Share of Voice in AI: How to Measure What Competitors Can't Hide

Share of voice (SOV) has always been a marketing metric: what share of the conversation about your category do you own? In the paid-media era it meant ad spend. In the search era it meant how often you appeared on page one.

In AI answers, SOV gets sharper — and more uncomfortable. Because an AI answer is a recommendation, not a list. When someone asks "best CRM for a 50-person sales team," the engine doesn't show ten results and let the user choose. It names a few. The question is no longer "are we in the conversation?" It's "when the machine makes the call, does it make it for us?"

What AI share of voice actually measures

For a given prompt, an AI answer can contain zero, one, or several brands. SOV in this context has three layers:

1. Presence — is your brand in the answer at all? (Mention rate: X of Y prompts.)
2. Position — when you're present, where? First-named brands get disproportionate attention; the first item in a list is the one people remember.
3. Citation — does the answer link to your site, or to a third party's?

The uncomfortable part: these three can diverge. You might be mentioned in 80% of answers but cited in 10% — the engine knows your brand but sends readers to a comparison site. Or you might be cited for documentation-style questions while your competitor owns the recommendation prompts.

Why competitors can't hide it

Traditional SOV measurement has a blind spot: you can see your own ads and your own rankings, but you estimate the rest. AI answers invert this. The answer is public text — anyone can ask the same prompt and read the same answer. That means:

• Your competitor's SOV is observable. Ask the prompt, count the names.
• The engine's preference is stable enough to track. Ask it daily and the pattern emerges.
• Changes are detectable in days, not quarters. A new comparison page on a competitor's site can shift an answer within a week.

This is the same logic behind Citedly's tracking: a fixed prompt set, asked across engines on a schedule, with mention/citation/position recorded per answer. The delta between this week and last week is your SOV movement.

How to build your own SOV baseline

You don't need a tool to start — you need a prompt set and a spreadsheet.

1. Write 15–30 prompts. Mix: category recommendations ("best X for Y"), competitor comparisons ("X vs Y"), and problem queries ("how to solve Z"). Use the words your customers actually type.
2. Ask each prompt in 2–3 engines. ChatGPT, Perplexity, and Gemini behave differently — an SOV number without an engine label is meaningless.
3. Record per answer: brands mentioned, order, links present, and a one-line sentiment note.
4. Repeat weekly for a month. Then compute: mention rate per engine, average position, citation rate, and your share of named brands across all answers.

After a month you'll have something no competitor can fake: a baseline. Everything you do next — content, schema, llms.txt, comparison pages — gets measured against it.

Where to go from a manual baseline

Manual tracking works until it doesn't: 30 prompts × 4 engines × daily is 120 answers a day, and nobody wants to read 120 answers by hand. That's the job Citedly automates — daily prompt runs across engines, mention/citation/position tracking, share-of-voice against the competitor brands you name, and alerts when your position drops.

Start with the free audit to see the deterministic side of your visibility, then set up tracking on a paid plan when you want the numbers to keep moving.