What is AI share of voice?

It is the percentage of AI answers, across a defined set of prompts and engines, in which a brand is mentioned or cited, usually compared with named competitors. It is the AI equivalent of share of voice in search or advertising.

How many prompts do I need?

Thirty to sixty prompts per client is a practical range for most categories. Fewer than twenty gives unstable results; more than a hundred adds cost and noise without much extra insight.

How often should I run the prompts?

Monthly is the most common cadence for client reporting. Run each prompt more than once per cycle if possible, because AI answers vary between runs.

What is the difference between a mention and a citation?

A mention is the brand's name appearing in the answer text. A citation is a link to the brand's own site used as a source. A brand can be mentioned without being cited, for example when a review site is cited instead.

Can I compare AI share of voice across clients?

Only if the method is identical: the same engines, run counts, settings and scoring rules, and prompt sets built the same way for each category. Even then, categories differ in how many brands AI answers mention, so compare each client against its own competitors and its own trend rather than ranking clients against each other.

AI Share of Voice: How to Measure Brand Visibility in AI Answers

Rankings tell you where a page appears in a list. AI answers do not have a list. They name a few brands, cite a few sources and move on. To know how a client is doing, you need a different measure: how often they appear in the answers that matter, compared with the brands they compete against.

That measure is AI share of voice. This guide sets out a method you can run manually or with a tool, so that the numbers are consistent from month to month and comparable across clients.

AI share of voice is the share of relevant AI answers that include a brand. It has three components, and it is worth tracking all three.

Component	Definition	Why it matters
Mention rate	Share of answers that name the brand	Exposure and recommendation
Citation rate	Share of answers that link to the brand's own site	Credit, traffic potential and trust
Position	Where the brand appears in the answer, such as first-named	Prominence

The difference between mentions and citations is important. A brand is often mentioned because a review site or directory says good things about it, while that third-party site gets the citation. Both are good, but they call for different work. More citations come from improving the brand's own pages; more mentions come from wider reputation. See entity SEO for LLMs.

Step 1: Build the prompt set

The prompt set is the foundation. A good one reflects what real customers ask, across the whole buying journey.

Where prompts come from

Sales and support teams. The questions prospects ask before they buy.
Search Console queries. Especially longer, question-style queries.
"People also ask" boxes for the client's core topics.
Community discussions on forums and social platforms where customers talk about the category.
Competitor positioning. The comparisons people make between the client and its rivals.

How to structure it

Aim for 30 to 60 prompts, grouped by stage.

Stage	Example prompt (accounting software client)	Share of set
Problem	"How do freelancers keep track of expenses for taxes?"	20%
Category	"Best accounting software for freelancers in the UK"	35%
Comparison	"Xero vs QuickBooks for a one-person business"	25%
Brand	"Is [client] good for freelancers?"	10%
Local or niche	"Accounting software that handles CIS for contractors"	10%

Write prompts the way people actually type or speak them. Avoid stuffing them with keywords. Keep the set fixed for at least a quarter so trends are meaningful; add new prompts as a separate group rather than replacing old ones.

Choosing competitors to benchmark

Share of voice only means something against a defined competitor set. Pick four to six brands:

Two or three direct competitors the client names in sales conversations.
One or two category leaders that AI answers mention most often, even if the client does not see them as rivals. Run a few category prompts first to find them.
One aspirational brand the client wants to be compared with, if relevant.

Record name variations for each brand, such as abbreviations and product names, so mentions are counted consistently. Keep the set stable for at least a quarter, and review it at each quarterly strategy meeting.

Step 2: Choose engines and settings

Track the engines the client's audience uses. For most clients that means ChatGPT with search available, Perplexity, Google AI Overviews and AI Mode, and often Gemini, Copilot and Claude.

Keep conditions consistent:

Clean sessions. Signed-out or fresh accounts without personal history, so personalisation does not skew results.
Consistent location. Set or record the location, since answers vary by country and city.
Consistent timing. Run each cycle in the same week of the month.

Different engines behave differently. Perplexity searches the web and cites on nearly every answer; ChatGPT sometimes answers from model knowledge without citations. That is why each engine is scored separately. See how ChatGPT and Perplexity choose sources.

Step 3: Run each prompt more than once

AI answers vary between runs, even with the same prompt and settings. One run can make a brand look absent when it usually appears, or present when it rarely does. Running each prompt two or three times per cycle and averaging the results gives a much more stable picture. If budget or time is tight, run fewer prompts more often rather than more prompts once.

Step 4: Record the results consistently

For every answer, record the same fields.

Engine, prompt, run number and date.
Brands mentioned, in order of appearance.
Domains cited, and which page of each.
Whether the client is mentioned, cited, both or neither.
Sentiment of the mention: positive, neutral or negative.

A spreadsheet works for a single client. For a portfolio, a tool or a database becomes necessary.

With the data recorded, the calculations are simple.

Mention rate = answers mentioning the brand ÷ total answers.
Citation rate = answers citing the brand's site ÷ total answers.
Share of voice = the brand's mentions ÷ all brand mentions across the competitor set.
First-mention share = answers where the brand is named first ÷ total answers.

Report each by engine and by stage. A brand can lead on comparison prompts and be absent from problem-stage prompts, which points directly to a content gap.

An illustrative example

Across 40 prompts, three runs each, on Perplexity (120 answers), a client is mentioned in 54 answers and cited in 31. Its three competitors are mentioned 70, 41 and 23 times.

Mention rate: 54 ÷ 120 = 45%
Citation rate: 31 ÷ 120 = 26%
Share of voice: 54 ÷ (54 + 70 + 41 + 23) = 29%, second of four

Numbers like these are easy for clients to understand and to track.

Step 6: Turn the data into actions

The value of share of voice is in what it points to.

Pattern	What it means	What to do
Mentioned but rarely cited	Third parties vouch for the brand; its own pages are not used	Rewrite key pages answer-first; check AI crawler access
Cited on Perplexity, absent in ChatGPT	Pages are strong; wider reputation is weak	Build presence on category review and comparison sites
Absent from a whole stage	Content gap	Create content for that stage's questions
Negative sentiment	Engines repeat a known complaint	Address it in content and on review platforms
Competitor dominates one engine	They have presence on that engine's favoured sources	Identify the cited domains and earn presence there

How to report it

Clients respond best to a simple view: share of voice against competitors, by engine, with the trend over time, followed by the two or three actions planned for the next month. Put it next to AI referral traffic and Search Console AI impressions, so the client sees exposure and visits together. See how to track ChatGPT traffic in GA4 and proving SEO value when clicks fall.

Common mistakes

Changing the prompt set every month. Trends become meaningless.
Running each prompt once. Variation between runs can swamp real change.
Using personalised sessions. Results reflect the tester, not the market.
Only counting citations. Mentions without citations are valuable and need different work.
Ignoring competitors. A 30% mention rate means little without knowing whether rivals have 10% or 80%.

Doing it manually versus with a tool

A manual process works for one or two clients and is a good way to learn the method. Beyond that, the work grows quickly: 40 prompts, three runs and five engines is 600 answers per client per month. Tools automate the runs and recording. Whatever you use, make sure you understand and control the prompt set, the engines, the run count and the scoring rules, so the numbers mean what you think they mean.

How Rankwyn helps

Rankwyn's AEO Citation Radar runs each client's fixed prompt set across engines on a schedule, records mentions, citations, position and cited domains against named competitors, and reports share of voice and its trend for every site in the portfolio.

Key takeaways

AI share of voice is the share of relevant AI answers that mention or cite a brand, compared with competitors.
Track mention rate, citation rate and position separately, by engine and by buying stage.
Build 30 to 60 realistic prompts and keep the set fixed for at least a quarter.
Run each prompt more than once per cycle in clean, consistent sessions, because answers vary.
Use the patterns in the data to choose actions: on-page work for citations, reputation work for mentions.

The bottom line

AI share of voice gives clients a clear answer to "how visible are we in AI search?" Build a realistic prompt set, keep the conditions consistent, run prompts more than once, score mentions and citations against competitors, and report the trend with the actions it points to. For the wider strategy, see our AEO and GEO guide.