AI Visibility: How to Measure Your Brand's Share of Voice in ChatGPT, Gemini and Perplexity

The dashboard gap

Marketing teams have rank trackers, search consoles, and attribution models for classic search. Then a channel growing roughly 4,700% year over year in retail traffic, by Adobe's measurement, shows up, and most teams have nothing. No baseline, no trend line, not even a habit of checking.

The result is that brands discover their AI problem anecdotally. A founder asks ChatGPT for "best protein powder for beginners," sees three competitors, and panics. That is a useful wake-up call, but one chat is not measurement. AI answers vary by phrasing, session, and day. To manage this channel you need the same rigor you apply to SEO, adapted to how AI actually works.

The three metrics that matter

1. AI referral sessions

The simplest signal: how much traffic ChatGPT, Gemini, Perplexity, and AI Mode actually send you. Most analytics setups can isolate these referrers today, and the trend tells you whether the channel is growing for you specifically. Shoppers arriving from AI conversations tend to land deep, already comparing or ready to buy, so even modest volumes can carry outsized revenue.

2. Share of voice per query

This is the heart of it. For each buying question that matters in your category, which products surface, in what order, and how often is one of them yours? Across a category's worth of queries, that rolls up into a single number: the percentage of relevant AI answers your brand appears in, next to the same number for each competitor.

Done properly, the measurement mirrors how assistants think. A shopper's question fans out into many sub-queries, price-bounded, attribute-bounded, use-case-bounded, so the tracking has to run that fan-out too, repeatedly, across each surface. Weekly runs at scale turn anecdotes into a trend you can act on.

3. Buy-link attribution

The metric nobody watches and everybody should. When an AI recommends your product, where does the buy button point? Your store, where you keep margin and the customer? Or a marketplace listing, where you keep neither? You can win the recommendation and still lose the economics. Track the destination of every buy link in every answer that includes you.

From measurement to diagnosis

A share-of-voice number on its own just tells you the score. The useful question is why you lose specific queries, and the answer is usually findable. Put your product card next to the competitor that beat you and diff them:

• A missing attribute: they list fabric weight, certification, or battery hours and you do not.
• A missing use case: their data says "for sensitive skin" or "for travel" and yours says nothing.
• A missing context line: their attributes explain who they are for; yours are bare specs.
• A missing review signal: they show aggregated proof; you show silence.

Run that diff per SKU and the abstract "we need better AI visibility" becomes a concrete work list. The fix side of this is covered in feed optimization for AI search and why ChatGPT doesn't recommend your products.

There is also a moving-target problem. AI platforms ship new feed fields, flags, and schema requirements continuously. A catalog that was complete in March can be quietly missing required fields by May. Platform gap tracking, knowing what each surface newly expects and whether you provide it, belongs in the same dashboard as share of voice.

Doing this manually vs. automatically

You can start by hand, and you should: ten queries, three surfaces, a spreadsheet, thirty minutes a week. That alone puts you ahead of most of your category.

What hand-tracking cannot do is scale to the fan-out. Real coverage of a category means hundreds or thousands of query variants per week, per surface, with positions and buy-link destinations recorded. That is machine work. Ziffi's Agentic Commerce Optimization Suite runs exactly this loop: a live dashboard of AI referral sessions, per-query share of voice across ChatGPT, Gemini, Perplexity, and AI Mode, and buy-link attribution, with the per-SKU gap diffs attached, so the dashboard tells you what to fix, not just what you lost. The same system then closes the gaps it finds, as described in the AEO playbook.

A starter scorecard

Whatever tooling you choose, start the baseline now. Share of voice compounds: the brands answering AI's questions today are collecting the reviews, signals, and sales that make them even harder to displace tomorrow.