How to Measure AI Visibility

An answer surface is any place where an assistant composes a reply instead of returning a list of links. Seven carry commercial answers: Google AI Overviews, ChatGPT, Perplexity, Gemini, Claude, Grok and Meta AI. Each surface names its own sources, so each surface is measured on its own.

1. Build a prompt set of 20 to 40 buying questions, written in the wording buyers use.
2. Run every prompt against each assistant on a fixed schedule, one clean session per assistant per run.
3. Record three values for every prompt: the businesses named in the answer text, the domains listed as sources, and the run date.

A baseline is the first complete run of the set, captured before any change is made to the site. Every later run is read against that baseline, which is what turns a single observation into a measurement.

A prompt set is the fixed list of questions every run uses without alteration. Frozen wording enables comparison: two runs answer the same question only when the question was identical in both.

Five criteria govern the set.

• Wording copied verbatim from buyer language — sales calls, form submissions, support email — never from a keyword tool's phrasing.
• Size held between 20 and 40 questions, wide enough to cover three intents and short enough to run by hand across seven assistants.
• Coverage split across three intents: purchase, comparison and diagnostic.
• Geography written inside the prompt where the business serves a defined area, because a prompt naming no place returns a national answer.
• Versioning applied to every addition — new questions open a second series, numbered separately, while the original series continues unchanged.

Platforms automate part of a run and leave part of it manual. The coverage each one reaches is set out in ai visibility tools, compared.

citation rate = (prompts naming your business ÷ prompts run) × 100

The prompt set produces the citation rate, and no other input is required. Hold the calculation to one assistant and one run. A 40 percent rate on Perplexity and a 0 percent rate on Gemini average to a number describing neither surface.

Share of voice is the second metric and a different calculation: prompts naming your business divided by prompts naming any business in the category, across the same run. Citation rate answers whether you appear at all. Share of voice answers how much of the answer space you hold against rivals.

Stated as a sentence, the worked example reads: named on 4 of 20 prompts is a 20 percent citation rate.

Those figures come from a documented case study published by Koray Tuğberk Gübür. The arithmetic is the lesson: a 30.5 percent click rise divided by a 98.6 percent impression rise multiplies the ratio by 0.66, and 4.1 percent multiplied by 0.66 is 2.7 percent.

Three metrics mislead on answer surfaces.

• Click-through rate falls as impressions rise, which records wider exposure rather than weaker work.
• Impressions record link displays. Google defines a generative AI impression as a link to your site shown inside an AI feature, a different event from the business being named in the answer.
• Average position is absent. Google's generative AI performance report carries four dimensions — pages, countries, dates and devices — position is not among them.

Search Console splits the two halves. Impressions for AI Overviews and AI Mode sit in the generative AI performance report; clicks from AI features fold into the Web search type of the main Performance report. A rate built from both divides numbers never counted together.

Weekly: Google AI Overviews, ChatGPT, Perplexity, Gemini. Monthly: Claude, Grok, Meta AI.

Cadence follows index refresh rate. The four weekly surfaces rebuild an answer from a live search on each query, so seven days register real change. The monthly three move slowly enough that weekly sampling returns variance instead of movement. Continuous coverage across all seven is the monitoring retainer.