Measuring Share of Voice in AI Answers
Share of voice in an LLM context is not the same as a brand-tracker metric. Here is how to define a prompt set, sample correctly, and track it week over week.
Summary — Share of voice in an LLM context is the percentage of brand-relevant prompts in which your brand is mentioned, divided by the total mentions across all competitors. To measure it correctly you need: a defined prompt set, multi-engine coverage, a stable competitor list, and a recurring sampling schedule. One-off audits don't count.
For the broader concept, see What is AEO?.
What "share of voice" actually means here
In classic brand tracking, share of voice (SOV) measures how much your brand appears in media relative to competitors. The AEO version is the same idea, applied to AI answers:
Across a defined set of buyer-relevant prompts run against AI engines, what percentage of all brand mentions are yours vs. competitors'?
There are three measurement variants you'll see used:
| Variant | Definition |
|---|---|
| Mention SOV | Your mentions / total mentions across all brands |
| Citation SOV | Your URL citations / total URL citations |
| First-mention SOV | Times you're the first brand named / total prompts |
Mention SOV is the most common headline metric. Citation SOV is more rigorous (it requires the engine to actually cite a URL). First-mention SOV is the most prestige-laden — it's a leading indicator of brand-of-default status.
Step 1: Build your prompt set
This is the foundational work, and the place teams cut corners. A good prompt set is:
- Buyer-relevant. What real prospects type, not what you wish they typed.
- Mixed by stage. Awareness ("what is X?"), consideration ("X vs Y"), decision ("best X for Y use case").
- Mixed by intent. Branded, comparative, category, problem-led.
- Sized for signal. Below ~50 prompts you're in noise; 200–500 is a healthy band.
A starter mix for a B2B SaaS:
| Bucket | Examples | Approx. share |
|---|---|---|
| Category / awareness | "what is AEO?", "how do I rank in ChatGPT?" | 30% |
| Comparison | "X vs Y", "best X for SaaS" | 30% |
| Use-case / decision | "best X for content marketing teams" | 20% |
| Problem-led | "why isn't my brand showing up in ChatGPT?" | 15% |
| Branded | "is X any good?", "X reviews" | 5% |
For B2B-specific guidance, see AEO for B2B SaaS.
Step 2: Pick your engines
In 2026, the four engines that move the needle:
- ChatGPT (web + browsing)
- Perplexity
- Google AI Overviews
- Claude
Don't measure only one. ChatGPT-only SOV is misleading; the same brand often dominates ChatGPT and underperforms in Perplexity.
Step 3: Lock the competitor list
Define 5–10 competitors up front. This list should:
- Include your obvious head-to-head competitors
- Include 1–2 emerging upstarts
- Include any "alternative category" you might be compared to
- Stay stable for at least a quarter so you can compare apples to apples
Without a fixed list, every change in the answer pool changes your SOV denominator.
Step 4: Sample on a schedule
LLM answers vary across runs. To get a defensible signal:
- Run each prompt at least 3 times per engine per measurement period
- Aggregate (e.g. "mentioned in ≥2 of 3 runs")
- Re-run weekly at a minimum, daily for high-stakes categories
This is exactly the work that AEO platforms automate. See the best AEO tools.
Step 5: Compute the metrics
For each measurement period:
mention_count(brand) = number of prompts where brand was mentioned (in ≥2 of 3 runs)
total_mentions = sum of mention_count across all tracked brands
mention_sov(brand) = mention_count(brand) / total_mentions
Track per-engine and total. The per-engine breakdown is where the strategic insight lives.
Step 6: Interpret responsibly
Some honest caveats:
- A 5% week-over-week SOV swing is usually noise. Don't redesign your content strategy on it.
- Trends across 4+ weeks are signal. Especially when correlated with content publishing or schema work.
- One viral mention can spike SOV temporarily. Decouple that from sustained gains.
- Your SOV isn't comparable to another industry's SOV. A 15% SOV in a 5-brand category is dominant; in a 50-brand category it's elite.
What actions move SOV
Across the AEO programs we've seen, the moves that move SOV the fastest:
- Publishing or improving the cluster page for prompts where you're absent (see the 25-step AEO checklist).
- Adding or upgrading Organization + Article + FAQPage schema (see schema markup for AEO).
- Earning a placement in 1–2 high-authority third-party listicles for the category.
- Updating evergreen content with fresh dates and 200+ words of new info.
A reporting template
For internal share-of-voice reporting, the cleanest format:
| Metric | This week | Δ vs last | Δ vs 4 weeks ago |
|---|---|---|---|
| Mention SOV (all engines) | XX% | +/- pp | +/- pp |
| Mention SOV (ChatGPT) | XX% | +/- pp | +/- pp |
| Mention SOV (Perplexity) | XX% | +/- pp | +/- pp |
| Mention SOV (AI Overviews) | XX% | +/- pp | +/- pp |
| Citation SOV (all engines) | XX% | +/- pp | +/- pp |
| First-mention SOV | XX% | +/- pp | +/- pp |
Pair the table with three things every week: top 5 prompts where you gained, top 5 where you lost, and one root-cause hypothesis per loss.
The bottom line
Share of voice in AI answers is the closest analogue to keyword rank tracking that AEO has. Build the prompt set carefully, sample with rigor, and read the trend — not the snapshot.
Set up free SOV tracking on ApexEcho to skip the spreadsheet phase entirely.