AI Share of Voice (SOV) Tracking: A Complete Guide for Marketers

Traditional Share of Voice measured how much of the advertising airtime in your category your brand owned. AI Share of Voice (AI SOV) measures something more important in 2026: how often ChatGPT, Claude, Perplexity, and Gemini mention your brand when a customer asks a category question.

It's the single most important marketing metric nobody tracked in 2024 and everyone will track by 2027. This guide covers the methodology, the math, the common mistakes, and how to tie SOV to actual pipeline.

Why AI SOV Is the Metric That Matters

Three reasons traditional marketing metrics don't capture the AI-search shift:

• AI recommendations carry more trust. When ChatGPT says "we recommend X," the user doesn't browse 10 alternatives — they often act on the recommendation. Conversion at 14.4% vs 2.5% for organic search proves this.
• Traffic metrics miss the channel entirely. AI queries don't always produce clicks. A user can ask ChatGPT about your brand, get their answer, and buy direct — your GA4 says "Direct" traffic. SOV catches the upstream signal.
• Ranking tools miss it too. Only 12% of URLs AI cites overlap with Google's top 10 results. Your Semrush rankings tell you nothing about AI visibility.

SOV is the metric designed for the reality that most brand discovery happens inside a conversation with an AI assistant.

The SOV Formula

The math is simple. The discipline is in the sampling.

Basic SOV

Run 50 category-intent prompts across 4 AI engines. Your brand appears in 18 responses. Your basic SOV is 18/50 × 100 = 36.

Weighted SOV (Preferred)

Basic SOV treats all mentions as equal. But an engine with 300M weekly users matters more than one with 3M; a brand that's the first recommendation matters more than one that's the seventh. Weighted SOV corrects for both:

A common weighting scheme:

And position weights within a given answer:

This produces a 0-100 score where being the first-mentioned brand in a ChatGPT answer contributes far more than a passing mention in Claude.

How to Pick the Right Prompts

Your prompt set should reflect how real customers ask AI about your category. Not how your SEO team writes title tags. Four prompt categories to include:

1. Category-Intent Prompts (60-70% of your set)

The user knows they want something in your category but not which brand. These are the gold-standard SOV prompts.

Examples for a project management SaaS:

• "Best project management tool for a 10-person startup"
• "What's the most affordable project management software?"
• "Project management tools that integrate with Slack"
• "Top project management tools for marketing agencies"

2. Comparison Prompts (15-20%)

The user is deciding between two specific brands. These test how you fare in head-to-head.

• "Asana vs Monday vs ClickUp"
• "Is Notion better than Asana for project management?"
• "[Your brand] vs [top competitor]"

3. Problem Prompts (10-15%)

The user describes a pain point, not a product category. Tests whether AI surfaces your brand for adjacent intent.

• "How do I track project deadlines across multiple teams?"
• "Our team is bad at communicating project status — what tool should we use?"

4. Brand Validation Prompts (5-10%, lower priority)

The user already knows your brand name. Tests how AI describes you.

• "Tell me about [your brand]"
• "Is [your brand] a good choice for [use case]?"

Brand validation prompts are the least predictive of SOV lift because the brand is already named in the query. Useful for reputation monitoring, not for tracking visibility.

Benchmarking: What's a Good SOV Score?

No universal benchmark — it depends on category. Rough guidelines from what we've seen:

The more concentrated the category, the higher the top SOV scores. Horizontal B2B SaaS (CRM, project management, video conferencing) has a handful of winners — each takes 60-80. Fragmented categories (DTC apparel, local services) have no single dominant brand and the top SOV is lower.

Your absolute SOV matters less than the delta between you and your named competitors — and your trajectory over 3-6 months.

Tracking SOV Over Time: The Discipline

A monthly tracking process that works:

Month 1 — Baseline

1. Lock in your prompt set (25-50 prompts, distributed across the 4 categories above)
2. Lock in your competitor set (3-5 named competitors)
3. Lock in your engines (typically ChatGPT, Claude, Perplexity, Gemini)
4. Run the scan. Record SOV for you + each competitor.
5. Record per-engine breakdown — often you're strong on 1-2 engines and weak on 1-2.

Months 2-12 — Same Prompts, Re-Run

The prompt set shouldn't change. If it changes, you're measuring a moving target and can't compare month over month.

Every month, capture:

• Your SOV (absolute)
• Delta vs last month (direction is what stakeholders care about)
• Competitor SOV for each named competitor
• Per-engine breakdown + any outlier moves
• The citation sources ChatGPT/Perplexity used for your mentions — these are your highest-leverage URLs

Quarterly — Prompt Audit

Once a quarter, review whether the prompt set still reflects how customers talk about the category. New use cases, new intent phrases, new comparison sets. Add 5-10 new prompts. Retire the stalest 5-10. This keeps the set current without destroying month-over-month comparability entirely (keep the 70% core stable).

Tying SOV to Pipeline

The #1 question from CFOs: "Does SOV correlate with revenue?"

Our position: yes, but with a lag, and triangulated. Four patterns we've observed:

1. Direct/branded search lift. ChatGPT users often double-check an AI recommendation by Googling the brand. A 10-point SOV lift typically produces a 15-25% increase in branded-search impressions within 30-60 days.
2. "Direct" traffic in GA4. Because AI referrals often come through without a clean utm_source, they show up as Direct. Sustained SOV improvement → sustained direct-traffic lift to the matching pages.
3. Self-reported attribution. Add "How did you hear about us?" to every conversion form. An explicit "ChatGPT / AI" option catches what no technical attribution can.
4. Sales call anecdata. Track how many sales calls start with "I saw you recommended by ChatGPT when I asked about..." In early-stage GEO, this is qualitative but very telling.

SOV is a leading indicator, not a revenue-accurate one. Treat it like brand awareness from the pre-digital era: you know it matters even if you can't attribute every sale to a specific mention.

The 5 Most Common SOV Tracking Mistakes

1. Running the same prompt twice in one day and assuming the result is stable. AI responses vary across sessions; take the aggregate of several runs, not a single snapshot.
2. Only tracking ChatGPT. It's the biggest single engine, but Claude, Perplexity, and Gemini behave very differently. Missing 3 of 4 engines means missing 55-75% of AI traffic.
3. Changing the prompt set quarterly. Kills comparability. Lock the core, iterate the edges only.
4. Not tracking competitors. Absolute SOV means less than relative. A +5 jump while competitors jumped +10 is a loss, even though your number went up.
5. Chasing a single-month regression. One bad month often reflects an engine update, not a real regression. Use 3-month rolling averages for signal, and investigate only if two consecutive months both regress.

Setting Up SOV Tracking in 30 Minutes

Fast path:

1. Identify your category. Write down 25 prompts an ideal customer might ask.
2. Name your top 5 competitors.
3. Run a tool like Teehoo Martech's free Brand Scan across all 25 prompts × 4 engines. ~2 minutes.
4. Record the SOV + per-engine scores + competitor deltas in a spreadsheet. ~10 minutes.
5. Set a calendar reminder for the same day next month. Same prompts. Same tool. Re-run.

That's it. The discipline is in doing it every month for 6 months — the first month is a baseline, months 2-6 are where the story emerges.

FAQ