visibility to ai (@visibilitytoai)

In depth: visibility to ai: Benchmarking Whether AI Assistants Name and Cite Your Company

Written from visibility to ai's own profile data — claims, answers and testimonials — in a measured human voice via ScriptGrain · last written 2026-10-02

Ask a chatbot for the best payroll software and it does not give you ten blue links. It gives you an answer, a short list of vendors, and, if you're lucky, a handful of sources. Either your company is in that answer or it isn't. There is no page two.

That's the problem visibility to ai was built to measure. It is an AI-visibility benchmark, part of Adapt Progress Evolve and based in London, and it launched on 14 September 2026. The premise is simple: take the questions a company's buyers would actually ask, put them to the main AI assistants, and record what comes back.

What a Run Actually Does

"AI visibility" sounds fuzzy until you pin it to a question. visibility to ai does the pinning. A run uses a set of approved buyer questions and puts each one to four engines: ChatGPT (GPT-5), Claude (Claude Sonnet 5), Gemini (Gemini 3.6 Flash) and Perplexity (Sonar), all with web search switched on. The models are as stated on the methodology page.

Every answer is read. Every mention is quoted. For each question, the report says whether the company was named, cited, or missing entirely.

Named and cited are not the same thing, and the product refuses to pretend otherwise. A mention means the brand appears in the text of an answer. A citation means the brand's own site shows up in the answer's source list. The two are recorded and reported separately, never blended into a single score. A company can be talked about constantly and never linked to; that gap is often the most useful finding.

Three Verdicts

Each vertical gets one of three verdicts. Recognised-not-cited means assistants name the company in a meaningful share of answers but rarely cite its site: a page problem, not a brand problem. Entrenched means one rival appears in the majority of answers. Open means no vendor is named in more than about a quarter of answers, so the field is still up for grabs.

It's a blunt taxonomy, deliberately so. A company in an entrenched market faces a different job from one in an open market, and a single blended number would hide that difference.

Evidence Over Assertion

Because AI answers are slippery, the method leans hard on receipts. Every extracted claim in a report carries a verbatim quote from the frozen answer. If the quote can't be found, the claim is dropped, and the drop rate is reported alongside the results. Nobody has to take the tool's word for what an assistant said; the words are right there.

The Tiers

Pricing is tiered by depth, not by engine. All four engines are included at every level, so nothing is sold per assistant.

The Snapshot costs £89 as a one-off: 15 questions, 60 answers. The Deep Dive 100 costs £249 as a one-off: 100 questions, 400 answers. For ongoing monitoring, Track is £59 a month or £616 a year, and Agency is £299 a month or £3,122 a year. US prices, in dollars, sit on a separate US page of the site.

The arithmetic is consistent: 15 questions across four engines gives 60 answers, and 100 questions gives 400. It suits companies whose buyers now start research with a chatbot rather than a search bar, which is exactly where a repeatable set of questions across four models earns its keep.

Free Benchmarks and Free Tools

Not everything sits behind a price. visibility to ai publishes free public industry benchmarks drawn from real runs: 8 unbranded buyer questions and 32 answers per category. The categories are HR software, payroll, expense management, field service, review management, customer onboarding, brand-voice AI writing tools and Cyber Essentials providers. They show which vendors assistants actually name in each, which is a useful look at the market even for someone who never buys a report.

There are also four free tools, none of which needs an account:

• An AI readiness check, made up of six deterministic checks on whether assistants can read a site.
• An AI-crawler robots.txt check.
• An llms.txt generator.
• A single-page AI citation check.

The first three deal with the plumbing: can crawlers get in, and can they make sense of what they find? The last answers the narrower question of whether one page gets cited at all.

Where It Sits

visibility to ai is an Adapt Progress Evolve product, built in the UK. Its scope is deliberately narrow: GEO measurement, citation tracking, competitor share of answer, and buyer-question research, all pointed at one question. When a buyer asks an assistant, does your company come up? The answer, per question and per engine, is either a quote or a blank.