Share of voice in AI search is the percentage of relevant AI-generated answers, from tools like ChatGPT, Google’s AI Overviews, and Perplexity, where your brand appears compared to competitors. You calculate it by running a fixed set of prompts across these platforms, logging every mention, and dividing your brand’s appearance count by the total number of brand mentions across all competitors. At Peak Marketing, this has become one of the first metrics we set up for clients who want to understand where they stand as search shifts from links to answers.
Traditional share of voice measured how often you showed up in search results or media coverage relative to competitors. The AI search version asks a different question: when someone asks an AI assistant a question in your category, does it mention you at all, and if so, how prominently?
This matters because the mechanics are different from ranking on a results page. An AI model doesn’t return ten blue links. It synthesizes an answer from whatever sources it trusts, and either includes your brand or it doesn’t. There’s no position two or three to settle for. You’re either part of the synthesized answer or you’re invisible to that user entirely.
Why This Metric Is Different From Traditional SOV
Classic share of voice tools pull from search engine results pages, social mentions, or ad impression data. They assume a stable, crawlable list of results you can rank in.
AI answers don’t work that way. The same prompt can return different results depending on the day, the model version, and even small wording changes in the question. A brand can appear in an answer on Monday and disappear from that same answer by Friday, with nothing on the brand’s end having changed. This volatility is part of why AI search share of voice needs to be tracked over time rather than checked once and filed away.
There’s also no guaranteed link back to your site. A mention in an AI Overview or a ChatGPT response might reference your brand name without ever sending traffic your way. That doesn’t make the mention worthless. Brand recall from an AI answer still shapes whether someone searches for you directly later, but it changes what you’re actually measuring.
Building a Prompt Set That Reflects Real Buyer Questions
The foundation of any AI share of voice tracking system is the prompt list. This is the set of questions you’ll run repeatedly across platforms to see who gets mentioned.
Good prompt sets pull from three sources:
- Actual customer questions from sales calls, support tickets, or intake forms
- Long-tail keyword data from tools like Ahrefs or Semrush, filtered for question phrasing
- Category comparison questions your prospects are likely asking before they buy
For a dental practice client, this might mean testing prompts like “what should I expect at a first dental implant consultation” rather than just “best dentist near me.” AI models tend to answer specific, needs-based questions differently than they answer broad local searches, so your prompt set needs to reflect the actual decision-making questions people ask, not just the keywords you’d target in traditional SEO.
Keep the list between 20 and 50 prompts for a first pass. Fewer than that and the data is too thin to spot patterns. More than that and manual tracking becomes unmanageable before you’ve even proven the exercise is worth automating.
Running the Prompts and Logging What Comes Back
Once the prompt list exists, run each one across the AI platforms that matter most to your audience. For most businesses that means ChatGPT, Google’s AI Overviews, and increasingly Perplexity, though the right mix depends on where your buyers actually spend time.
For each response, log four things: whether your brand appears, where in the answer it appears (first mention, buried in a list, footnoted as a source), which competitors appear alongside you, and whether the AI cites a specific page or just references your brand name generically.
That last point matters more than it might seem. A citation with a link tells you which piece of content the model pulled from, which means you can go look at that page and figure out why it earned the mention. A bare name-drop with no citation tells you the model has some general awareness of your brand but nothing you can directly optimize toward yet.
Do this on a recurring schedule, not as a one-off audit. Monthly is a reasonable cadence for most businesses. Categories with fast-moving competitive dynamics, like legal services in a specific metro area, may warrant a tighter check-in.
Calculating the Actual Share of Voice Number
Once you have a few rounds of data, the math is straightforward. Count how many prompts returned a mention of your brand. Divide that by the total number of prompts you ran. That gives you your raw appearance rate.
To turn that into a true share of voice figure, you need the competitive context. Count every brand mention across all competitors for the same prompt set, then divide your brand’s mention count by that total. If you appeared in 12 of 40 prompts, and your top three competitors combined appeared in 60 total mentions across those same 40 prompts, your share of voice is roughly 17 percent of the competitive conversation, not 30 percent of your own prompt list.
This distinction trips people up. A high appearance rate against your own prompts can still mean a small share of voice if competitors are dominating the same answers alongside you.
What to Do With a Low Score
A low AI search share of voice usually traces back to one of a few root causes: thin or generic content that doesn’t answer the specific question being asked, a lack of structured data or clear factual statements the model can cite confidently, or simply not having authoritative third-party mentions that would give an AI model a reason to trust your brand as a source.
The fix isn’t a single tactic. It’s usually a combination of tightening existing pages so they answer specific questions directly near the top, building out content for the gaps your prompt testing revealed, and picking up mentions on industry sites or directories that AI models already treat as trustworthy sources.
This is the kind of tracking and content work Peak Marketing builds into ongoing SEO programs for clients across law, healthcare, and retail, because the brands showing up in AI answers today are rarely the ones scrambling to catch up after the fact. Start with a prompt list this month, run it consistently, and you’ll have real data instead of a guess about where you stand.
