peaksly explains · 02

Fundamentals

What is AI visibility?

AI visibility does not describe whether a brand is “visible on the internet”. It describes what was actually observed for fixed search prompts in fixed AI search engines: mention, observed position and sources, each with a denominator.

6-minute readUpdated 5 Sep 2026with a fictional example

Answer of an AI search engine to prompt 2 of 10 · search intent comparison · fictional and shortenedFictional teaching example, not a measurement

“Which suppliers deliver laser measuring systems for quality inspection in metal processing?”

For quality inspection in metal processing, three providers are mentioned above all: Nordwerk Industrie for measuring systems directly in the production line, Example Provider B for stationary measuring cells and Example Provider C for mobile handheld devices. Which solution fits depends on cycle time and part size.

Highlighted mention: Nordwerk IndustrieMeasurement path: direct API, one run. An API measurement path is not a visible consumer interface.

Observation · prompt 2 of 10

Mention: Yes
Brand: Nordwerk Industrie
Measurement path: direct API, run 1 of 1

Observed position and source follow separately in steps 1 and 2.

All names, prompts and answers are fictional. The highlight shows only the mention; observed position and source are recorded separately in steps 1 and 2.

In short

Counted, not estimated. Every number with a denominator.

AI visibility is the documented presence of a brand or domain in answers to a fixed set of search prompts. It only becomes meaningful with a visible denominator, AI search engine, point in time and measurement path.

Step 1 · Measurement frameFictional example: Nordwerk Industrie

Without a measurement frame, every percentage is incomplete.

A statement like “30 percent visible” sounds clear but leaves the most important question open: 30 percent of what? A traceable measurement frame names at least the prompts used, the AI search engines checked, market and language, the period and the measurement path.

The question set should cover different situations. A general provider search, a product comparison and a specific problem to solve can produce very different answers.

The denominator belongs to the result. “Mentioned for 3 of 10 prompts” is more verifiable than “30 percent visible”, because it stays immediately clear how many questions were actually examined.

One observationfictional

Prompt: 2 of 10, search intent comparison
AI search engine: fictional AI search engine, model version documented
Measurement path: direct API, run 1 of 1
Time: 14 Aug 2026, 09:10, fictional

Three separate values

Mention: YesNordwerk Industrie
Observed position: 1of 3 named providers
Source of the mention: own pagenordwerk.example

Step 2 · Measured valuesFictional example: Nordwerk Industrie

Three observations, three different questions.

Mention

Was the agreed brand or domain recorded in the answer? The mention rate relates recorded mentions to all valid observations. The review status shows separately whether an observation was checked by machine or fully by hand.

Observed position

Where did the brand appear in this specific answer? That is not a Google position and no guarantee for the next answer. It is a time-bound observation within one specific answer.

Sources

Which linked sources or domains did the AI search engine supply? A source can be present although the brand is not mentioned, and a brand mention can appear without a directly visible source.

The same count as in the observation sheet on the knowledge page. Numerator and denominator stay visible; no overall score is derived from them.

Finding for Nordwerk Industrie · 10 prompts · 1 measurement path · 1 run · 10 evaluable observationsFictional teaching example, not a measurement

1No
2Yes
3No
4No
5Yes
6No
7No
8Yes
9No
10No

Prompt 1filled = mentionPrompt 10

Mention: 3 of 10observations mention the brand.
Observed position: 1, 3, 3Only in the three answers with a mention.
Source of the mention: 2 own, 1 independentnordwerk.example twice, one trade source once.

Step 3 · LabellingFictional example: Nordwerk Industrie

A small number can be meaningful if it is labelled honestly.

The same number, labelled twice: “30 percent visible” leaves open how many prompts, which AI search engine, which measurement path and which point in time are meant. “3 of 10 prompts with a mention” carries these details and can be checked and repeated.

Together, the three values tell more than an overall score. But they do not yet show why the brand was mentioned or which change would produce a different answer.

Without denominator

30 % “visible”

How many prompts?
Which AI search engine?
Which measurement path?
Measured when?

With denominator

3 of 10 prompts with a mention, mention rate 30 percent

10 prompts, 1 run
fictional AI search engine, model version documented
direct API, no consumer interface
14 Aug 2026, fictional

Step 4 · Measurement pathsFictional example: Nordwerk Industrie

An API does not automatically measure the visible web interface.

A direct provider API, a public guest interface and a visible consumer interface can deliver different answers. Models, search functions, personalisation, localisation and interface experiments can differ. Which six answer surfaces peaksly distinguishes and how they can be measured is shown in the article on answer surfaces.

Never silently mix measurement paths. API measurement and measurement at the consumer interface need separate labels and time series. An API result must not be presented as if a user had seen exactly the same answer.

Prompt 2 of 10 on two measurement pathsfictional

Path A direct API model version documented · 14 Aug 2026, 09:10 Yesposition 1

Path B visible consumer interface checked manually · 14 Aug 2026, 09:12 Yesposition 2

In the report

own label per measurement path
own time series per measurement path
not offset against each other

Honest limit

What AI visibility does not prove.

The value lies in comparability: the same prompts, clearly separated measurement paths, visible denominators and evidence a person can check. Everything beyond that remains interpretation and should be named as such. What a mention is worth when nobody clicks is shown in the article on zero-click.

Not proven

future mention, position, reach or revenue
why a source was selected
a substitute for search engine optimisation or a professional content review
a development over time without repeated measurement

What stays comparable

the same prompts
separate measurement paths
visible denominators
verifiable evidence

Next: think SEO and GEO together.

The next article shows where both disciplines use the same foundation and where their observation surfaces differ.

SEO & GEO explained Open methodology →Look up the terms →