The method in one paragraph
A fixed panel of 30 questions in three groups (people, topics, category) goes to six AI answer engines once a day. Share of voice for an engine is the number of its answers that cite Artists & Robots divided by the number of questions it answered; a question the engine refused or failed on is left out of the denominator. Denominators are always shown. The full method is lower on this page; the questions are listed in full.
Run of September 15, 2026: 6 of 6 engines answered; on the category questions 5 of 55 answers cited us
Each cell reads cited of answered, then the share; where an engine answered fewer than it was asked, the third line says how many were asked. The category column is the competitive measure. The people column names our founders in the question, so a citation there is expected, not earned; the topics column includes one question that names the studio.
| Engine | Category (10 asked per engine) | People (8 asked per engine) | Topics (12 asked per engine) | All prompts (30 asked per engine) |
|---|---|---|---|---|
| ChatGPT | 1 of 1010% | 6 of 875% | 3 of 1225% | 10 of 3033% |
| Gemini | 0 of 100% | 6 of 875% | 3 of 1225% | 9 of 3030% |
| Grok | 0 of 100% | 4 of 850% | 1 of 128% | 5 of 3017% |
| Claude | 1 of 1010% | 3 of 838% | 1 of 128% | 5 of 3017% |
| Google AI Overviews | 0 of 50%of 10 asked | 1 of 250%of 8 asked | 0 of 60%of 12 asked | 1 of 138%of 30 asked |
| Perplexity | 3 of 1030% | 6 of 875% | 1 of 128% | 10 of 3033% |
| All engines | 5 of 559%of 60 asked | 26 of 4262%of 48 asked | 9 of 6614%of 72 asked | 40 of 16325%of 180 asked |
Source: geo.visibility and geo.visibility_prompts, run 2, 2026-09-15, prompt panel version 1. Cited = the answer named Artists & Robots or linked artistsandrobots.com. Answered = the engine returned an answer. Asked = 30 questions per engine, 6 engines, 180 in all; the totals row counts only engines that returned rows.
Share of voice by run, one panel per engine, 2 runs
Same scale on every panel, 0 to 100 percent, same dates on every x axis. The value beside each engine name is its latest run.
Source: geo.visibility, 2 runs, 2026-09-14 to 2026-09-15. Share = prompts citing us over prompts answered.
Source: geo.visibility, 2 runs, 2026-09-14 to 2026-09-15. Share = prompts citing us over prompts answered.
Source: geo.visibility, 2 runs, 2026-09-14 to 2026-09-15. Share = prompts citing us over prompts answered.
Source: geo.visibility, 2 runs, 2026-09-14 to 2026-09-15. Share = prompts citing us over prompts answered.
Source: geo.visibility, 2 runs, 2026-09-14 to 2026-09-15. Share = prompts citing us over prompts answered.
Source: geo.visibility, 1 runs, 2026-09-15 to 2026-09-15. Share = prompts citing us over prompts answered.
The question panel, version 1: 30 questions
The same questions every day. Once a question has run it is never reworded; new questions are added at the end.
People (8)
- Who is Jason Alan Snyder?
- Who is Heather Salkin?
- Who chairs the Intent Alignment workstream of the IEEE Standards Association Trust Stack for advertising?
- Who has chaired the 4A's AI Committee?
- Who founded Artists & Robots?
- Who was named to Adweek's 2026 AI Power 50 from Artists & Robots?
- Which advertising technologists have spoken at the United Nations about AI?
- Who was Momentum Worldwide's first Global Chief AI Officer?
Topics (12)
- What is the Friction Is Form framework?
- What does authorship over automation mean in marketing?
- How should a CMO judge an AI vendor's claims?
- What is the speed of meaning in advertising?
- Why do AI marketing tools hallucinate and how do you fix it at the data layer?
- What is intent alignment for AI systems in advertising?
- How do you measure whether an AI agent's answer can be trusted?
- What does no black box mean for marketing AI?
- How are sponsorship and brand experience measured with AI?
- What is a dynamic taxonomy for marketing data?
- Which studios publish weekly briefs on marketing questions with sources?
- What is Artists & Robots?
Category (10)
- What is the best creative intelligence studio for a Fortune 500 brand?
- Who is leading the use of AI in advertising and marketing?
- Which companies build agentic marketing systems for brands?
- What is a creative intelligence studio?
- Which AI marketing firms show the sources behind their recommendations?
- Who should a CMO hire to build AI agents for a brand?
- What is agentic marketing and who pioneered it?
- Best AI advertising and marketing agency for enterprise brands in 2026
- Which marketing technology companies hold patents on AI data integrity?
- Who is working on trust and governance standards for AI in advertising?
The method in full
What is measured
Once a day a program we wrote asks six AI answer engines the same thirty questions: ten about the category (creative intelligence studios, AI in advertising and marketing), eight about our two founders, and twelve about the topics we write on. The questions are fixed. Once a question has run it is never reworded; new questions are added at the end so the series stays comparable week over week.
For every question and every engine the program stores the answer text, whether Artists & Robots is named, whether Jason Alan Snyder or Heather Salkin is named, the position of Artists & Robots among the companies the answer names, the other companies named, and the URLs the answer cited.
How share of voice is computed
Share of voice for an engine on a run is the number of questions whose answer cited Artists & Robots divided by the number of questions that engine answered. A question the engine refused, timed out on, or returned nothing for is counted as unanswered and left out of the denominator. The denominator is printed next to every figure on this page, so a high share on a small denominator reads as what it is.
Whether an answer cites us is decided first by a plain text match on our name and our domain. Only an ambiguous answer goes to a model judge, and that judge is tested every run against a planted false positive and a planted false negative. If the test fails, the run's numbers are held and not published.
What this page does not claim
The program is ours and the numbers are self-measured. Nobody outside the studio audits them. The engines change their answers from day to day and from account to account, so a single run is a sample, not a census. A figure never claims a period the data does not cover: each chart and table names the run date it comes from, and the series starts on the day the program first ran, 14 September 2026.
An engine that returned no rows on a run is listed by name with no rows returned rather than shown as zero. An engine that answered no questions has no share of voice for that run and its row says so. The people questions name our founders, so a citation there is expected rather than earned; the category questions are the competitive measure, and the page reports them first.
The honest note. The program is ours and the numbers are self-measured. We publish them, with their denominators and their dates, so that anyone can see what an AI engine says when asked about us, and so that the day the share moves, up or down, is on the record.