How can a cybersecurity vendor improve its share of voice in AI assistants?

A cybersecurity vendor can improve its share of voice in AI assistants by making the public evidence that connects it to a buyer's question easier for each assistant to find, read and cite, and by measuring the same questions again to see whether it was named more often. First, though, the phrase needs a definition. "Share of voice" comes from advertising and search, where it means your share of the visible space. An AI answer has no fixed space to divide, so the phrase can mean several different things, and a number labelled with it is only useful if you know which one it is.

On this page we explain what we count in our UK B2B technology study, show the figures for the two categories that serve security buyers, and set out what a security vendor can do. We do not call any of our figures a share of voice, and nothing here judges a vendor's product.

What "share of voice" can mean in AI answers

Three readings are in common use, and they are not the same.

1. How often you are named. Across a set of buyer questions, in how many answers does your name appear? This is a count of mentions and it is the simplest to read.
2. How much of each answer you take. Some measures weight a name by the role the answer gives it, such as the one put forward as the fit, an alternative, or a name that only sits in a list. Our platform has a measure of this kind, a weighted share of recommendation, and its definition is on our methodology page. The study does not use it.
3. How often your pages are cited. A page of yours can be used as a source without your name appearing in the answer text. That is a citation, a different outcome. Mention, citation and qualified recommendation shows how the three outcomes differ and why we count each on its own base.

The first reading is the most direct answer to how visible a vendor is in AI assistants, so that is what the study reports. It stays a count of how often a vendor is named. It is not a share of anything, and it is not a measure of recommendation.

What our study counts

Our UK B2B technology study put the same buyer questions to three assistants through their APIs, each with its own web search: Gemini with Google Search, GPT through Azure OpenAI with Bing search, and Perplexity through its Sonar Pro model. Each category has a fixed list of vendors, chosen before any answer was collected. For every answer, we recorded which listed vendors it named.

A vendor counts as named when the answer names its product, one of its listed aliases, or an edition or plan of it. A company named only as an owner or parent does not count.

Each figure is written as named in N of M answers: M is the number of that assistant's answers for the category that were stored and classified, and N is how many of them named the vendor. Every figure belongs to one assistant. We do not blend the three into one number, because the assistants search differently and answer differently.

Every figure also carries an interval, a range that says how high or low the true rate could plausibly be. Answers to one question tend to agree with each other, so the interval resamples whole questions rather than single answers. Reading an AI visibility figure honestly explains why we show figures this way.

The security figures

Two categories in the study serve security buyers. The first covers security awareness and human risk: the training and phishing-simulation platforms a company buys to reduce risk from its own staff. The second covers identity and access management: single sign-on, multi-factor authentication and identity governance for a workforce.

What these figures show

Read the two tables as you would read any small sample, as a direction.

• A small group of vendors tends to be named in nearly every answer, and the rest far less often. That is the pattern the study reports across most categories, and it is visible in each table above.
• The assistants do not agree. A vendor can be named in most of one assistant's answers and in few of another's, so one assistant's figure says little about the others. A security vendor should read the three columns separately.
• The intervals are wide. Each figure rests on the answers to one category's questions, and two vendors whose intervals overlap may not differ at all.

What these figures do not show

• They are not about recommendation. An answer can name a vendor to recommend it, to compare it or to warn against it. The study counts only that it was named.
• They are not about the consumer apps. These are API answers, and the apps can answer differently because they carry a person's history, location and settings. We have not yet compared these figures with answers from the apps.
• They are a snapshot. Assistants change their models and their search often, so the same questions asked later may name different vendors.
• They cover the tracked list only. A vendor outside a category's list is not counted, however often it is named.
• They say nothing about a product. Being named often is not the same as being the right choice, and the study makes no judgement of any vendor's product.

What a security vendor can do

Nobody outside an assistant's provider controls what it names. What a vendor controls is the evidence an answer can draw on, and the order of work is the same as for any B2B technology company. We cover each step in a guide of its own and do not repeat them here.

1. Write down the questions your buyers ask, and keep them fixed. Security buyers tend to arrive with constraints: a framework they must meet, such as Cyber Essentials or ISO 27001, a size of company, a way of working. An answer can only connect you to a constraint through something public. Building a buyer-question set shows how to collect the wordings and freeze them, so that next month's measurement compares like with like.

2. Let each assistant's search reach your site, and publish pages that answer the question. How a B2B tech company gets recommended more often by ChatGPT and Claude covers the search crawlers each company documents, which are separate from its training crawlers, and the pages that make a claim easy to find. For a security vendor, the pages that state what the product covers, which frameworks it supports and which integrations it has are the ones a buyer's question usually turns on.

3. Trace the evidence behind an answer that names someone else. Pick one answer for a question that matters to you, read its citations, sort each source by kind and owner, and compare what the named vendor has with what you have. What evidence AI assistants rely on when they recommend B2B vendors sets out that trace step by step.

4. Turn the gaps into briefs. Rank the gaps by what they cost you in deals, and write one brief for each. Turning AI visibility data into a content plan shows how.

5. Measure the same questions again, on each assistant separately. A rise in how often you are named, with its counts and its interval, is the only honest sign that the work moved anything. A single answer proves nothing.

Read the study, or request the report

The findings page sets out every figure in every category, the method and its limits. The full report is free with your work email.

If you would rather have the measurement run for you, our services page lays out the steps, from a first read of your own buyer questions to monthly measurement.