AI Search Visibility Metrics KPIs: What to Track and Report

Your client asks, "Are we showing up in ChatGPT?" You open ChatGPT, type a question, and the brand is there. That looks like an answer, but it is a sample of one. Ask again tomorrow, or ask the same question worded differently, and the answer changes.

This guide covers the numbers that replace that screenshot. After reading it, you can choose which KPIs to track, say what each one means in a client call, and build a one-page report that doesn't promise more than the data shows. It runs in this order: what the KPIs are, why classic SEO numbers miss them, the core four, source metrics, sample size, segmentation, and reporting.

What are AI search visibility metrics KPIs?

AI search visibility metrics KPIs are the numbers that show how often, how prominently and from which sources AI answers mention your brand. They are measured across many answers to a fixed list of prompts, over time, and compared with competitors. A single answer is an observation. A KPI is a count across many of them.

The group has two halves. The first half describes the answers themselves: how often your brand is named, how it compares with competitors, and where in the answer it appears. The second half describes the sources behind the answers: which websites the engine cites, and whether yours is one of them.

Here is a practical test for whether you need them. Pick the five questions your client's buyers most often ask, run each one five times in fresh temporary chats, and count how often the brand appears. If the count is the same every time, a screenshot is enough. If it moves, and it usually does, you need KPIs with a sample size.

Why classic SEO KPIs miss AI answers

Rankings, impressions and clicks tell you how a page performs in a list of links. An AI answer has no list of ten positions to be in. It is one block of text that may name three brands, five, or none, and it may cite a few pages as sources. That is why teams that already run solid SEO reporting still need a second set of numbers, and that is what AI search monitoring is for.

AI answers change from run to run

The same prompt can return a different answer each time you run it. Brands appear, drop out and swap order. This is how the category behaves, so any number built from one run is a sample of one. A KPI built from many runs, with the count shown next to it, is the fix.

A mention is not a visit

Your brand can be named in an answer without anyone clicking anything. Classic analytics won't show it, because nothing reached your site. Mention-based KPIs exist to cover that gap. They tell you whether you are part of the conversation the buyer is reading, before any click.

The core AI search visibility KPIs

Four numbers cover most of what clients ask about. Each is easiest to read when it carries its sample size, so we'll show that every time.

Visibility: how often answers name you

Visibility is the share of answers that name your brand: the answers naming the brand divided by the number of answers counted. For example, if the fictional brand Brindlepath is named in 12 of 40 answers, its visibility is 12 of 40, or 30%. The "12 of 40" matters as much as the 30%, because it tells the reader how much to trust the figure. Read how Aivting calculates visibility for the exact definition.

Share of voice among the brands you track

Share of voice is your brand's mentions as a share of all mentions of the brands you track, counting one mention per brand per answer. Say you track Brindlepath plus two fictional competitors, Tarnfield and Corvane, across 40 answers. Brindlepath is named in 12 answers, Tarnfield in 20 and Corvane in 4. That is 36 mentions in total, so Brindlepath's share of voice is 12 of 36, or 33%, while its visibility is 12 of 40, or 30%.

The good thing about share of voice is that it puts you next to rivals. The less good thing: it depends on who you track. Add a competitor and your share falls even if nothing about your own presence changed. Aivting lets you track up to 10 competitors per project, so decide on the set first and keep it stable when you compare months. A competitor you add is counted only in answers collected after you add it; answers already collected aren't re-checked. A competitor you remove drops out of every count, past ones included.

Visibility and share of voice can move in opposite directions. If your brand is named in more answers but a competitor is named in many more, visibility rises while share of voice drops.

Average position inside the answer

Position is the order in which your brand is first named among the brands you track: 1 means it was named before any tracked competitor. Brands you don't track don't count. The average position is that figure across the answers that name you. Being named first in a short recommendation is a different outcome from being the fourth of five brands in a long one.

One caution. An average built from one or two answers is noise, so Aivting shows average position only once your brand has been named in at least 3 answers. Any tool or spreadsheet you build should apply a similar floor.

Sentiment, and why to treat it with care

Sentiment is an average tone score from -100 to +100 across the answers that name the brand, describing the tone in which it is mentioned. It's the easiest KPI to over-read, because tone is subjective and a small sample swings it. In Aivting, sentiment is not available yet on live projects. We show it here so you know what the term means when you meet it in other reports.

Source metrics: which sites AI answers cite

Answers often cite the pages they drew on. Source metrics count those citations. They answer a different question from the first four: not "do the answers name me?" but "which websites do the answers lean on?" These AI search visibility metrics point to action, because a source is something you can work on, while an answer is something you can only observe. Google describes how AI features in Search use links in its Search Central documentation.

Your own site as a source

This metric compares your visibility in answers that cite your domain with your visibility in answers that don't. If Brindlepath is named in many of the answers that cite its site and in few of the answers that don't, that is a pattern worth noting. It is not proof that the citation caused the mention. The mention might have caused the citation, or both might follow from a third thing. Aivting shows this comparison only when at least 10 answers cite your site and at least 10 don't, and labels it a pattern, not cause and effect.

Gaps: sources that cite competitors but not you

A gap is a domain or page that is cited in answers naming a competitor and in no answer naming your brand. It never includes your own site or a competitor's own site. These are your shortest to-do list: sites the engines already cite for this topic, where the answers that cite them never name you. Read how gaps are found for the exact rule, including which domains are left out.

Listicles you are missing from

Many cited pages are list articles: "best tools for...", "top 10...", "X alternatives". Aivting flags a page as a listicle from words in its URL or title such as best, top plus a number (top 10), alternatives, vs, compared, comparison, review or reviews, and "tools for" ("best practices" pages are excluded). Turn on Listicles only and Gaps only together to see list pages that answers cite alongside competitors but never in an answer that names you. That is a practical outreach list: each page is one that engines already cite when answering your buyers' questions.

Why every KPI needs its sample size (n)

A rate without its count is a claim you can't check. "30% visibility" could be 3 of 10 answers or 300 of 1,000. The first can swing wildly on the next run. The second barely moves.

The sample size, written as n, is the number of answers that came back usable. Count it carefully, because there are two ways to get it wrong. An answer that failed to arrive, came back empty or was blocked is not an answer that "didn't mention you". Aivting counts those separately and never adds them to the "not mentioned" side. When n is zero, it shows no rate at all rather than 0%. Read sample size in Aivting for how n is counted.

A worked example shows why this matters. Suppose the fictional brand Pelliwick's visibility is 30% in January and 40% in February, and the client reads that as progress. If each month rests on 10 answers, the whole move is one extra mention (3 of 10 against 4 of 10). If each month rests on 200 answers, the same 10 points means 20 more mentions and deserves a closer look. The percentages are identical. Only n tells you which story you are in.

There is no universal n above which a number becomes safe. The honest approach is to show n beside every figure, widen the date range when n is small, and avoid reading much into a change between two small samples.

How to segment the numbers

A single blended visibility number hides more than it tells. Four cuts do most of the work.

By engine. ChatGPT and Google AI Overviews are different products with different behavior. A brand can do well in one and poorly in the other, and a blended figure will hide that. Aivting reports each engine separately: [ENGINE AVAILABILITY].

Branded vs non-branded prompts. A prompt that includes your brand name will name you far more often than one that asks about the category. Mixing them inflates the headline number. Report them apart.

By intent. Aivting tags prompts as informational, commercial, comparison or transactional, using rules on the wording, so treat the tags as a filter, not a measurement. Commercial and comparison prompts sit closest to a purchase decision, so they often matter most to a client.

By date range. Compare equal periods, such as this month against last month, and keep the prompt list and competitor set the same, or the comparison breaks.

How to report AI visibility KPIs to clients and leadership

Reporting is where careless numbers do the most damage. The fix is a short, repeatable format that always shows n and always names what it can't tell you. For what each plan includes, see plans and pricing.

A one-page monthly report

A page like this is enough for most clients:

1. Visibility with n, this month against last month, split by engine.
2. Share of voice against the named competitors, with the competitor set listed.
3. Average position, only where there are enough answers.
4. Trend, drawn from daily data across the period, not a single day.
5. Top gaps, the three or four sources that cite competitors but not the client.
6. Next actions, one line each, tied to a gap or a prompt.

Aivting exports a report with your agency's or your client's logo, and creates read-only share links that need no login. Share links are off by default, expire after 30 days and can be revoked at any time.

What not to promise

Don't say a change in a KPI was caused by your work unless you can rule out the alternatives: a run of unusual days, a competitor change, a different prompt set. Don't promise placement, either. Aivting measures and advises. It doesn't guarantee that any engine will name or cite a brand.

Set a baseline, then track change

The first month of data is a baseline, not a score. Collect it, write down the n behind each figure, and only then start comparing. The daily graph adds a point for each day that has collected answers [CADENCE: how often collection runs automatically], but compare weeks and months rather than single days because individual days vary.

If your first KPI is ChatGPT, the practical question is how to monitor brand mentions in ChatGPT in a way that produces these numbers. We cover it step by step in our guide to monitor brand mentions in ChatGPT, by hand and with a tool.

FAQ about AI search visibility metrics KPIs

Which AI search visibility KPI should I track first?

Visibility, shown with its sample size. It's the share of counted answers that name your brand, and every other KPI builds on it. Add share of voice once you've decided which competitors to track.

How is share of voice different from visibility?

Visibility compares your mentions with all answers. Share of voice compares them with all mentions of the brands you track. If your brand appears in 12 of 40 answers and the one rival you track appears in 20, visibility is 12 of 40 (30%) and share of voice is 12 of 32 mentions (38%). Add more rivals and share of voice falls; visibility doesn't move.

How many answers does a number need before it means something?

There's no universal threshold. Aivting shows the number of answers (n) next to every figure and shows average position only once a brand is named in at least 3 answers. When n is small, widen the date range and read the figure as a rough signal.

Can I track these KPIs for Google AI Overviews as well as ChatGPT?

Yes. Each engine is a separate product, so report each one on its own. [ENGINE AVAILABILITY: list only the engines Aivting collects on the publish date]

How often should I report AI visibility KPIs?

Weekly or monthly. Aivting's graph is drawn by day, but single days vary, so trends over weeks tell you more than any one day. This is our advice, not a rule: pick the rhythm your client already uses for other reports.