ChatGPT visibility audit

See what ChatGPT says about your team before your competitors do.

A fixed question set, asked repeatedly, with every answer preserved — then classified for who was recommended, who was cited, and where your evidence is missing.

Prompts
64 buyer and seller questions
Repetitions
4 per question
Provider
ChatGPT (OpenAI)
Capture date
August 2026
Valid answers
Published with each benchmark (256 maximum)

What the audit measures

How often the team is recommended
Counted per captured answer, reported with the valid-answer denominator.
Which competitors appear instead
Ranked by how many answers recommend them across the same question set.
Which questions are won and lost
Buyer, seller and neighborhood questions, reported individually.
Which websites ChatGPT cites
Every cited URL and domain recorded against the answer it appeared in.
Whether the team is described accurately
Brokerage, markets, specialties and team composition, as stated in the answers.
Which evidence gaps matter most
Ranked by frequency across answers, confidence in the observation and difficulty to close.

Why one prompt, or one answer, is not enough

Ask the same question twice and you can get two different lists of teams. A single answer is one draw from a distribution, so a screenshot of a good answer and a screenshot of a bad answer are equally uninformative on their own.

The audit therefore fixes the question set before capture, asks each question several separate times in independent sessions, and reports every figure against the number of valid answers actually captured. Repetition is what turns an anecdote into a rate.

Recommendation
The answer puts the team forward as someone to hire or contact.
Mention
The team is named, but not put forward — for example, listed as one of many.
Citation
A source is linked or explicitly attributed in the answer, with a resolvable URL.
Omission
The team does not appear at all in that answer. Counted, not ignored.

How the audit is run

  1. 01

    Define market and entity

    Agree the market, the team name, its variants and the competitor set to be observed.

  2. 02

    Build the question set

    Real buyer and seller questions for that market, locked for the measurement window.

  3. 03

    Capture repeated responses

    Each question asked multiple separate times rather than once.

  4. 04

    Preserve raw answers

    Full text, timestamp, provider and citations stored for inspection.

  5. 05

    Classify recommendations

    Recommended, mentioned, or absent — assessed per answer.

  6. 06

    Extract citations

    Every cited URL and domain recorded and classified by source type.

  7. 07

    Compare with production

    Measured recommendation frequency set against verified closed production.

  8. 08

    Prioritize evidence gaps

    Sequenced by importance, frequency, confidence and difficulty.

  9. 09

    Remeasure

    The same locked question set rerun under comparable conditions.

What the audit cannot conclude

The audit observes outputs. It does not observe how the provider retrieves, weighs or ranks anything, and it cannot tell you why a particular team appeared.

What this cannot prove

  • It cannot show the provider's ranking logic — that is not observable from outside.
  • It cannot prove that a cited source caused a recommendation.
  • It cannot guarantee a ranking, a recommendation or a citation, and no work that follows it can either.
  • It cannot predict what the same question will return next month.
  • It measures one provider per engagement; other answer engines are measured separately.

A preview of the output

Anonymized measured example · Jersey City, NJ · residential. Every figure below traces back to a preserved answer.

Jersey City, NJ · residential · ChatGPT (OpenAI) · captured August 2026 · 64 questions × 4 repetitions · 256 valid captured answers
TeamAnswers recommending
The team (you)4 / 256
Team A116 / 256
Team B52 / 256
Team C48 / 256
Team D40 / 256

Anonymized measured benchmark, published with permission of nobody it identifies because it identifies nobody. Source: Recommended First, Jersey City AI Visibility Report. Exact model metadata is unavailable for this benchmark. Not a verified client outcome.

What this cannot prove

  • Counts describe the captured window, not a permanent state.
  • A competitor appearing more often is an observation, not an explanation.
  • Nothing here demonstrates how the provider ranks sources internally.

See who AI recommends in your market — and where you stand.

One client per market. Start by telling me where you work.