What is the formula for AI share of voice?

AI SoV (%) = your weighted mention score divided by the sum of all tracked brands' weighted scores, times 100, where each mention scores position x sentiment x platform weight.

Does a higher AI share of voice guarantee more citations or sales?

No. It measures visibility in your sample and does not promise any outcome from an AI engine.

Knowledge Center

AI Share of Voice: A Replicable Formula for Measuring Brand Visibility Across ChatGPT, Gemini, and Perplexity (2026)

Published November 13, 2026 · AEO Spotlight Editorial

By AEO Spotlight Editorial | Last updated: November 11, 2026

AI share of voice is the percentage of weighted brand visibility your brand earns, out of all weighted brand visibility in a fixed set of prompts run on AI platforms. The replicable formula is: AI SoV (%) = (Your weighted mention score ÷ Sum of weighted mention scores for all tracked brands) × 100. Each mention's score is Position weight × Sentiment weight × Platform weight, summed over every prompt run in your sample. You choose a prompt set, run it on each platform, record which brands appear, score each appearance, add the scores per brand, and divide. Everything below can be done in a spreadsheet with no software.

This guide publishes every weighting choice so you can inspect and change it. No proprietary benchmark, survey, or original dataset is claimed here. The worked example uses clearly labeled illustrative numbers.

TL;DR

Fix a prompt set, a brand list, and a platform list before measuring anything.
Score each brand mention as position × sentiment × platform weight.
Sum scores per brand, then divide by the total across all brands.
Repeat each prompt several times, because AI answers vary between runs.
Keep weights constant between reporting periods so trends are comparable.

AI share of voice measures how often, and how favorably, a brand appears in generated answers relative to named competitors, for a defined set of prompts. Platforms covered usually include ChatGPT, Perplexity, Gemini, Brave AI, and Google AI Overviews. It is a relative metric. A score of 30% means your brand holds 30% of the weighted visibility among the brands you chose to track, nothing more.

It does not measure traffic, revenue, or how many people saw an answer. It also does not describe all of AI search, only your sample. In my experience, the biggest source of confusion is treating a sample result as a market-wide fact. Your prompt set defines the market you are measuring, so choose it deliberately.

Platform behavior also differs. Some assistants show citations and links with answers and some may answer from model knowledge alone, so check each platform's current help pages before deciding what counts as a mention in your method. Record that decision in writing.

The formula has two layers.

Mention score: S = P × T × W

P = position weight
T = sentiment or recommendation weight
W = platform weight

Brand total: Total_brand = Σ S over every run where the brand appears.

Share of voice: AI SoV_brand = Total_brand ÷ Σ Total_all_brands × 100

In a spreadsheet, build one row per brand mention per run, with columns for prompt ID, platform, run number, brand, P, T, W, and S. A pivot table by brand then gives you the totals. A printable formula reference sheet and a template with these columns would be a natural companion to this guide; build yours from the column list above so the logic stays visible.

Step 1: Define Your Prompt Set and Platform Matrix

List the prompts a real buyer would type, covering category, comparison, and problem-based questions. Write them once and freeze them. Then list the platforms (for example ChatGPT, Gemini, Perplexity) and the brands, including yours. Your matrix is prompts × platforms × runs. Running each prompt more than once per platform helps smooth out variation in generated answers. Pick a number of runs and keep it constant.

Step 2: Count Brand Mentions per Prompt Run

For each run, note every tracked brand that appears by name. Decide up front how to handle edge cases: a brand appearing only in a cited link, a product name without the company name, or a brand named only as something to avoid. Write the rule down. If you only want raw mention share, stop here and divide your brand's mention count by total mentions. That simpler version is useful as a baseline.

Step 3: Apply Weighting Factors (Position, Sentiment, Platform Reach)

Here are the default weights used in the example below. They are my illustrative choices, not an industry standard, so change them if your situation warrants.

Factor	Condition	Weight
Position (P)	First brand named	1.0
	Second or third	0.7
	Fourth or later	0.4
Sentiment (T)	Recommended or favorable	1.0
	Neutral listing	0.6
	Negative or cautioned	0.1
Platform (W)	Set by your own audience mix	e.g. 1.0 baseline

Platform weight is the least objective. If you have no basis for differences, set every platform to 1.0 and add weights only when you can justify them from your own analytics.

Sum each brand's scores, add all brand totals together, and divide each brand's total by that grand total. The shares across brands then add to 100%. Normalizing keeps the metric relative, which is the point of share of voice.

How Do You Calculate It? A Worked Example for Three Brands Across Two Platforms

Illustrative data only. These numbers are invented for demonstration and are not real measurements.

Setup: 2 prompts, 2 platforms (Platform A, Platform B), 1 run each, so 4 runs. Brands: Alpha, Beta, Gamma. All W = 1.0.

Run	Brand	Position	Sentiment	P	T	S
1 (Prompt 1, A)	Alpha	1st	Favorable	1.0	1.0	1.00
1	Beta	2nd	Neutral	0.7	0.6	0.42
2 (Prompt 2, A)	Beta	1st	Favorable	1.0	1.0	1.00
2	Alpha	2nd	Neutral	0.7	0.6	0.42
2	Gamma	3rd	Neutral	0.7	0.6	0.42
3 (Prompt 1, B)	Alpha	1st	Favorable	1.0	1.0	1.00
3	Gamma	4th	Neutral	0.4	0.6	0.24
4 (Prompt 2, B)	Gamma	1st	Favorable	1.0	1.0	1.00
4	Beta	2nd	Neutral	0.7	0.6	0.42

Totals:

Alpha: 1.00 + 0.42 + 1.00 = 2.42
Beta: 0.42 + 1.00 + 0.42 = 1.84
Gamma: 0.42 + 0.24 + 1.00 = 1.66
Grand total: 5.92

AI SoV:

Alpha: 2.42 ÷ 5.92 = 40.9%
Beta: 1.84 ÷ 5.92 = 31.1%
Gamma: 1.66 ÷ 5.92 = 28.0%

The shares sum to 100%. Raw mention share would give Alpha 3 of 8 mentions (37.5%), so weighting moved the result by a few points here.

Mention Position and Contextual Prominence

Being named first in an answer generally reads as a stronger endorsement than appearing in a long list. Position weights encode that judgment. Whether the answer is a ranked list or prose affects how you assign position, so state your rule. A brand mentioned in the opening sentence of a prose answer can be treated as position 1.

Sentiment and Recommendation Strength

A mention that says a brand is a good fit for the user's need is different from one that lists it with a caveat. Score recommendation strength with a small scale, and have the same person (or a written rubric) do the scoring so results stay consistent. Subjective scoring is the weakest link in any manual method, so keep the scale coarse.

Platform and Prompt-Set Coverage

Platform weights should reflect where your audience actually asks questions, and you should be able to defend them. Prompt coverage matters as much. A prompt set heavy on one product line will inflate scores for brands strong in that line. Balance prompts across the topics you care about, and report results per platform as well as combined, since ChatGPT, Gemini, Perplexity, Brave AI, and Google AI Overviews can surface different brands for the same question.

Changing the prompt set between periods. Trends then reflect your edits, not the market.
Single runs. One answer is one sample of a variable output.
Leaving competitors out. Share needs a complete denominator; omitting a major rival inflates your number.
Mixing weighting schemes. Freeze weights and log any change in a changelog.
Counting branded prompts. Prompts that name your brand guarantee a mention and skew the result. Use unbranded category and problem prompts.
Ignoring negative mentions. Without sentiment weighting, a warning counts the same as a recommendation.
Overstating precision. With a small sample, treat differences of a point or two as noise.

The manual method works for a few dozen prompts. Beyond that, repeating runs across platforms, logging answers, and scoring mentions becomes slow. AEO Spotlight is an online brand monitoring and content creation platform that tracks brand visibility across AI platforms including Gemini, ChatGPT, Perplexity, and Brave AI. You can learn how it handles prompt tracking and view share-of-voice dashboards to apply the same formula at scale. Check that any tool you use lets you see and set the weights, so the arithmetic above stays reproducible.

Whichever route you take, keep the manual version as an audit check: run a small subset by hand and confirm the tool's numbers match.

AI SoV (%) = your weighted mention score ÷ the sum of all tracked brands' weighted scores × 100, where each mention scores position × sentiment × platform weight.

Mention share counts appearances only. AI share of voice adds weights for position, sentiment, and platform, so it reflects prominence and favorability as well as frequency.

How many prompts do I need?

There is no verified universal number. Use enough prompts to cover your main topics and repeat each several times. Larger samples reduce noise, and you should state your sample size whenever you report a score.

Should I weight platforms differently?

Only if you can justify it with your own audience data. Otherwise set all platform weights to 1.0 and report each platform separately.

Does Google AI Overviews count?

Yes, you can include it as another platform in the matrix, with the same mention rules applied to its generated summaries.

How often should I recalculate?

Choose a fixed cadence, such as monthly, and keep prompts and weights unchanged so periods are comparable.

No. It is a measurement of visibility in your sample, and it does not promise any outcome from an AI engine.