We ask, they answer, you see the words.
A reading puts a set of questions to four assistants, several times each. Every answer is quoted, checked against your profile, scored, and turned into a plan.
Your profile is the ground truth.
You give your name, your profession, and your city or region. You can add your website, your LinkedIn address, and up to five facts that are true about you. "Founded Marsh Estate Law in 2014" is a fact. "Great with clients" is not. You can also list names you are often confused with.
Every claim an assistant makes is checked against that profile. A claim that matches it counts for you. A claim that contradicts it counts against you. A claim the profile cannot settle is marked as such and counts for nothing.
Ten questions, in the words a client would use.
-
Who is {name}, the {profession} in {city}?
-
What is {name} known for?
-
Is {name} a reputable {profession}? Are there any concerns?
-
What do clients say about {name}?
-
What is {name}'s background and experience?
-
Who are the best {profession}s in {city}?
-
I need a {profession} in {city} for {typical need}. Who would you recommend?
-
How can I contact {name}?
-
How does {name} compare with other {profession}s in {city}?
-
What has {name} published, said, or been featured in?
The free check asks questions 1, 3, and 6. The typical need in question 7 comes from your profession. An estate attorney gets a will and a trust. A financial advisor gets retirement planning. An executive coach gets preparing for a first executive role.
Which assistants answer, and how.
Each assistant is asked through its developer interface, in the mode closest to the app, which means with web search on.
| Assistant | Model | Mode |
|---|---|---|
| ChatGPT | gpt-5.6-luna | Responses interface with web search |
| Gemini | gemini-3.5-flash-lite | Gemini interface with Google Search grounding |
| Perplexity | perplexity/sonar | Perplexity interface with web search |
| Claude | claude-sonnet-5-5 | Messages interface with web search |
The model ids are printed in every reading. When a provider changes a model, the reading says which one answered on that date.
Answers vary, so we show the spread.
The same assistant, asked the same question on the same day, can answer differently. It may search different pages, or none. It may find you on one try and a namesake on the next. One sample would hide that.
The Full Reading asks each question three times per assistant. Where the three agreed, the reading shows "all 3 tries." Where they differed, it says "2 of 3 tries" and shows each answer. Monitor and the 30-day re-check ask twice, because they are looking for change rather than a first picture.
The consumer apps can also answer differently from the interfaces we use, and every assistant changes over time. A reading is a dated record of what was said when we asked.
Every answer is read the same way.
For each answer, a separate step records four judgments. Who the answer is about: you, a different person, a mix, or no one it knows. Each factual claim with a verdict of matched, contradicted, or could not be checked. Your position in any recommendation list. The pages cited.
A small model makes those judgments in a fixed answer form. The form is checked against a schema and retried once if it is wrong. They are judgments, and the quotes are always there so you can judge too.
How each score is computed.
-
Recognition
The share of identity answers that knew who you are. A mixed answer counts half.
-
Accuracy
Of the claims that could be checked, the share that matched your profile. Claims the profile cannot settle count for nothing.
-
Identity
The share of answers about a person that were about you, not a namesake. A mixed answer counts half.
-
Visibility
The share of list questions where you appeared, weighted by position. First to third counts in full. Fourth and fifth count 70 percent. Sixth or later counts 40 percent.
-
Sources
Of the answers that cited pages, the share that cited your own website or profile.
Seventy and above is shown in green, forty to sixty-nine in amber, under forty in red. A score with too little to compute shows a dash and explains why.
Findings become fixes.
Each finding maps to a fix from a library of nine. The first four are a consistent bio sentence, a readable About page, Person schema with sameAs links, and an llms.txt file. The next three are a Google Business Profile where it applies, the directories for your profession, and disambiguation lines for a namesake. The last two are updating the stale profiles the assistants cited, and earning mentions on podcasts and in publications.
Each fix is written for you. It says what to do, where, and why it helps the specific finding, with the effort and the likely impact. The plan is ordered by impact, then effort. It ends the same way every time. These steps improve the odds, and no one controls what an assistant says.
Start with the free check.
One assistant, three questions, two minutes. You see the quotes and a short score before anyone asks you to pay.