What Is Share of Voice in AI Search?
Share of voice in AI search is the percentage of your tracked questions where an engine mentions or cites your brand, measured against competitors. It's the core AEO metric because AI answers replace clicks with citations. Here's how to define, calculate, and track it — and why a single reading lies.
Share of voice in AI search is the percentage of your tracked questions where an engine mentions or cites your brand, measured against competitors. It's the core AEO metric because AI answers replace clicks with citations — so what matters isn't where you rank in a list, it's how often you're named in the answer itself.
The short answer
Share of voice (SOV) = the share of your tracked questions where the engine mentions or cites you, vs competitors. Run 100 prompts, appear in 30 answers → 30% SOV. It's the AEO version of rank tracking — and the only honest reading is a trend across a fixed prompt set, not a single snapshot.
What does share of voice mean in AI search?
Share of voice in AI search is the proportion of your tracked questions for which an AI engine names or cites your brand in its answer. It reframes the old SEO idea of share of voice — your slice of total search visibility for a keyword set — for a world where the "result" is a synthesized answer, not a ranked list. Instead of "what position do I hold?", SOV asks "how often am I in the answer at all?" That's the visibility that counts when AI search replaces clicks with citations.
How do you calculate AI share of voice?
You calculate AI share of voice by running a fixed prompt set, logging your brand's appearances, and dividing. There are two views worth tracking:
- Brand share of voice = answers mentioning you ÷ total prompts. (Out of every question you track, how often do you show up?)
- Competitive share of voice = your appearances ÷ all brand appearances across you and your competitors. (Of the brand mentions in your space, what slice is yours?)
| Metric | Formula | Answers the question |
|---|---|---|
| Brand SOV | Your mentions ÷ total prompts | How often am I in the answer? |
| Competitive SOV | Your mentions ÷ all brands' mentions | What's my slice of the conversation? |
| Citation SOV | Times your site is cited ÷ total prompts | How often am I the linked source? |
Compute each per engine, because engines cite very different sources — a blended average hides where you're winning or losing. That's important enough to have its own guide: why per-engine measurement beats a blended average.
A worked example
Say you track five questions on one engine and log who each answer cites:
| Prompt | You | Competitor A | Competitor B |
|---|---|---|---|
| Best X for small teams? | ✓ | ✓ | |
| X vs Y — which is better? | ✓ | ✓ | |
| Is X worth it? | ✓ | ||
| Top X tools in 2026 | ✓ | ✓ | ✓ |
| How do I choose an X? | ✓ |
You appear in 3 of 5 answers, so your brand SOV is 60% (3 ÷ 5). Across all brand appearances — yours (3) plus Competitor A's (4) and Competitor B's (2), nine in total — your competitive SOV is 33% (3 ÷ 9). Run the same five prompts next week and the numbers will move; that's why you track the trend, not the snapshot.
Why is share of voice the metric that matters?
Share of voice matters most because AI answers compress many results into one response, so being named beats being ranked. In classic search you could rank #3 and still earn clicks; in an AI answer, if you're not cited, you're invisible regardless of where you'd have ranked. SOV measures the thing that now drives visibility: presence inside the answer. It pairs with two supporting metrics — AI referral traffic (visits from chatgpt.com, perplexity.ai, and the like) and branded-search lift — but citation share is the headline number.
What counts as an appearance — a mention or a citation?
Decide up front what counts, because "appearing" in an AI answer has three distinct levels and conflating them muddies the metric:
- Mention — your brand name appears in the answer text, with no link. It builds awareness and signals authority, but the reader can't click through.
- Citation — your site is listed as a source for the answer (the linked footnote or source card). This is the highest-value appearance: it credits you as the source and earns the rare click.
- Sentiment — how you're named. Being cited as the recommendation is a different result from being named as the cautionary example, even though both count as an "appearance."
Track mention SOV and citation SOV separately — being widely mentioned but rarely cited (or the reverse) calls for different work. Position matters too: a name in the first sentence carries more weight than one buried in a closing list, mirroring the extractability bias toward the top of the answer.
How do you track it?
Track share of voice with a fixed prompt set, a regular cadence, and per-engine logging — the same method whether you do it by hand or with a tool.
- 1
Build a fixed prompt set
15–30 real, high-intent questions your customers ask. See how to build an AI prompt set — there's a downloadable template.
- 2
Choose your engines
Track the engines your audience actually uses (ChatGPT, Perplexity, Google AI Overviews, etc.) — and score each separately.
- 3
Run on a cadence
Daily or weekly. Because answers are non-deterministic, frequency is what turns noisy readings into a reliable trend.
- 4
Log mentions and citations
For each answer, record whether your brand is mentioned, whether your site is cited, and which competitors appear.
- 5
Compute SOV and watch the trend
Calculate brand and competitive SOV per engine, and judge the direction over weeks — never a single reading.
The full manual method is in how to measure your AI visibility; to automate it, see the best AI visibility tools; to benchmark rivals, see how to track competitor AI citations.
What's a good share of voice?
A "good" share of voice is higher than your direct competitors' on the questions that matter to you — not a fixed percentage. Absolute SOV numbers run lower than people expect, because AI citations are spread thin: Evertune found no single domain exceeds about 5% of citations in a given space. So a 15–20% brand SOV on your core buying questions can be category-leading, while a flashy 60% on awareness questions nobody buys from is worth little.
Judge SOV three ways: relative (you versus named competitors), weighted by intent (decision-stage questions count more than awareness), and as a trend (up-and-to-the-right beats a high but sliding number). Set your baseline from the first month of readings, then target steady gains on the decision questions that precede a purchase.
Why does the number keep moving?
The number keeps moving because AI answers are non-deterministic and citations are volatile — the same prompt can return different sources from one run to the next. Semrush found 40–60% of LLM-cited sources change month to month, and analysts like Rand Fishkin of SparkToro have repeatedly stressed how unpredictable and zero-click AI search has become. The implication for SOV is firm: one measurement is a coin flip, but a fixed prompt set sampled repeatedly reveals a real, defensible trend.
Don't trust a single reading
Because the same question can yield different citations run to run, a one-time share-of-voice number is noise. Measure the same prompt set on a schedule and report the trend. Treat exact percentages as directional, and watch the slope over weeks — that's the Adaptability pillar applied to measurement.
What are the most common share-of-voice mistakes?
The most common mistakes turn a useful metric into a misleading one — and most of them are easy to avoid once you know the trap.
What corrupts a share-of-voice number
Measuring once: a single reading is noise, not a number. Churning the prompt set: swap questions and you lose comparability — freeze the list and add deliberately. Blending engines: averaging across engines that barely overlap hides where you actually stand (see per-engine vs blended). Counting mentions and citations as one thing: they're different wins and need separate tracking. Chasing vanity questions: high SOV on questions that never precede a purchase flatters the dashboard and changes nothing.
Where this fits in the Canon
Share of voice is how you instrument the Adaptability pillar — you can't adapt to what you don't measure, and SOV is the measurement. It depends on a good prompt set (Alignment) and is most honest when computed per engine.
Next: why per-engine measurement beats a blended average, how to build an AI prompt set, and how to measure your AI visibility.
Frequently asked questions
- What is share of voice in AI search?
- Share of voice (SOV) in AI search is the share of your tracked questions where an AI engine mentions or cites your brand, relative to competitors. If you run 100 prompts and your brand appears in 30 of the answers, your share of voice is 30%. It's the AEO equivalent of rank tracking — it measures presence inside the answer rather than position in a list of links.
- How do you calculate AI share of voice?
- Run a fixed set of questions across the engines your audience uses, log whether each answer mentions or cites your brand, and divide your appearances by the total. Brand SOV is your appearances over all prompts; competitive SOV is your appearances over the total brand appearances across you and your competitors. Always calculate it per engine, then track the trend over time.
- Why is share of voice the main AEO metric?
- Because AI answers reduce clicks, so position in a link list matters less than whether you're named in the answer at all. Share of voice measures exactly that — how often you're the cited source — which is the visibility that survives in a zero-click, answer-first world. Rankings tell you where you sit in a list nobody may scroll; SOV tells you whether you're in the answer.
- Why does my share of voice keep changing?
- Because AI answers are non-deterministic and citations are volatile. The same prompt can yield different sources run to run, and Semrush found 40–60% of cited sources change month to month. Commentators like Rand Fishkin have stressed how unpredictable AI answers are. So a single reading is noise — measure a fixed prompt set repeatedly and judge the trend, not any one result.
- What counts as an appearance — a mention or a citation?
- They're different and worth tracking separately. A mention is your brand name in the answer text (awareness, but no click); a citation is your site listed as a source (the high-value appearance that credits you and can earn the rare click). Sentiment matters too — being named as the recommendation beats being named as the cautionary tale. Track mention share of voice and citation share of voice separately.
- What's a good AI share of voice?
- Higher than your direct competitors on the questions that matter — not a fixed number. Absolute share of voice runs lower than people expect because citations are spread thin (Evertune found no single domain exceeds about 5% in a given space), so 15–20% on core buying questions can be category-leading. Judge it relative to rivals, weighted toward decision-stage questions, and as a trend.
Last updated .