FAQ
What is AI share of voice and how is it calculated?

AI share of voice is the proportion of AI answers, across a fixed set of prompts and runs, that name your brand compared with the competitors named in the same answers. The questions below cover how we calculate it, how reliable a single check is, and how to read the number once you have it.

What the metric is
What is AI share of voice?
AI share of voice is the share of AI-generated answers that mention your brand, measured against a defined set of prompts. There is no "AI ranking position" in answer engines. There is being named, or not, which is why we use share of voice, prompts won and "cited as the source" as our measurement language. Every report shows the same defined prompts, every cycle, every engine, with screenshot evidence of where the client appears and who appears instead.
Why does it matter more than a ranking position?
AI share of voice matters because the click has largely disappeared from the journey, so being named inside the answer is the outcome that counts. Around 68% of Google searches now end without a click, and under 1% of those end up on a website. Rankings and impressions can hold steady while sessions fall, so a rank report tells you little about whether prospects meet your brand.

How the calculation works
How do you calculate AI share of voice?
You calculate it by dividing the answers that name your brand by the total answers sampled across a fixed prompt set. The method that makes the figure defensible is consistency. We run the same defined prompts every cycle on every engine, so the number can be compared like for like. Our sampling runs at around 1,500 sampled answers per source per cycle across 30 prompts.
Why does the prompt set have to stay fixed?
A fixed prompt set keeps changes in the score attributable to real shifts in visibility rather than to changes in what you asked. If you swap prompts between cycles, you are measuring a different question each time. The same logic applies to engines: results differ across ChatGPT, Gemini, Perplexity and the rest, so each engine is tracked in its own right.
Is it calculated per engine or combined?
We report it per engine and in combination. The per-engine view shows where you are named and where a competitor is named instead, and it matters because the engines behave differently from one another. For our own business, the priority engines are ChatGPT, Google AI and Perplexity. Any claim should state which platforms it covers, because a ChatGPT result says nothing certain about Gemini.
How many runs do you need before the number means anything?
You need a large sample, because a single check is not reliable. A single check carries ±9 to 13 points of uncertainty, which is why one screenshot proves nothing. At around 1,500 sampled answers per source per cycle, with 30 prompts and Wilson confidence intervals, the published margin of error is roughly ±1 to 2 points.

Reading the number
What should a baseline report include?
A baseline states three things: the figure, the period and the margin. A bare percentage with no period and no margin cannot be compared with next month's number. We apply this discipline to ourselves, running a monthly own-name visibility tracker with 22 defined prompts, screenshot evidence and first-answer counts.
Is a trend more useful than a single score?
Yes. A single score is a snapshot with a margin of error, while a trend across repeated cycles on the same prompts shows whether your structuring and content work is changing what the engines say. Because every cycle uses the same prompts and engines, movement between cycles reflects real change rather than noise.
Can I check ChatGPT by asking it a few questions myself?
You can, but the result only tells you what one answer said on one day. That is why we do not treat one answer as evidence of where you stand. A manual check is a useful starting prompt, and the measurement becomes meaningful once you repeat it across a fixed set of prompts and runs.

Making the number actionable
Does a high share of voice produce enquiries?
Share of voice shows visibility, and connecting it to enquiries needs separate attribution. Attribution tracking requires a "How did you hear about us?" enquiry field with individual AI engines listed. Without that field, a rising share of voice and a rising enquiry count remain two unlinked numbers.
What do I do if my share of voice is low?
Start with access, then structure. Roughly 22% of web traffic sits behind Cloudflare, and content behind a default block is invisible to AI engines however well it is structured, so verify crawler access first. After that, put a complete, standalone answer in the opening sentence under each heading, because 44.2% of LLM citations come from the first 30% of a page. A dashboard can report the gap, but closing it takes content and structural changes you ship yourself.
To put numbers on it, read how we measure AI visibility.


