Merchant Center AI Performance Insights: Read Your AI Share of Voice
What actually shipped at GML 2026
At Google Marketing Live 2026, Merchant Center gained a feature called AI performance insights. It shows you how your brand performs on AI surfaces, then compares your share of voice against a set of similar brands. For the first time, the question “am I even showing up in AI answers” has a number attached to it.
The timing is not random. Product research has shifted hard. Traditional Google search accounted for roughly 89% of product research in 2023; by 2026 that is down to about 67%. The missing third now flows through ChatGPT, Perplexity, Copilot, and Google’s own AI Overviews. If you are absent from those surfaces, you are losing a third of intent you previously could not even measure.
Here is the number that should worry feed managers most. In mid-2025, the top-10 organic rankers captured 76% of AI Overview citations. By early 2026 that dropped to about 38%. The link between your Google ranking and whether AI cites you is weakening fast. Inferring your AI-surface presence from SEO rankings no longer holds, which is exactly the blind spot AI performance insights is built to fix.
The value here is the first reliable read on whether your products earn a voice inside AI answers, with a peer benchmark so the number means something.
How to read the share of voice number
Open Merchant Center, go to AI performance insights, and you will see your share of voice alongside a benchmark for comparable brands. Do not act on the headline figure alone. Break it down three ways first.
Read the absolute level. If your share of voice sits well under the peer benchmark, AI surfaces rarely mention you when answering relevant shopping questions. The cause is usually incomplete feed data or product descriptions that do not match how people phrase queries conversationally.
Read the trend. A static number tells you little. What matters is direction. If you edited your feed last week and share of voice climbed, the change worked. If it is flat or sliding, check whether competitors are crowding you out.
Read the category split. Share of voice varies widely across categories in the same account. You might do fine on “wireless earbuds” and be nearly silent on “bluetooth speakers.” That gap is your roadmap for what to fix next.
| Metric you see | What it tells you | Action to take |
|---|---|---|
| Share of voice under peer benchmark | AI answers rarely cite you | Prioritize feed completeness and richer descriptions |
| Falling share in one category | Competitors are pushing you out of AI answers | Rewrite those products with conversational attributes |
| Share rising after a feed edit | Your direction is correct | Replicate the change across other categories |
| Flat but low overall | No hard data gaps, just not conversational enough | Fix description tone, not missing fields |
Remember that share of voice is relative. A rise can mean a rival weakened rather than you improving, and a drop can mean a competitor pushed hard rather than you slipping. Always pair the number with trend and category before making a call.
Turn the data into a fix queue
Once you can read it, the next move is triage. Do not refurbish the whole catalog at once. Rank work by share-of-voice gap, largest first.
Subtract your share from the peer benchmark per category and pull the widest gaps to the top. These are your highest-leverage fixes: rivals are visible in AI answers and you are absent, so a small improvement moves the needle a lot. In your Shopify feed, check whether those products have GTIN, brand, material, size, and use-case fields populated. Those structured attributes are exactly what AI assembles answers from.
Next, look at categories that have a voice but are declining. These usually do not lack fields; their copy lags how people ask conversationally. A shopper asks for “a cleanser for sensitive skin,” but your description still reads “deep-cleansing foam.” The AI cannot connect the two.
Only then touch categories where you already lead. Maintain them and move on. Time spent on a wide gap returns far more than polishing a category that is already winning.
Run this on a two-to-four-week loop: edit a batch of feed entries, let the data refresh, check whether share of voice moved, then decide the next batch. The insights update on a rolling basis, so treat it as a feedback loop rather than a one-time audit.
Conversational attributes: write copy AI can use
Knowing where you lack voice only gets you halfway. You also need a way to add it, and Google shipped conversational attributes at the same event for exactly that.
Traditional feed descriptions are written for keyword matching: “waterproof bluetooth 5.3 battery 30 hours.” But shoppers ask AI surfaces in full sentences, such as “are there earbuds for running that handle sweat and last a half marathon?” Conversational attributes let you structure product information to answer questions like that.
In practice, start with the products in your widest share-of-voice gaps. Rewrite descriptions from spec lists into “scenario plus problem solved.” Sticking with earbuds: instead of “IPX5 rated,” write that they handle workout sweat and light rain without water getting in. Same fact, but the second phrasing is what an AI is more likely to lift into an answer.
After editing, go back to AI performance insights and watch the relevant category. If share of voice rises within two weeks, the approach holds and you can roll it out at scale. If nothing moves, you likely still have field gaps, so return to the structured data.
Ask Advisor: a collaborator inside Merchant Center
Last, Ask Advisor. This is the agentic collaborator Google is building into retailer tools, and it lives inside Merchant Center as something you can simply ask.
It pairs naturally with AI performance insights. When you see a category lose share of voice, you do not have to comb the feed line by line for the cause. Ask Advisor directly: “why is my bluetooth speaker category share of voice so low?” It reads your feed data and performance together and returns a diagnosis, such as which products miss key fields or which descriptions lean too keyword-heavy.
Keep the right framing when you use it. It offers recommendations, not orders. Ask Advisor cannot see your inventory strategy, margins, or brand positioning, and only you can. If it suggests adding a material field to a batch of products, that is almost certainly right, so do it. If it suggests repositioning a hero product, weigh that against your wider plan first.
Chain the three together and the workflow clicks: AI performance insights tells you where you lack voice, Ask Advisor diagnoses why, and conversational attributes are the tool you use to fix it. See, ask, edit. Close each loop by returning to the share-of-voice data to confirm the change worked, and the AI surface that used to be invisible becomes something you can monitor and tune on a routine.
Related Articles
Perplexity Buy with Pro Economics: Does the AI Channel Actually Make Money
Perplexity's Buy Now agent keeps adding shoppers, but it reportedly takes an 8-12% GMV commission on top of Stripe processing. This is not a listing guide — it is the margin math: at what AOV and gross-margin structure the channel pencils out, and when you are just working for the platform.
Google Images Turns 25 With a Pinterest-Style Feed: What E-Commerce Sellers Need to Do Now
Google Images just got its biggest redesign in 25 years, replacing the empty search box with a personalized, continuously updated visual discovery feed. For e-commerce sellers, product image exposure shifts from query-driven to feed-driven. This article covers what changed, a practical image SEO checklist, and why the new Collections feature is a purchase intent signal worth paying attention to.