How to Measure Your Share of Voice in AI Answers

Short version: Share of voice (SoV) in AI answers is the one number that tells you whether AI assistants name your brand when a buyer asks "who's a good supplier?" — the moment that now happens before Google. Most brands can't quote that number because nobody measures it. This is the reproducible way to do it: build the prompt set, pick the engines, decide what counts as a mention, name the competitors, and turn it into a weekly trend line. One real result at the end — a Vietnamese distributor at 45.0% SoV — as the worked example.

#Why share of voice is the number, not traffic

The buying journey now starts one step earlier than your website. A clinic's purchasing lead, a DTC shopper, a B2B buyer — they ask ChatGPT or Gemini to shortlist suppliers before they open a single tab. The brands the assistant names make the next round. The ones it doesn't are gone before they knew there was a race.

Traffic and rankings measure the old first step. Share of voice measures the new one: across all the times an assistant answers a category buying question, how often is it your name in the answer versus your competitors'? You can't move a number you don't measure — so measure this one first.

We ran exactly this for a B2B dental-materials distributor; the result is the last section. But the method is the part you can reuse, so here it is first.

#Step 1 — Build a prompt set that mirrors real buying intent

Write 20–30 category prompts a real buyer would type before they know who to trust: "where to buy reputable [category] in [city]," "authorized [category] distributor in [region]," "top [category] suppliers in [country]," plus a few comparison prompts. One rule decides whether the whole exercise is honest: if your brand name appears in the prompt, delete it. You're measuring discovery, not ego — a prompt that already names you can't tell you whether the assistant would have found you. Then freeze the set. The trend only means something if week 5 asks the same questions as week 1.

#Step 2 — Pick the engines your buyers actually use

Run every prompt on the assistants your buyers live in — for most B2B and DTC categories that's ChatGPT and Gemini; add Perplexity or Claude if that's where your audience asks. Run each prompt on each engine, and log which engine produced which answer. Mixing engines without labeling them is the single most common way this number quietly lies — one engine can love you and another ignore you, and a blended average hides both.

#Step 3 — Decide what counts: mention vs citation

Two different signals, scored separately:

• Mention — your brand is named in the answer text.
• Citation — your domain is linked or sourced.

Write the scoring rule down before you start, so week 5 is scored exactly like week 1. Vague phrasing ("a store like yours," "several local suppliers") counts as nothing — only a named brand counts. This is the boring discipline that makes the number survive someone poking at it later.

#Step 4 — Name the competitor set

Share of voice is comparative. A number with nothing beside it is a testimonial, not a measurement. List the 5–8 rivals the assistants actually name in your category — you don't have to guess, the answers tell you who they are. Every mention of every competitor goes into the same log as yours. Now your number has a context: not "we got mentioned a lot," but "we got mentioned more than the next three combined."

#Step 5 — Compute it

Share of voice = your mentions ÷ total mentions (you + all competitors), across the full prompt × engine set, in one tracked week.

Don't stop at one percentage — it hides too much. Track three numbers together:

1. Share of voice — your slice of all brand mentions.
2. Mention rate — the % of answers that name you at all.
3. By-engine split — your SoV on each engine separately.

A brand can have a healthy overall SoV while being invisible on one engine. Three numbers show you that; one hides it.

#Step 6 — Run it weekly, on the same prompts

One week is luck. A line is a position. Same frozen prompts, same engines, every week, logged so any number opens back to the exact prompt, engine and week behind it. That auditability is the entire point: a share of voice you can click into and verify is a measurement; one you can't is a screenshot. Put it on a dashboard — even a spreadsheet — so the trend, not a single good day, is what you report.

#The four ways this number lies

Before you trust any SoV figure — yours or an agency's — check it against these:

1. Brand-name prompts inflate it. If the prompt already says your name, you measured recall, not discovery.
2. A one-off screenshot instead of a trend. One week proves nothing; ask for the line.
3. No competitor set. A percentage with nothing to compare against is meaningless.
4. Engines blended without labels. A single average can hide being absent on an entire engine.

#What a real result looks like

Here's the worked example — a B2B dental-materials distributor in Vietnam (Oroca), measured on this exact method, week 37 of 2026:

Reading it the way the method demands: 45.0% share of voice (49 mentions) — number one outright, more than the next three combined; 82% mention rate across the tracked prompts; by engine, ChatGPT 76.7% and Gemini 86.7% — present on both; and a rising line, not a spike — 0 at launch → 41.2% (W36) → 45.0% (W37), same 30 prompts each week. Full context and the dashboard behind it: 0 to 45% Share of Voice in AI Answers — the Oroca case.

We're not promising you 45%. Oroca's category, starting point and competitive field are its own. What travels is the method: measure first, name the competitors, score it honestly, track it weekly — and only then decide what to fix.

Want to know where your store stands in AI answers right now? Send one store URL and we'll run a free async AI-visibility read — we ask live AI assistants the category questions your buyers ask and send back where you actually appear versus your competitors, mention by mention: luma-e.com/ai-readiness. No call required.