What is an AI visibility agent and how does it measure share of voice in ChatGPT and Perplexity answers?

An AI visibility agent is software that repeatedly asks answer engines such as ChatGPT and Perplexity the questions your buyers ask, records which brands each answer names and which sources it cites, and then works out what to change. Share of voice is your slice of those answers: the answers that mention or cite you, divided by the answers that mention or cite any brand you track, on the same fixed prompts, engines and dates.
What is an AI visibility agent, and how is it different from a tracker?
A tracker reports. An agent also acts, within limits you set.
A tracker runs prompts, stores answers, and shows mention rates, citations and sentiment on a dashboard. Semrush's AI Visibility Toolkit and Ahrefs Brand Radar are good examples of strong measurement tools.
An agent does that, then diagnoses why you are missing, drafts fixes such as a direct answer block or structured data, ranks them, and routes each for approval or applies it where it has permission.
The line is blurring. Peec AI now lists recommended actions and an agent with skills on its plans, and Profound's agents draft and publish content after an approval step. The useful question is no longer "tracker or agent" but "what does it change, and who signs off?" Nagent's own agent, DRIS, is one example, and it gets its own section below.
What does share of voice mean in AI answers?
In classic marketing, share of voice is your share of the conversation in a category. In AI answers it narrows to a measurable ratio, but vendors define it differently, so read the formula before comparing numbers.
Ahrefs defines AI share of voice as "the share of AI visibility your brand captures compared with the competitors you track, across the prompts you care about", and gives the basic formula as responses mentioning your brand divided by responses mentioning any tracked brand (Ahrefs). Brand Radar then weights each prompt by an estimated AI search volume.
Semrush reports an AI Visibility Score based on how often your brand is mentioned compared with the median for your top competitors, and analyses share of voice in Brand Performance (Semrush).
Peec AI tracks visibility, position, sentiment and share of voice daily against competitors (Peec AI).
Two distinctions matter more than the brand of tool:
Mention or citation. A mention is your name in the answer text. A citation is your URL in its sources. Perplexity lists its sources with its answers; ChatGPT shows sources when it searches the web. You can be mentioned without being cited, and cited without being recommended.
Weighted or unweighted. Counting every prompt equally treats a niche question the same as a high-demand one. Weighting by estimated volume is closer to reality, but the weights are themselves estimates.
How is AI share of voice calculated? A worked example
Say you track 40 buyer prompts on ChatGPT and Perplexity, asked once a week for four weeks: 40 × 2 × 4 = 320 answers. You track two competitors.
Answers that mention any of the three brands: 200.
Answers that mention you: 50. Competitor A: 120. Competitor B: 90. (Some answers name more than one.)
Mention share of voice, per answer: you appear in 50 of the 200 answers that name anyone, which is 25%. Another common variant divides each brand's mentions by all brand mentions: 50 ÷ (50 + 120 + 90) = 19%. Both are valid; they are not interchangeable. Pick one and keep it.
Now citations. Of the 320 answers, 140 cite a site belonging to one of the three brands, and 35 of those cite yours. Your citation share is 35 ÷ 140 = 25%. If Perplexity cites you in 30 answers and ChatGPT in 5, the blended 25% hides an engine-level problem.
Why do AI share of voice numbers jump around?
Because answer engines do not give the same answer twice, and because measurement choices shift the denominator.
Sampling noise. One run per prompt per week is thin. Peec's measurement guide cites a practitioner's rule of thumb of about 10 responses per prompt for a quick estimate (Peec AI).
Panel changes. Adding 40 new prompts adds citations without you being any more visible. Report per-answer rates, not raw counts.
Competitor list changes. Adding a competitor lowers your share overnight.
API versus app. Most tools query engine APIs. Answers in a signed-in app with memory can differ.
Decay. Profound found the median page loses half its citation share within 11 days of peaking (Profound), so last month's win may already be fading.
A trustworthy report states the prompt count, engines, run count and window beside every figure, and refuses to call a change when the ranges overlap.
What should an AI visibility agent measure besides share of voice?
Coverage: the share of your prompts on which you are cited at least once.
Recommendation position: where you appear when the answer lists options.
Sentiment and accuracy: whether what the answer says matches your confirmed facts.
Source gaps: sites the engines cite for your competitors but not for you, such as review sites or forums.
AI referrals: sessions from chatgpt.com or perplexity.ai in your analytics.
Crawler access: whether OAI-SearchBot and PerplexityBot can fetch your pages. OpenAI says sites that block OAI-SearchBot will not appear in ChatGPT search answers (OpenAI), and Perplexity asks sites to allow PerplexityBot to appear in its results (Perplexity).
How do you set up share of voice tracking in a week?
Day 1: pick the prompts. Pull 30 to 50 questions from sales calls, Search Console queries and People Also Ask boxes. Keep them unbranded unless you want to test how engines describe you.
Day 2: pick the competitors. List three to five you actually lose deals to, plus any site the engines keep citing beside you.
Day 3: pick the engines. Start with the two or three your buyers use. Ask sales which assistants prospects mention.
Day 4: fix the definitions. Write down whether share of voice counts mentions or citations, per answer or per brand mention, weighted or not. Put that line on every report.
Days 5 to 7: run the baseline. Ask each prompt several times per engine, store every answer and its sources, and calculate the first figures.
From then on, change prompts only on a set date, and note it on the chart.
What should you ask a vendor about share of voice?
Is your share of voice based on mentions, citations or both?
Is it weighted by search volume, and where do the weights come from?
How many answers sit behind each figure, and do you show the range?
What happens to history when I add a prompt or a competitor?
Do you query engine APIs, logged-out apps or logged-in apps?
How does Nagent's DRIS measure share of voice?
We make Nagent, and its AI visibility agent is DRIS. It reports to MIRA, the AI Marketing Chief of Staff, and asks ChatGPT, Claude, Gemini, Perplexity and Grok the prompts you accept.
DRIS separates its measures so they are not confused:
Citation share: answers citing you, over answers citing you or any competitor on your list, each competitor counted once per answer.
Mention share: answers whose text names you, linked or not, over every tracked brand's, each brand once per answer.
Share of voice (per tracked page): your citations against the competitor pages linked to that page, on the same prompts. With no competitor page linked, it reads 0.
Coverage: the share of a page's prompts on which at least one answer cited you.
Its Analytics view (Beta) compares each figure with the same length of time before it, and shows a change only when both periods hold at least eight observations and their 95% ranges no longer overlap; otherwise it reads "no clear change". One answer citing three brands counts as one observation. Repeated samples and failed captures are left out.
Beyond measuring, DRIS drafts fixes ranked against an objective you set, and nothing on a live page changes until a person approves. Limits: answers come from engine APIs with a simulated memory, not signed-in apps, and a competitor tracked only by domain is matched on its domain's first word, which can overcount brands with common-word names.
Frequently Asked Questions
How do I track whether ChatGPT and Perplexity mention my brand?
Write 20 to 50 buyer questions, ask them on a fixed schedule, and record mentions, cited URLs and competitors for each answer. A tool automates this; a spreadsheet works for a first month.
What is a good AI share of voice?
There is no universal benchmark, because definitions differ. Compare yourself with the same competitors, prompts and engines over time, and aim to lead on the prompts closest to a purchase.
What is the difference between AI visibility and AI share of voice?
Visibility is how often you appear at all. Share of voice is how often you appear relative to competitors on the same prompts.
Can an AI visibility agent change my website?
Some can. Ask whether changes publish automatically, after a review step, or only after a named person approves, and whether index or canonical changes need a separate sign-off.
See how DRIS reports share of voice on your own prompts: book a Nagent demo.