Google's AI Share of Voice in Merchant Center: what it measures and why it matters now

Google added AI Share of Voice to Merchant Center as part of the GML 2026 announcements. Most brands have registered the headline and moved on. The practical implication is worth sitting with, because this metric is measuring something none of the current standard analytics stacks can track, and the gap between what it shows and what brands currently measure is going to widen as AI-assisted discovery becomes a larger part of the purchase funnel.

What AI Share of Voice actually measures

The metric tracks how often your brand appears in Gemini's AI-generated product recommendations. When a user asks Gemini about a product category and Gemini surfaces a list of recommended items, AI Share of Voice captures your presence in that shortlist, across categories, over time.

This is structurally different from auction-based impression share. Traditional paid impression share tells you how often your ads appeared when a user searched a relevant query in Google's auction system. AI Share of Voice tells you how often your products were surfaced by a model's shortlisting logic, before any auction took place, and in many cases before the user had reached a search results page.

The commercial significance: if Gemini is recommending products in your category and your brand is not in those recommendations, you are invisible at the decision point. The user forms a consideration set, makes a choice, and arrives at a competitor's site having never encountered your brand. Your conversion data never records this absence because the user never arrived. The absence is invisible to your current measurement infrastructure.

The measurement gap in your current analytics stack

Your analytics setup almost certainly tracks clicks, sessions, conversions, and revenue. It may track assisted conversions, view-through attribution, and cross-channel path analysis. None of these capture a user who was presented with AI-generated product recommendations, did not see your brand listed, and converted elsewhere.

This is not a minor edge case. It is the difference between knowing your paid search is converting at a given rate and understanding why you are losing a portion of the decision-making process before paid search even becomes relevant. The conversion data is accurate as far as it goes. It represents a shrinking share of the actual decision funnel if AI-assisted discovery is becoming a meaningful pre-filter on which brands users even consider.

The analogy to organic search is useful here. For twenty years, brands tracked organic rankings and worked to appear on the first page of search results. AI Share of Voice is the equivalent signal for AI-generated shortlists: a measure of your visibility in a different kind of result, one that increasingly sits at the top of the consideration funnel before traditional search rankings apply. See GA4 channel groupings: what changed and why it matters for the broader pattern: new platform behaviours require new measurement disciplines, not just new columns in an existing report.

Who actually needs to care about this metric

The natural assumption is that AI Share of Voice belongs to the paid media team alongside impression share and auction metrics. This assumption will result in the right people never seeing the data, and the data never being acted on.

The factors that drive AI Share of Voice are different from the factors that drive auction performance:

• Product feed completeness and accuracy. Gemini's product recommendations draw on structured data. Brands with clean, complete, well-categorised feeds are more likely to be shortlisted. This is a product data and operations problem, not a bid management problem.
• Structured data quality on the product pages themselves. Schema markup, accurate pricing, detailed product attributes, and review aggregation all factor into how well the model can characterise your products against a user's query. This is a technical SEO and development problem.
• Product review volume and recency. The model uses review signals to assess product credibility. A product with 200 recent reviews is a safer recommendation than one with 12 reviews from two years ago. This is a customer experience and retention problem.
• Content in crawlable sources. Editorial coverage, published comparisons, and authoritative category content that mentions the brand all contribute to the signals the model uses when building its shortlists. This is a content and PR problem.

None of these are bid optimisation levers. The paid media team cannot improve AI Share of Voice by adjusting a target CPA or increasing a bid modifier. The teams that need to see this data and have the mandate to act on it are product, content, technical SEO, and PR. Most brands do not have a workflow for getting this metric to those teams, or a clear owner for the improvement work.

What drives AI Share of Voice versus auction impression share

The operational difference between the two metrics is worth naming precisely, because it affects the management cadence and the budget implications.

Auction impression share responds to bids, budgets, quality scores, and audience signals. If you want more auction impression share, the primary lever is financial: bid higher, budget more, improve quality score through better landing pages and ad relevance. The feedback loop is fast. Changes in bids produce changes in impression share within days.

AI Share of Voice responds to data quality, content, reviews, and structural signals. The feedback loop is slow. Improving your product feed completeness does not produce a measurable shift in AI Share of Voice the following week. It produces a directional shift over months as the model re-evaluates your product catalogue against updated signals. This means the brands that invest in these capabilities now are building an advantage that compounds over time and is difficult to close quickly once the gap is established.

This is also why AI Share of Voice is a strategic signal rather than a tactical one. Checking it weekly and making adjustments is not the right management cadence. Reviewing it quarterly, tracking directional trends, and connecting those trends to product data and content investments is the right use of the metric.

What to actually do

If you have access to AI Share of Voice data in Merchant Center, the first-order priority is establishing a baseline:

• Pull your current AI Share of Voice for your top five product categories. This is the number you are starting from. Record it with a date. Without a baseline, all subsequent numbers are uninterpretable.
• Pull the same data for your primary competitors where available. The gap between your share and their share in a given category is the clearest early signal of where AI-assisted discovery is already costing you, before it shows up in conversion metrics.
• Audit your product feed for completeness in the categories where your share is lowest. Are all required fields populated? Are descriptions detailed enough to differentiate products from generic alternatives? Are attributes correctly categorised so the model can match them to query intent? Feed gaps are usually the lowest-effort, highest-impact lever for improving AI visibility.
• Review structured data on your product pages. Use Google's Rich Results Test or a comparable tool. Confirm that schema markup is correct, prices are current, and review aggregation is firing correctly. These signals are the connective tissue between your product catalogue and Gemini's recommendation logic.
• Set a quarterly review cadence for this metric. AI Share of Voice will not change fast enough to warrant weekly attention. A quarterly review where you track directional movement, connect it to the data investments you have made, and assess competitive gaps is a sustainable discipline that reflects the speed at which the underlying signals actually change.
• Route the data to the teams who can act on it. Product, content, and technical SEO need to see AI Share of Voice data. If it sits only with the paid media team, it becomes an interesting number without an owner for improvement. The conversation with those teams is about feed completeness and structured data health, not auction strategy.

The competitive dynamic

Most brands will check this metric once after reading an article about it and not build a cadence around it. The ones that do build that cadence are making a specific bet: that AI-assisted discovery will become a meaningful share of the purchase funnel across product categories, and that early visibility in AI-generated shortlists will be harder to displace once consideration-set patterns are established.

The parallel to organic search investment is instructive, even if not perfect. The brands that built organic authority in 2005 did not have to fight for it in 2015. The brands that waited until organic search was obviously important found themselves trying to close a gap that the early movers had spent years building through content, link equity, and structural improvements. The mechanism is different for AI Share of Voice, the signals are feed quality and review volume rather than backlinks, but the compounding dynamic is similar.

The metric is newly measurable. The underlying behaviour it tracks, AI systems shortlisting products before users reach a traditional search interface, is already happening at scale. Getting a baseline now and identifying which teams own the improvement work is the minimum responsible response before the gap becomes visible in conversion data.

If you want help understanding your current AI Share of Voice and what your product data and content gaps look like relative to your category, book a free audit. We will look at your Merchant Center data, your feed completeness, and your structured data health, and give you a prioritised view of what to improve first.