AI visibility tracking and reporting

Why a rank is the wrong unit, and share of voice is the right one

Traditional rank tracking works because a search result is an ordered list that is broadly stable between identical queries. An AI answer is neither. It is generated, it varies between runs of the same question, and there is no position to occupy.

Research into answer stability keeps finding the same thing: a substantial share of the entities named in a response do not reappear when the identical query is run again, and only a small fraction of citations show up in more than one engine. So a report claiming you moved from fifth to second in ChatGPT is describing noise as progress.

Share of voice across many prompts is the honest alternative. Run a large enough prompt set often enough and the instability averages out into something you can actually track. It answers the question a business cares about, which is not where do I sit but how often do the engines think of me.

• No position exists to report, so any AI ranking number is invented
• Individual answers vary between identical runs, so single checks mislead
• Frequency across many prompts is stable enough to trend
• Variance is reported alongside the average, not hidden inside it

Building a prompt set that measures the right thing

The prompt set is the whole measurement. Get it wrong and you produce a confident number about a question nobody asks.

We build it from how buyers actually phrase things, which is closer to conversation than to keywords. Someone does not type best CRM software into an assistant, they describe a situation and ask what to use. So the set covers the problem as customers describe it, direct comparisons against named competitors, questions about your category where you should be a credible answer, and your own brand name to catch anything inaccurate.

It stays fixed once agreed. Changing prompts between runs makes the numbers incomparable, which is the most common way this kind of reporting quietly becomes meaningless. When the set genuinely needs to change, we start a new baseline and say so rather than pretending the trend continues.

What we report, and what we refuse to claim

Each month you get how often you appeared, which competitors appeared alongside you, what the engines said about you, which sources they cited, and what changed since last month.

The citation sources are usually the most actionable part. When an engine answers a question in your category, the pages it draws on are visible, and they are frequently third-party: a comparison article, a forum thread, a review site. That tells you where influence actually sits, which is often nowhere near your own website.

What we will not do is claim credit for a specific mention. Citation is a long tail spread across an enormous number of sources, and no agency can prove its own work caused one appearance rather than a review posted the same month. We show you the trend and describe what we think drove it, with the uncertainty attached.

Wiring it to the rest of your measurement

Tracking on its own tells you about visibility, not about business. The value increases when it sits next to everything else.

That means your analytics, so AI referral sessions are identifiable rather than mixed into direct. Search Console, which now reports on generative surfaces. Bing Webmaster Tools, which matters more than its market share suggests because Bing indexing feeds some assistants. And your server logs, which are the only place you can see AI crawlers fetching your pages at all.

One expectation to set early: AI referral traffic is currently tiny next to the search traffic it is displacing. Treating AI visibility as a traffic channel leads to disappointment. Treating it as influence over how buyers arrive already informed, and measuring it accordingly, does not.

• Analytics configured so AI referrals are not counted as direct
• Search Console generative reporting included in the monthly read
• Bing Webmaster Tools, because Bing indexing feeds some assistants
• Server logs, the only place AI crawler fetches are visible

What this service cannot do

It cannot make you appear in AI answers. It measures whether you do. The work that changes the number is content, entity clarity and third-party corroboration, and if you are not doing any of that, tracking will faithfully report a flat line.

It cannot give you a guaranteed cadence of improvement, because the engines re-read the web on their own schedule and some of what influences them is outside your control entirely.

And it cannot be audited by anyone else, which is a real weakness of every tool in this space including ours. There is no independent source of truth for AI visibility the way Search Console is for Google. We reduce that by keeping the prompt set and the method open, so you could reproduce our numbers yourself if you wanted to.