AI share of voice: measure your brand in AI answers (2026)

AI share of voice is your brand's proportion of brand mentions across a defined set of AI answers, compared with the other brands counted. Divide your mentions by the total, then multiply by 100. Inspect that total first. Fix the questions, competitors, engines and markets before comparing periods. Otherwise, a rising percentage can hide unchanged visibility.

I'm Brenden House. I run Searchmaxxed, an SEO and AI SEO agency. This guide explains how to build a number your marketing team can defend, and how to turn the gaps into website work.

In summary

• Count each brand once per answer. Repeating a name doesn't create another recommendation.
• Report share of voice beside mention rate. One compares brands; the other measures how often your brand appears.
• Separate Australia from the US, and ChatGPT from other engines. Combine them only with stated weights.
• Keep the original answers. A chart without its questions, dates and counting rules is difficult to check.
• Track enquiries separately. AI share of voice measures sampled visibility, not revenue or market share.

How do you calculate AI share of voice correctly?

Calculate AI share of voice by dividing your brand-answer appearances by all eligible brand-answer appearances in the same sample. Count each brand once per answer.

That last rule matters. An answer naming your software repeatedly should not outweigh another answer recommending a competitor alongside several alternatives to the same prospective buyer.

AI share of voice = your brand-answer appearances ÷ all eligible brand-answer appearances × 100.

An eligible brand is one that meets your written category rules. A brand-answer appearance means that eligible brand occurs in an answer's own text, regardless of repeated mentions.

Count the answer itself. Keep source-panel links separate.

Here's an illustrative B2B SaaS panel, calculated for this guide on 27 October 2026. These aren't client results.

The complete panel contains 100 answers collected for this hypothetical comparison. Some name several brands; others name none. Your company appears in 30% of answers but receives 25% of the brand appearances.

Different denominators, different questions.

Semrush's measurement guide gives a mentions-based formula, while its product scores also incorporate prominence. Its guide was published on 17 July 2026 and checked on 27 October 2026.

Ask what your score counts. The label alone won't tell you.

How is share of voice different from AI visibility?

AI share of voice compares your presence with other brands; mention rate measures how often you appear in the sampled answers. AI visibility is broader.

Keep the measures separate:

Multiply the rate calculations by 100 for percentages. Define “recommending” before tagging answers, then inspect ambiguous cases manually rather than accepting every automated label.

A mention can be negative. A citation can be unrelated to choosing your product. Neither is automatically a recommendation.

For Searchmaxxed, those distinctions change the work. Missing mentions suggest a different problem from an answer naming your company but describing a product you stopped selling.

Can your share rise without your brand becoming more visible?

Yes, your AI share of voice can rise when competitors lose mentions, even if your own mention rate stays flat. The denominator explains it.

Return to the illustrative SaaS company. Keep its 100-answer panel unchanged across both periods, with the same prompts, engines, markets and counting rules used throughout the comparison.

Share rose by five percentage points. Your brand reached no additional sampled answers.

That's a relative gain, worth knowing about, but it isn't evidence that more buyers saw your company or that any extra enquiries came through.

Read both columns. Searchmaxxed's monthly competitor measurement belongs beside enquiries and website changes, so the report explains what moved rather than celebrating whichever percentage happens to rise.

Which prompts and competitors should you track?

Track the questions your buyers ask about their problems, requirements and alternatives, with a fixed category definition for eligible brands. Separate branded from unbranded prompts.

Start with customer language. Search Console queries, sales questions and product documentation give you material; a keyword list alone won't capture every constraint an enterprise buyer brings to research.

These are example prompts. Replace their assumptions with actual product facts before testing, especially where security, public-sector procurement or regulated manufacturing requirements determine whether you qualify.

Keep branded questions apart. Naming yourself in the prompt makes that answer unsuitable for measuring whether an engine discovers you without being told your name first.

Next, define the competitor set. List direct alternatives, map product names to their parent brands, and decide how to treat acquisitions before the first reporting period starts.

Record unexpected competitors too. A brand outside your sales team's usual shortlist can still compete for the same answer, so preserve it in an “other eligible brands” group.

Don't quietly drop it. If eligibility rules change, recalculate historical results under the new rules or start a clearly labelled series.

Our guide to tracking AI visibility covers the collection workflow. This guide's focus is keeping the resulting comparison honest.

Should Australia, the US and different engines share one score?

Report AI share of voice by market and engine before producing any combined score. Different samples can produce different scores without any change in performance.

Keep these separate initially:

• ChatGPT, with the collection method and search setting recorded.
• Gemini and Perplexity, each with their own answer records.
• Google AI Overviews and AI Mode, treated as separate search experiences.
• Australian and US panels, with market, language and collection settings retained.

The settings matter. OpenAI's search documentation, checked on 27 October 2026, says location and saved memories can influence search queries.

A market named in a prompt isn't proof that the collection location matched. Record both, and keep account history or personalisation consistent wherever the interface allows.

For a neutral baseline, use fresh conversations. Test follow-up conversations separately, with their preceding messages saved, rather than mixing them into the single-question benchmark without a label.

Google needs another distinction. Its AI features documentation, checked on 27 October 2026, says AI Overviews often don't trigger.

No Overview isn't a collection error. Record it separately, report how often Overviews appeared, and state whether your citation rate uses all searches or only searches showing an Overview.

How should you combine the scores?

Use stated, fixed weights if leadership needs a combined score. Keep the underlying rows visible.

An equal-weight average describes the panels you chose, giving each the same influence regardless of its audience size or commercial importance. It doesn't describe audience share.

If the US panel contains more answers, pooling every answer gives it more influence. That may reflect research effort rather than the commercial importance of either market.

Choose the weights deliberately. Document whether they reflect equal treatment, target revenue or customer research, and don't present a business priority as measured AI usage.

How many prompts and repeat runs do you need?

There is no sample size that makes a hand-picked prompt panel representative of every AI conversation. Start with category coverage, then measure repeat-run variation.

Here's a proposed starting design for an enterprise team, rather than an industry benchmark. Adjust it around the products and buying questions that matter commercially.

Those counts are design assumptions. More repeats help assess stability; more distinct questions broaden coverage. Neither fixes a prompt set built around what you already rank for.

Save every answer. Retain prompt ID, full wording, timestamp, market, engine, model when available, collection method, answer text, citations and classification.

Log errors separately. A failed request isn't a brand absence; an answer that genuinely names no brands is still a valid observation for mention-rate reporting.

If no eligible brands appear anywhere in a segment, its share of voice is undefined. Report “no brand mentions”, not a fabricated zero-percent competitive ranking.

How do you tell movement from noise?

Repeat the same panel under comparable conditions and report its observed spread before declaring a small change meaningful. Inspect which questions changed, not just the average.

Three runs don't establish certainty. Closely related prompts and repeated answers aren't independent samples of the entire market, even when the dashboard treats them as separate rows.

For formal uncertainty estimates, have an analyst resample complete prompt groups, retaining their repeats together. Describe those intervals as uncertainty within the panel, not all buyers.

Keep time windows comparable. Record model changes and collection outages, and compare the matched sample if missing data would otherwise change the mix.

What can AI share-of-voice tools miss?

AI monitoring tools can miss context, misclassify brands or conceal changes in their sampling. Require raw answers and a documented calculation before relying on their score.

Check these points first:

• Does the tool collect the consumer interface, an API response or its own search-assisted configuration?
• Can you export answer text, timestamps and source URLs?
• Does it deduplicate repeated brand names?
• Does a parent company absorb its product brands?
• Are source-panel links counted as mentions?
• Can you distinguish recommendations from negative references?
• Does it preserve history when prompts, models or competitor lists change?

Same model name isn't enough. Treat different collection methods as separate series until you've tested whether they produce comparable answers for your own questions.

Searchmaxxed's AI visibility tools guide covers software selection. Here, the important requirement is that another person can reproduce the number from the exported answers.

Why inspect the sources behind the score?

Sources show where an answer gets its supporting information, which helps distinguish a missing website page from missing coverage elsewhere. Our own research illustrates the difference.

Across 900 ChatGPT answers about Australian agencies, a single third-party "best agencies" site was cited in 168 of them. The next three were a second listicle, Semrush's agency directory and Clutch. If you want to be recommended, the job is getting onto the handful of pages the model already trusts.

That was Searchmaxxed's agency-question panel from 2 October 2026. It isn't a SaaS benchmark, and those citation counts aren't evidence that a particular brand was recommended.

The useful question travels. Which sources support the answers in your category, and what do those pages actually say about your company?

Read them before prescribing work. A manufacturer may need accessible product specifications; a software company may need a comparison page that explains an integration accurately rather than merely naming it.

What should an AI share-of-voice board report contain?

A board report should show comparable movement, its likely cause and the action proposed. Pair relative share with absolute coverage, then report commercial outcomes separately.

Keep the headline readable. Put the detailed methodology behind it, with enough information on the page to stop a director mistaking the sample for the whole market.

For the illustrative example, the headline would be: “Share rose from 25% to 30%; mention rate stayed at 30%.” Then explain why.

Competitors appeared less often. That's the story.

Don't turn the five-point increase into a revenue estimate. Use analytics, CRM records and customer feedback to understand enquiries; the answer panel cannot supply missing sales attribution.

How does Searchmaxxed turn missing mentions into work?

Searchmaxxed connects AI-answer observations with technical SEO, content and competitor research, then builds the pages or fixes the evidence gaps. Tracking informs implementation.

Start with the answer. If a competitor appears for a requirement you genuinely meet, we check your product page, supporting documentation and the sources used in that response.

Then check access. Google's guidance requires a page to be indexed and eligible for a snippet before it can appear as a supporting AI-feature link.

We fix crawl and indexing problems, internal links and page structure, then write the missing explanation using product facts your team can verify before publication.

The page earns its detail. For B2B SaaS, that might mean an integration or alternatives page; for manufacturing, a clear explanation of specifications and suitable applications.

Repeat the checks afterwards. A changed page and a changed answer are observations to investigate together, not proof that every movement came from that single edit.

This sits within our AI search optimisation service. You own the website, content and data, and measurement feeds weekly implementation rather than ending at a software export.

Is Searchmaxxed right for your team?

Searchmaxxed fits teams that want SEO and AI-search measurement connected to website implementation. It's less useful when you only need software to monitor answers.

Our managed scopes differ:

Bring your actual requirements. A manufacturing group may need separate product divisions; a venture-backed software company may need Australian and US competitors kept apart in every report.

Public-sector teams should also define approved sources, data handling and publication responsibilities before work begins. Those requirements belong in the agreed scope, not an assumption.

If your internal team already handles research, implementation and measurement well, a tracking tool may be all you need. We aren't selling another dashboard login.

Frequently asked questions

What should you do next?

Start with the buying questions, markets and competitors you need to measure, then decide who will fix the gaps. Searchmaxxed can connect that measurement to implementation.

Bring an existing report if you have one. We'll look at what it counts, what your buyers are asking, and which pages need work.

Let's talk pricing.