Best AI visibility tool for a cybersecurity vendor, and the security review you will run on it

A cybersecurity vendor is the only buyer of an AI visibility tool who will put the tool through a security review before signing. That changes the shortlist. The criterion worth buying on is what each vendor publishes that you can verify without asking a salesperson, and for you that includes its own security posture, its sub-processors and where it stores the answer text it collects. Read on 20 October 2026, three of the six vendors we checked publish something on that; three publish nothing we could find, and our own domain publishes least of all. That last sentence is the point of the page.

Start with the review your own team will run

An AI visibility tool receives a list of the prompts you care about, which is a map of your commercial anxieties, and it stores the full text of what assistants say about you and your competitors. In your company, that is a vendor assessment, a data processing agreement and a question about sub-processors. Ask for those three before a demo, not after.

Of the vendor pages we read on 20 October 2026, Profound and Scrunch name a recognised security certification on their own pages, and Peec publishes a trust page and a changelog. Otterly's pricing page names enterprise grade compliance and single sign on at its top tier. Semrush's AI visibility feature page and the Ahrefs Brand Radar page carry no security statement in the document we read, which is a statement about those pages rather than about those companies. AI Knows Us publishes no trust centre, no status page and no sub-processor list as of 20 October 2026. On the narrow question of published security posture, four rivals are ahead of us and we are not going to write a page that hides it.

Where we do rank first, and on what

On disclosed method and published failure. We publish the question bank, the number of readings, the engines behind each reading, the domains cited, and the results that went against us. None of the six vendor documents we read on 20 October 2026 published a sample size, a prompt set, or a negative result about its own product.

Our own numbers, so you can judge the method by what it reports about us: 1,260 readings across ChatGPT and Gemini in India, the United Kingdom and the United States, 20,143 citations across 4,947 distinct domains, and our own name cited in 6 of those 1,260 readings, zero of them in the United States and zero on Gemini. The exports behind our published claims sit on our proof page with the date they were taken.

What the engines actually do with a security question

In those readings, the question what is the best way for a cybersecurity vendor to appear in AI answers named Semrush in four readings, Profound in three, Otterly in three, Goodie in one and Peec in one. Our own name appeared in none. A separate question about what a security firm should do named Otterly twice, Profound once, Peec once and Semrush once.

Those counts tell you something practical. The engines answer a security vendor's visibility question with general marketing tools, not with security specialists, because nobody in this category has published for a security buyer. The page you are reading is our attempt at that, and the measurement is what makes it worth reading rather than the fact that we wrote it.

The surfaces to measure, which are not the usual five

A security buyer's shortlist is assembled from documents a marketing tool does not watch. In our citation data, vendor owned pages carried 82.0% of all citations and review directories carried 0.7%, and in this vertical the vendor owned page that matters is rarely the home page.

• Your trust centre and compliance pages. The pages that state what you are certified for and when the audit was done. These are the pages a procurement question resolves to.
• Advisory and disclosure pages. Your own write-ups of vulnerabilities you found or fixed. These are cited because they are specific and dated.
• Your documentation. One documentation host appeared in 446 of our 1,260 readings. For a product with an agent or a sensor, the install and hardening guides are read more than the brochure.
• Analyst and framework mappings. Pages that map your controls to a named framework, in text rather than in a downloadable chart, because an assistant cannot read the chart.
• Your help centre, if it is hosted. One help host that appeared in 269 of our readings refuses crawlers its own robots file allows. Check what yours returns before you measure anything.

Ask a vendor to show the source list for a prompt on your category and see whether those five kinds of page appear in it. If the list is all news and directories, the tool is reading the wrong half of your site.

The limits that set the tier, for a security vendor

Every vendor on this list meters the same four things, and in your vertical the second one is the one that catches people out.

• Prompt allowance. Security categories fragment into product lines, so the prompt count rises fast: endpoint, identity, cloud posture and managed detection each need their own set.
• Personas. A chief information security officer and a procurement analyst ask different questions and get different answers. One vendor publishes a persona limit on its public page and its entry tier includes one. If you need two, the entry tier does not fit you whatever the headline figure says.
• Engines in the base plan. Several vendors ship a set and sell the rest as add-ons, each priced separately.
• Export and API access. You will want the raw answer text inside your own environment rather than in somebody else's dashboard. Check whether that sits above your tier.

A note on checking vendor pages yourself, because it cost us time. otterly.ai/pricing/ returned 200 to a macOS Safari agent and 403 to a Chrome on Windows agent on 20 October 2026, with the agent string as the only variable. Change agent before you record a refusal. In your own line of work you will recognise the pattern.

Questions a security vendor asks before buying an AI visibility tool

Does sending our prompt list to a vendor create a disclosure risk?

Treat the prompt list as commercially sensitive and scope it accordingly. You can measure a category without naming an unreleased product, and the first month's set rarely needs to.

We are not in an analyst quadrant. Does that end the conversation?

No. In our readings the engines cited vendor owned pages far more than any third party ranking. A dated, specific page about a control you implement is read; an absence from a quadrant is not something an assistant notices.

Should we publish findings from client work to get cited?

Publish the method and the anonymised pattern rather than the client. The specificity is what earns the citation, and the client name adds none of it.

Our product is sold through partners. Who gets named?

Often the partner, and that is worth measuring explicitly. Run the same prompt set with and without your product name and record which entity the answer names. The gap between the two is your channel's share of your own visibility.

How much of this can we do ourselves?

The first reading, entirely. Ten prompts, three runs each, the names and the cited URLs written into a sheet. The reason teams buy a tool is the repeat, not the first reading. The work of getting named for this trade sits on its own page: how a cybersecurity firm gets recommended by AI.

Which engines matter for a security buyer?

Ask your own buyers. Our data covers ChatGPT and Gemini across three markets, and the answer differed by engine often enough that a single engine reading is a partial reading. A vendor that includes only one of the two in its base plan is selling you half the picture at the headline figure.

The check that is worth more than a demo

Before you compare vendors, ask an assistant the procurement question your own buyers ask, with no company names in it, and read which pages it cites. If every source is a news article and a directory, the gap is that your trust centre and your control documentation are not being read, and no amount of dashboard will change that until they are. AI Knows Us runs that reading against the five surfaces listed above and keeps the answer text, and we publish what it says about us as readily as what it says about anybody else.