How to Measure Your Share of Voice When AI Is Doing the Talking

For twenty years, share of voice was a percentage. You owned some slice of impressions or clicks for a keyword, a competitor owned another slice, and you fought over the remainder. There was a page with ten blue links and you could point at your position on it.

That page doesn't exist anymore for a growing share of questions. Someone asks ChatGPT "who's the best HVAC company in Plano" and gets three names in a sentence, maybe with a line about why. There's no page ten to rank on. There's no position four. Either your name shows up in that sentence or it doesn't.

Share of voice in this world is simpler to define and harder to measure: out of every relevant question a buyer might ask an AI model, what percentage of the time does your business get named. That's the whole metric. The hard part is that "every relevant question" spans dozens of phrasings across at least three engines people actually use, and the answer changes week to week without you touching anything.

SEO share of voice tools work because search results are structured and stable enough to scrape. A rank tracker hits Google, parses the page, and logs your position. Run it daily and you get a clean trend line.

AI answers don't hold still. Ask the same model the same question twice and you can get different competitors named, different phrasing, sometimes a different verdict on who's best. The model isn't looking up a cached ranking, it's generating a response based on what it has absorbed about your industry and your specific business, and that absorption shifts as it retrains, as your content changes, and as your competitors publish new material.

That means a share of voice number from a single query is close to meaningless. You need repeated sampling across a spread of real buyer phrasings, tracked over time, before a pattern means anything. One good mention doesn't mean you're visible. One bad session doesn't mean you're invisible. The trend across weeks is the signal.

A lot of AI visibility tools produce one composite score and call it your AI visibility rating. It looks clean in a dashboard. It's also lying to you, because it's averaging together three models that don't agree with each other.

We built OMG around per-model scoring instead of union inflation, meaning your ChatGPT number, your Perplexity number, and your Gemini number stay separate. This matters in practice more than it sounds like it should. We've seen businesses get named constantly in Perplexity, which leans hard on live web citations, while being nearly invisible in ChatGPT, which draws more from what it learned during training. Blend those two into one score and you get a mediocre-looking average that hides both the win and the problem. Keep them separate and you know exactly where to spend effort.

Starter plans on OMG track two engines, Pro tracks three, and Enterprise expands to five, because the right panel size depends on how much of your traffic is actually AI-referred versus how much is still coming through classic search.

Getting cited by an AI model isn't about publishing more content. It's about giving the model something specific and unambiguous to repeat. Vague marketing copy doesn't quote well. A sentence like "we're passionate about quality service" gives a language model nothing to extract, because it can't tell if that's true or just what every competitor's homepage also says.

Specific claims get cited. "Licensed in Texas since 2011, average response time under 45 minutes for emergency calls" is a sentence a model can pull into an answer and attribute to you, because it reads as a fact rather than a slogan. Same goes for pricing ranges, service area boundaries, certifications, and named staff with credentials. If a fact appears in one consistent form across your site, your Google Business Profile, and any directories that list you, models converge on it faster.

This is also where a lot of businesses lose ground without realizing it. If your website says one thing about your service area and your GBP listing says another, the model has two conflicting signals and often just picks the more frequently repeated one, which might not be yours.

Run the same batch of buyer questions your prospects would actually type, across each engine you're tracking, and read who gets named alongside you or instead of you. Do this often enough and a pattern shows up fast: usually one or two competitors keep appearing because they've been specific about something you've left vague, like turnaround time, licensing, or a niche they've claimed.

Tools like Profound, Otterly, and the AI visibility features inside Semrush all do some version of this competitive scan. The differences come down to how granular the model breakdown is and whether the tool stops at reporting or actually helps you close the gap. A report that tells you competitor X shows up 40% more often in Perplexity is useful information. It doesn't fix anything by itself.

Knowing your share of voice is a diagnostic, not a strategy. The strategy is closing the specific gaps the diagnostic finds, and that's where most businesses stall out, because writing a dozen fact-dense service pages and getting the technical files right, things like llms.txt, robots.txt, and schema, is a project most owners don't have bandwidth for on top of running the business.

That's the reasoning behind our DIFM add-on. For $399 a month on top of a Pro or Enterprise plan, OMG auto-publishes the technical groundwork like llms.txt and robots.txt directly, and queues blog and FAQ drafts addressing the specific gaps your share-of-voice tracking surfaced, ready for your approval before anything goes live. You're not staring at a dashboard wondering what to do with the number. You get draft content built around exactly what's missing.

If you manage AI visibility for multiple clients, a single share-of-voice number per client isn't enough to justify a retainer, and clients notice when a report doesn't explain why the number moved. OMG's agency tier supports white-label reporting across multiple client accounts, so you can hand a client a breakdown by engine, by competitor, and by the specific content gap driving the difference, under your own agency's branding rather than ours.

The businesses that get real value from tracking this aren't the ones checking a dashboard once a month. They're the ones treating a dip in one engine as a prompt to go fix a specific, findable gap, then checking again in a few weeks to see if it moved.