AI share of voice: how to measure it against your competitors
Share of voice in AI answers is a count, not a feeling. Here is how to pick the prompts, run them across the engines, count who gets named and track it over time.
On this pageShow 9Hide
Share of voice used to mean ad impressions or the slice of a category's search clicks you won. In AI search it means something simpler: when a buyer asks an assistant about your category, how often does the answer name you, and how often does it name your competitors instead?
That is a count. You can do it by hand for one topic, and you should, because doing it once by hand is the fastest way to understand what a tool is counting for you later.
What AI share of voice measures
AI share of voice is the share of answers, across a fixed set of prompts and engines, that name a brand. Run the same prompts through ChatGPT, Gemini, Perplexity and Google AI Overviews, record which brands each answer recommends, and divide.
Two numbers matter, and they should be kept apart:
- Mention share: the share of answers that name your brand at all.
- Citation share: the share of answers that link to one of your own pages as a source.
A brand can be named often and cited rarely, for example when the engines describe it from reviews, comparison pages and roundups written by other people. The reverse happens too. If you average the two, you cannot tell which one moved.
Check your AI visibility free
See which buying questions ChatGPT, Perplexity, Gemini and Google's AI answers name you for, and who they name instead. No call, no card.
Step 1: pick the prompts your buyers actually use
The prompt set decides the result, so build it before you look at any answers. For each topic you sell into, write five prompts from different angles:
- Category: "best invoicing software for freelancers"
- Positioning: the thing you claim to do differently, asked as a need
- Sector: the category for a specific kind of buyer
- Outcome: the result the buyer wants, in their words
- Comparison: "alternatives to" a competitor you lose deals to
Write them the way a buyer types, not the way your homepage talks. Keep the set fixed once you start measuring. If you change the prompts every week, you are measuring your prompt writing, not your visibility.
Step 2: run every prompt on every engine
Put each prompt to each engine you care about. Four engines and five prompts gives you 20 answers per topic. For each answer, record:
- every brand the answer names or recommends
- whether your brand is one of them
- the pages the answer cites, where the engine shows sources
- the date
Use a clean session with memory and chat history turned off where you can. OpenAI says that with Reference chat history on, ChatGPT can use past conversations to personalize responses, and you want the answer a stranger would get.
Step 3: count it
Here is a made up example, so the arithmetic is easy to follow. Three brands compete in one topic. You ran five prompts on four engines, so there are 20 answers.
- Brand A is named in 12 of the 20 answers: 60 percent mention share.
- Brand B is named in 8: 40 percent.
- Your brand is named in 5: 25 percent.
Shares across brands do not add up to 100, because one answer can name several brands. That is fine. You are measuring how often each brand is in the answer, not splitting a fixed pie.
Now count citations the same way. Suppose 3 of the 20 answers link to one of your pages: 15 percent citation share. Brand A might have a high mention share and almost no citations, which suggests its visibility comes from what other sites say about it.
Then look at the engines one by one. A 25 percent overall share can hide a brand that is named in four of five ChatGPT answers and none of the Gemini ones. That split shows you where to look first.
Step 4: find out why the leaders are named
The count tells you where you stand. The sources tell you why. Open the cited pages in the answers where a competitor is named and you are not, and sort them by type: roundups of the best options, comparison pages, review sites, forum threads, or the competitor's own pages.
Those pages are the input. If the engines keep citing a handful of roundups that list three competitors and not you, the fix is to get onto those pages, not to rewrite your homepage. Our guide to LLM visibility goes through what to look for in the sources.
Step 5: measure it again, on a schedule
The models behind AI answers are not deterministic. OpenAI's API documentation says outputs may differ from request to request, so one reading is a sample, not a verdict. Measure on a schedule with the same prompts and you get a trend. Weekly is the minimum that tells you anything; daily is better when you are actively working on a topic, because you can see which change moved which answer.
At this point the spreadsheet usually breaks. Five prompts across four engines is 20 answers for one topic. Ten topics is 200 answers per reading, each with its brands and sources to log. That is the job AI rank trackers exist for.
Where alwayscited fits
We built alwayscited for this problem, so weigh this section accordingly. Our free checker runs five buyer prompts across Google AI Overviews, ChatGPT, Gemini and Perplexity and shows every answer, the pages each engine cited and the brands named instead of you. That is enough to get your share of voice free for one topic today.
The alwaystracked plan runs the same check every morning: 10 clusters, each one a Google keyword joined to five prompts, so 50 prompts and 10 keywords checked daily across those four engines. It starts from $129 a month, with a 14 day free trial (card required). It reports; it does not do the placement work. That sits in the higher tiers.
Common mistakes
- Measuring one prompt. One prompt is an anecdote. Use at least five per topic.
- Mixing mentions and citations. Track them as two numbers.
- Changing the prompt set mid stream. Add new prompts as a new set, and keep the old ones running.
- Ignoring the engine split. An average across engines hides the engine where you are missing.
- Treating one reading as final. Answers move. Look at the trend.
Questions
- What is a good AI share of voice?
- There is no universal benchmark, because it depends on how many brands compete in your category and how specific your prompts are. Compare yourself with the brands you lose deals to, on the same prompts, and watch the direction over a few weeks rather than chasing one number.
- How many prompts do I need to measure AI share of voice?
- Enough to cover the ways buyers ask about your category. Five prompts per topic, written from different angles, is a practical minimum. A single prompt is too easy to over read.
- Should I count mentions or citations?
- Count both, separately. A mention means the answer names your brand. A citation means the answer links to one of your pages as a source. They can move independently, so blending them hides what changed.
- How often should I measure it?
- The models are not deterministic, so the same prompt can return a different answer and a single reading is noisy. Daily or weekly readings on a fixed prompt set give you a trend you can act on.
Check your AI visibility free
See which buying questions ChatGPT, Perplexity, Gemini and Google's AI answers name you for, and who they name instead. No call, no card.