Share of Voice in AI Answers: The Metric Your CMO Doesn't Have Yet

The pipeline loss you cannot see

Every analytics tool you own is built on a click. Sessions, sources, conversions, they all assume someone clicked through to your site, leaving a trace. AI-mediated discovery breaks that assumption. A buyer asks an assistant for a recommendation, the assistant names three brands, the buyer goes straight to one of them, and if you were not named, there is no referral, no impression, no log line. The loss is real and it is also completely invisible to your stack.

This is the structural blind spot of the AI era. Rank trackers tell you where you sit in ten blue links, but with AI Overviews in roughly half of Google searches and hundreds of millions of people asking ChatGPT directly, the question that actually decides pipeline has shifted from "do we rank?" to "do we get named?" Those are different questions with different answers, and only one of them is on your CMO's dashboard.

What AI Share of Voice actually measures

AI Share of Voice (SOV) is straightforward in concept: across a representative set of prompts your buyers would ask, what percentage of AI answers mention your brand? Per Alex Birkett's framework for measuring AI SOV, if your brand appears in 300 of 1,500 responses across ChatGPT, Gemini, and Perplexity, your AI Share of Voice is 20%. But the headline number hides two distinct things worth tracking separately:

• Entity-based SOV, how often your brand is named as a recommendation or option in the answer itself. This is the "did the model say our name" metric.
• Citation-based SOV, how often your content is cited as a source the answer drew from. This is the "did the model use our page" metric, and it connects directly to your AEO work.

The two can diverge sharply. You might be cited as a source (your content is good) but not named as a recommended option (the model recommends competitors and merely uses your comparison article). Or named but not cited (you have brand mentions across the web but the model is not pulling from your own pages). Tracking both tells you whether your problem is content the model can use, or positioning the model believes.

The four signals worth tracking

A useful AI SOV program goes beyond a single percentage. Drawing on the measurement approaches in Siftly's 2026 SOV guide and the tooling landscape, four signals matter:

• Mention rate, what share of relevant prompts name you at all.
• Share of voice vs. competitors, your mentions as a fraction of all brand mentions on the same prompts. This is the competitive number a CMO understands instantly.
• Sentiment and accuracy, when the model describes you, is it positive, and is it correct? A confident wrong description ("they're the budget option" when you are premium) is worse than absence.
• Position and prominence, are you named first, or buried in a list of seven? Are you the recommendation or an also-ran?

How to measure it

The mechanics are conceptually simple: assemble a representative prompt set, query the AI platforms repeatedly (answers vary run to run, so you sample), and parse each response for brand mentions, competitors, citations, and sentiment. The variance is real, the same prompt yields different answers, which is exactly why this needs systematic sampling rather than a one-off check.

# Skeleton of an AI Share-of-Voice measurement run
import re

PROMPTS = [
    "What's the best AI quality testing platform?",
    "How do I load test an LLM API?",
    "Tools to scan a website for accessibility and SEO issues",
    # ...a representative set of buyer-intent queries
]
BRAND = "alt.qa"
COMPETITORS = ["CompetitorA", "CompetitorB", "CompetitorC"]
SAMPLES = 5  # repeat each prompt; answers vary run to run

def analyze(answer):
    named = bool(re.search(rf'\b{re.escape(BRAND)}\b', answer, re.I))
    comp_hits = [c for c in COMPETITORS
                 if re.search(rf'\b{re.escape(c)}\b', answer, re.I)]
    # position: where in the answer is the first brand mention?
    pos = answer.lower().find(BRAND.lower())
    return {'named': named, 'competitors': comp_hits, 'first_pos': pos}

results = []
for prompt in PROMPTS:
    for _ in range(SAMPLES):
        ans = query_ai_platform(prompt)      # ChatGPT / Perplexity / Gemini
        results.append(analyze(ans))

total = len(results)
mention_rate = sum(r['named'] for r in results) / total
all_brand_mentions = sum(r['named'] for r in results) + \
    sum(len(r['competitors']) for r in results)
sov = sum(r['named'] for r in results) / max(all_brand_mentions, 1)
print(f'Mention rate: {mention_rate:.0%}   Share of voice: {sov:.0%}')
# Track these over time and per platform, the trend is the signal.

Several commercial platforms now do this at scale, Profound, Otterly, Siftly, and others track brand mentions across ChatGPT, Perplexity, and Gemini, and even HubSpot has added AEO visibility tracking with sentiment and competitive benchmarking. Whether you build a lightweight version or buy one, the point is the same: instrument the answers, because they are now part of the funnel and nothing else measures them.

From metric to action

Measurement is only useful if it drives work. A low or declining AI SOV points back at fixable causes, most of which are the AEO fundamentals: can the answer engines crawl and render your pages, are you allowing the right AI crawlers, is your structured data valid, is your content fresh enough to clear the recency gate, and do your pages actually contain the direct, citable answers to buyer questions? AI SOV is the outcome metric; the AEO scan findings are the inputs you change to move it. Track the metric, diagnose with the scan, fix the inputs, and watch the trend, per-platform, over time, against named competitors.

The bottom line

The highest-intent discovery moment in modern marketing, a buyer asking an AI for a recommendation, produces no analytics event, which means most companies are losing pipeline they literally cannot see. AI Share of Voice closes the blind spot: track mention rate, competitive share, sentiment, accuracy, and position across ChatGPT, Perplexity, and Gemini using a sampled prompt set, because answers vary run to run. Then connect the metric to its causes, crawlability, structured data, freshness, and citable content, and work the inputs. Your competitors are starting to build this dashboard. The CMO who has it knows whether the brand is winning the new front page; the one who does not is flying blind on the channel that increasingly decides the click.