Want to see how much of the AI shortlist your products are winning?
Gravton measures product-level AI-shopping performance across markets and engines, then keeps the evidence underneath the headline result so ecommerce teams can see where recommendation performance changed and what deserves investigation next.
VISIBILITY & CONTENT STRATEGY