For PR and comms teams
Five questions to ask of an AI-visibility report
For anyone who receives or makes one: an in-house team, an agency or a tool. Ask them in writing before the score reaches a decision.
This page lists what to ask of an AI-visibility report, not how to measure. It is written for both sides of the table: the team that receives a report and the team, agency or tool that makes it. This website sells nothing and rates no tool, agency or report. Back to AI answers and PR.
One AI answer is one draw, not a measurement.
First: what decision is this report for?
Ask this before the five questions. A report for a monthly check needs less than a report that moves budget or goes to the board.
Many AI-visibility reports count how often AI answer engines such as ChatGPT or Gemini name you, your share of voice, tone and cited sources. That is useful, but being named is not being described correctly.
The five questions
Q1. How many times was each question asked?
Why: answers change from one ask to the next.
A good answer sounds like: "Many times per question on each engine. Here is the exact number."
Q2. Checked against which facts, written down by whom, and when?
Why: presence is not correctness, and a friendly tone is not accuracy.
A good answer sounds like: "Against facts your team signed off, written down before the first question. Wrong claims are counted separately from mentions."
Start small: begin with five facts you can sign off this week: your name, what you do, where, one product fact and one contact. Facts that change by decree or tariff can wait.
Q3. What is the normal spread, and is the change bigger?
Why: a move inside the spread across repeats and across runs on different days is noise, not news.
A good answer sounds like: "Here is the range across repeats and across runs. This quarter's change is inside it, so we do not call it a trend."
Q4. Which language, which country, and whose words?
Why: a report asked in English from one place may not show what your buyers in Germany, Austria, Switzerland or Hungary see.
A good answer sounds like: "In German and Hungarian, asked from those markets, phrased as your buyers ask. The prompts are attached; gaps are listed."
Q5. What does the score count, and could someone rerun it?
Why: tools define the headline score differently. If engines, model versions, dates or the definition change, this quarter cannot be compared with the last.
A good answer sounds like: "Engines, versions and dates are logged. The definition is written out and has not changed."
For press releases, also ask: does the report separate the release, coverage of it and your own website? Measure what AI does with a press release shows one way.
For both sides of the table
"Not yet, and here is our plan" is a good answer too. Agencies and in-house teams can run the same five questions on their own dashboards before anyone asks. The questions are a way to read a report the right way, not a test of the people who made it.
Why one ask is not enough, with a calculator: how far to trust your own log.
Checklist to copy
Cite one question as "makefield.io/pr/reports Q3", or link it as makefield.io/pr/reports#q3.
What to ask of an AI-visibility report
makefield.io/pr/reports, version 1, last checked 26 October 2026
First: what decision is this report for?
- How many times was each question asked?
- Checked against which facts, written down by whom, and when?
- What is the normal spread, and is the change bigger?
- Which language, which country, and whose words?
- What does the score count, and could someone rerun it?
Press releases: are the release, coverage of it and our website reported separately?
What industry bodies have published
- AMEC GEO Principles (May 2026): repeatable prompts, documented methods, transparent assumptions; AI outputs are "directional evidence rather than absolute truth"; do not rely on a single score, platform or tool (as reported by B&T).
- PRCA, "Beyond Visibility: How to Measure GEO" (June 2026): warns against vanity metrics and names accuracy as an aim.
- IAB, "Measuring Visibility in the AI Era" (August 2026), page 32: "Single-response measurement is not measurement." IAB also asks for ranges and confidence, rates of hallucination and factual errors, and results per market, and says "the source of truth must be brand-supplied".
- Answers vary: in a January 2026 test by SparkToro and Gumshoe, two companies that sell in this field, so a vendor study (about 600 volunteers, 12 prompts, nearly 3,000 runs), the odds of getting the same brand list twice were under 1 in 100 (as reported by Search Engine Land).
Words on this page
- GEO: Generative Engine Optimisation, work on how AI answers describe you.
- Share of voice: your mentions against competitors' mentions.
- Hallucination: content an AI answer invents.
- Spread: how much results differ when the same question is asked again.
What this page does not claim
- It does not say any report, tool or agency is bad, and it rates none.
- It does not replace setting objectives.
- Makefield's own small test of how answers vary is the floor test. It compared wording, not facts.
How to cite
Denes Csaszar, "Five questions to ask of an AI-visibility report", Makefield, makefield.io/pr/reports, last checked 26 October 2026, version 1.
Last checked 26 October 2026. What changed: version 1, first published 28 October 2026.
Makefield is an independent research project on AI answer accuracy, based in Berlin, Germany. This website sells nothing.
The answer and the report: How to read your own AI-visibility audit asks four questions of one AI answer. This page asks five of the report built from many.
Also for PR teams: Ask the AI about your organisation · Measure what AI does with a press release