The wording of the question is part of the measurement.
These are families of prompts for observing how an entity is described, with numbered templates. Each one answers a different question, and none of them is a trick for obtaining a mention.
A prompt library is the fixed set of questions put to AI systems in order to observe how they describe an entity. This one, version 1 of October 4, 2026, holds thirteen templates in two groups: entity prompts, which name the subject, and category prompts, which do not. A set is built by choosing the templates that match the question being asked, filling in the brackets, and then leaving each prompt unchanged from one run to the next.
How to read the templates
Square brackets mark what you fill in once: [name], [role], [organization], [other organization], [category], [city or market], [need]. Once filled in, a prompt stays as written, in each run and each engine.
Prompts are for observing. Nothing here is designed to make a system say something, and a prompt that names the answer it hopes for measures the prompt.
Run person prompts only about yourself, or about someone who has asked you to. An answer about a person can contain false statements, and such answers are not published.
Entity prompts
These apply to any person or organization. They show whether the entity is identified, and whether the facts are right.
| No. | Family | Template | What it shows |
|---|---|---|---|
| E1 | Identity | Who is [name]? | Whether the entity is recognized, and whether it is the right one |
| E2 | Identity, with context | Who is [name], [role] at [organization]? | Whether a namesake explains an error in E1. The prompt supplies the role and the organization, so neither is coded from it |
| E3 | Activity | What does [organization] do? | Category, offer and positioning as described |
| E4 | Record | What is [name]'s background? | Roles, dates and sequence |
| E5 | Current role | What is [name]'s current position? | Whether a former role is given as present |
| E6 | Standing | What is [organization] known for, and according to which sources? | What the description rests on |
| E7 | Sources | What are the main sources of information about [name]? | Which sources the system names. This is the system's own account: it is not evidence of the sources used, and it may name sources that do not exist |
Run E1 and E2 as a pair. The difference between the two answers helps separate a problem of recognition from a problem of confusion: see confused with someone else.
Category prompts
These apply to organizations, and to people only where a real category exists. They show whether the entity appears when it is not named.
| No. | Family | Template | What it shows |
|---|---|---|---|
| C1 | Category | Which firms specialize in [category]? | Presence without being named |
| C2 | Category, by market | Which [category] providers operate in [city or market]? | Variation by market |
| C3 | Comparison | How does [organization] compare with [other organization]? | The terms on which the entity is compared |
| C4 | Alternatives | What are the alternatives to [organization]? | Which others are associated with it |
| C5 | Use case | I need [need]. Who should I consider? | Presence in a question phrased as a need |
| C6 | Selection criteria | How should I choose a [category] provider? | The criteria the answer uses, which the entity's evidence may or may not meet |
A set meant to produce a share-of-voice figure draws on these templates and needs volume; the thresholds are given in what a valid reading requires.
Wording rules
Write a full sentence, as someone asking an assistant would.
Do not lead. "Why is [organization] the best choice for…" contains its conclusion. A yes-or-no question about credibility plants a doubt; E6 asks what the entity is known for.
Leave out superlatives and years unless a real user would add them. The system may add them itself. In an analysis of five million query fan-outs dated April 2026, Peec AI found that ChatGPT added words the user had not typed, "reviews" among the most frequent, and added the current year in 5.44% of prompts (Peec AI, 2026). Adding them yourself tests a different question.
One question per prompt. A compound prompt gives an answer that cannot be coded against either part. E6 is the one exception, and its second half is read only for the sources named.
Keep a control. Include one organization whose facts are well established. If the answers about it are wrong too, that points to the engine or the conditions. A correct control proves nothing about a less documented entity.
Treat each language as its own prompt. Translate for meaning, have a native speaker check the result, and freeze each language version separately. Results are not merged across languages.
Freeze the set. Date the version, and record every change as a new version.
Building a set
| Purpose | Templates | Size |
|---|---|---|
| How is one person described? | E1, E2, E4, E5, E7 | A small deliberate set; a diagnostic |
| How is one organization described? | E1, E3, E6, E7, C3 | A small deliberate set; a diagnostic |
| Does a brand appear in its category? | C1, C3, C4, C5, C6 | Large enough for a share-of-voice reading |
| Does the description differ by market? | C2, and the same set per language and location | The same set, repeated per market |
On this site, a private assessment uses a 30-prompt set and the monthly engagement follows up to 50 prompts. Both are diagnostics of how one entity is described; neither is a share-of-voice measurement. How many prompts a reading needs, by purpose, is explained in how many prompts make a reading. The four families named there, entity, category, comparison and use case, are the ones these templates detail.
What this library is not
It is not a list of the questions people ask. It models them.
It is not complete. A field has its own vocabulary, and the prompts that matter are found by listening to the people who ask.
It is not a result. No answer to any of these prompts is reported on this page.
Where it sits
The library serves the protocol in methodology; changes to it are entered in the change log, and its terms are defined in the glossary. It is part of Research, and it supplies the first layer of The Authority Architecture System, Perception Intelligence.
Frequently asked questions
What prompts should I use to check what AI says about me?
Start with the identity pair, "Who is [name]?" and the same with your role and organization, then the record and current-role prompts. Run each several times in each assistant and keep the answers.
Should prompts be the same in every engine?
Yes. The same wording, in the same language, so that wording is not one of the causes of a difference between answers.
Can a prompt make an assistant mention me?
A prompt that names you will produce an answer about you. That shows how you are described, not whether you appear unprompted. Category and use-case prompts, which do not name you, show the second.
How often should the set change?
As rarely as possible. Each change breaks comparison with earlier runs, which is why a change is recorded as a new version.
Understand how search and AI currently interpret your authority.
Request a confidential assessment of your visibility, entity signals, narrative consistency and source authority.
Request a Private AssessmentSelective engagements for leaders and organizations with material reputational stakes.