When a buyer asks AI who to hire in your market, does your team show up?

Recommended First measures which real-estate teams AI recommends, compares those answers with verified production, and shows you the evidence behind the gap.

Read the methodology

Verified production · Preserved answers · Methodology published · No ranking guarantees

Benchmark A — Greenville, South Carolina

Anonymized measured example

Verified production

Your team
$78.9M

Top-5 team by volume

Competitor
$33.0M

Outside the top 20 by volume

AI recommendations · of 256 valid captured answers

Your team
0 / 256
Competitor
17 / 256

You outproduce them. AI recommends them more.

Market
Greenville, SC · residential (anonymized)
Benchmark date
August 2026
Provider
ChatGPT (OpenAI)
Prompts
64 buyer and seller questions
Repetitions
4 per question
Sample
256 valid captured answers
Production source
RealTrends 2025 closed volume

OpenAI-powered search response; exact model metadata unavailable for this benchmark. Every figure in this exhibit describes the same captured run.

What we do

Three connected things, in one measured loop.

We do not reverse-engineer a private algorithm. We reverse-engineer the observable evidence environment around AI recommendations.

  1. 01

    Measure who AI recommends

    A fixed question set is run repeatedly against ChatGPT (OpenAI). Every answer is preserved word for word and each recommendation is classified per answer.

    Review the measurement methodology
  2. 02

    Investigate the evidence behind the answers

    Citation Intelligence records the pages and domains that appear around those recommendations, classifies each source, and compares that evidence footprint with your own coverage.

    See how Citation Intelligence works
  3. 03

    Improve legitimate coverage, then remeasure

    Gaps are prioritized, corrections and additions are documented, and the same question set is rerun so the change log sits beside the result.

    See how AI search optimization works

Figure 0Measure, investigate, improve — each step traceable to a preserved answer.

02The visibility gap

Market position and AI recommendations are two different things.

Plotting verified production against measured recommendation frequency isolates the case that matters: teams with real authority that the answers keep overlooking.

The addressable position

High authority, low AI visibility. The credibility already exists — the sources just don’t carry it.

Figure 01Verified authority vs. measured recommendation visibility

Verified authority →

AI Blind Spot

Market Leader

Low Priority

AI Overperformer

Your team
Competitor

Measured AI visibility →

Authority = verified 2025 production (RealTrends). Visibility = recommendation frequency across 256 captured answers · ChatGPT (OpenAI) · captured August 2026. OpenAI-powered search response; exact model metadata unavailable for this benchmark.

03The evidence

Every number traces back to an answer you can read.

Questions are fixed in advance, answers are preserved word for word, recommendations are classified per answer, and the final count can be inspected one response at a time.

Benchmark B — Jersey City, NJ (anonymized)

Every number has a receipt.

Every aggregate traces back to individual captured answers. Questions are stored, answers are preserved, recommendations are classified per response, and the final count can be inspected answer by answer.

Market
Jersey City, NJ
Provider
ChatGPT (OpenAI)
Captured
August 2026
Prompts
64 × 4 runs
Valid answers
256
  1. Figure 02Captured answer → recommendation count

  2. 01

    Question asked

    Who are the best real estate teams in Jersey City?

  3. 02

    Captured answer excerpt

    ChatGPT (OpenAI) · Jersey City, NJ · August 2026

    For Jersey City, a few teams come up consistently. Team A is generally regarded as the strongest for both buyers and sellers, with heavy activity downtown and along the waterfront. Team B is also well reviewed for condo sales, and Team C handles a good share of high-rise listings. Larger brokerage offices in the area are another option if you'd prefer a bigger platform.

  4. 03

    Recommendations extracted

    • Team Arecommended
    • Team Brecommended
    • Team Crecommended
    • The team (you)not recommended

    Classified per answer as a recommendation (named as who to hire) rather than a passing mention.

    OpenAI-powered search response; exact model metadata unavailable for this benchmark.

  5. 04

    Aggregate this answer feeds

    4 / 256recommendations across all valid captured answers

04Why the gap exists

Three things we investigate when AI overlooks a strong team.

Whether your track record is represented, whether AI can tell who you are, and whether the sources it reads substantiate any of it.

01

Authority

Is your track record represented?

Your production record may be strong without being consistently represented in the sources that surface around recommendations.

02

Entity clarity

Does AI understand who you are?

Name variants, stale brokerage attribution and split profiles fragment one team into several.

03

Source coverage

Do trusted third parties substantiate it?

When independent sources don’t clearly substantiate the team, the available evidence can be thinner than the real-world track record.

The sources behind the answers

A short list of source types keeps appearing in the answers.

Across benchmark B, certain source classes repeatedly appear alongside recommended teams.

Figure 03Source classes observed in captured answers — indexed frequency

Source classes ranked by indexed appearance frequency behind recommendations
Source classIndexed frequency (listing portals = 100)
Listing portals100
Brokerage sites44
Local publications22
Community forums12
Video7
The team's own siteNot cited in the captured benchmark responses
Unit
Indexed source-appearance frequency (most frequent class = 100).
Numerator
Citations belonging to that source class within the captured answers.
Denominator
Citations belonging to the most frequent class (listing portals).
Multiple citations
A single answer can cite several sources, so citations are counted per citation, not per answer.
Market
Jersey City, NJ · residential (anonymized)
Provider
ChatGPT (OpenAI)
Captured
August 2026
Sample size
64 buyer and seller questions × 4 per question = 256 valid captured answers

Indexed frequency expresses how often each source class appeared around recommendations relative to the most-cited class in the same captured set. One answer can cite multiple sources. Observed association only; no causal claim is made. OpenAI-powered search response; exact model metadata unavailable for this benchmark.

Citation Intelligence

See the evidence behind the recommendations.

See which pages and domains appear around AI recommendations, then compare that evidence with your own coverage.

Figure 04The Citation Intelligence workflow

  1. 01Prompt
  2. 02AI answer
  3. 03Citation
  4. 04Source classification
  5. 05Competitor comparison
  6. 06Evidence gap
  7. 07Action
  8. 08Remeasurement
  • Measure who AI recommends across a fixed, repeated question set.
  • Record the sources cited inside each captured answer.
  • Map the domains and source types that recur behind recommendations.
  • Identify entity and evidence gaps against your own coverage.
  • Prioritize legitimate improvements by importance, frequency and difficulty.
  • Remeasure using comparable prompts and conditions.

Explore Citation Intelligence

How results become evidence

What we measure once an engagement starts.

We don't publish case studies we can't evidence. Same questions, same repetitions, same provider — measured before and after.

  1. 01

    Baseline

    Recommendation frequency, competitor presence, source coverage and how accurately you're described — captured before anything changes.

  2. 02

    During engagement

    Every change we make is logged with the evidence behind it, so each intervention can be traced to a reason.

  3. 03

    Remeasurement

    The same controlled prompt set, the same repetitions, the same provider — new answers captured and counted the same way.

  4. 04

    Outcome

    We report what changed and what didn't. A flat number is reported as a flat number.

05Why trust this

Published method, stated limits, one name on the work.

Methodology

How the number is built.

64Buyer + seller questions

4Repetitions

256Maximum captured answers

  1. 01Raw answers preserved verbatim
  2. 02Recommendations classified per answer
  3. 03Compared with verified RealTrends production
  4. 04Every claim traceable to a stored response

Measured results and our interpretation are kept separate. Observations are recorded facts, inferences are labeled, and nothing is presented as a guaranteed ranking outcome.

The valid-answer denominator is published with each benchmark.

View the full methodology

What stands behind the work

No testimonials yet, and none invented. What exists is the record of the work itself.

Verified production data
Third-party market authority (RealTrends 2025), not self-reported volume.
256 preserved AI answers
Measured across repeated runs rather than collected anecdotally.
Published methodology
Prompt set, repetitions, provider, capture date and denominator are stated in full. Read the methodology
Sample audit available
The deliverable can be inspected before any conversation happens. View a sample audit
One client per market
A commercial commitment, not a scarcity device: two clients in one market would compete for the same answers.

Code of practice

What we won’t tell you

We don’t guarantee rankings.
AI answers are probabilistic and change over time. Nobody outside the model can promise a position in them.
We don’t call one screenshot a trend.
Recommendations are measured across repeated prompts and repeated runs, and reported with the denominator.
We don’t publish numbers we can’t trace.
Every published metric connects back to a captured answer that can be inspected on request.
We don’t invent case studies.
Until a client engagement has a measured before and after, there is no case study to show.
We don’t claim credit for everything that moves.
Observation and interpretation are labeled separately. A flat number is reported as a flat number.

Who is accountable for the work

Built and reviewed by Francisco Zuluaga

Founder, Recommended First · Brooklyn, NY

I built Recommended First because real market performance and AI recommendations are starting to diverge in ways that aren’t obvious from traditional search rankings. A team can lead its market on closed volume and still be missing from the answer a buyer actually reads.

My background is in data science, so the work starts with measurement: preserve the answers, compare them with verified production, and only make claims we can trace back to the underlying data. Every benchmark I send has been read by me before it goes out. If a number in your report looks wrong, the captured answer behind it is available on request.

francisco@recommendedfirst.com

Research

Published benchmarks, with their limits stated.

We do not publish findings that cannot be traced to preserved evidence.

  • Proprietary benchmark

    ChatGPT Real Estate Visibility Benchmark

    The primary citable dataset: how often ChatGPT (OpenAI) recommends real estate teams across repeated captured answers, with the prompt taxonomy, share definitions, exclusion rules and source-class taxonomy published in full.

    Market:
    Greenville, SC and Jersey City, NJ · residential (both anonymized)
    Sample:
    256 valid captured answers per market edition
    Updated:
    September 2026
    Read the report
  • Market benchmark

    Jersey City AI Visibility Report

    A single-market benchmark comparing verified production against how often each team was recommended across repeated captured answers, with the cited sources behind those answers.

    Market:
    Jersey City, NJ · residential
    Sample:
    256 answers maximum
    Updated:
    September 2026
    Read the report
  • Source analysis

    AI Citation Sources in Real Estate

    Which classes of sources appear most often around AI recommendations in a measured real estate market — reported as observed association, not causal proof.

    Market:
    Jersey City, NJ · residential
    Sample:
    256 answers maximum
    Updated:
    September 2026
    Read the report
  • Reference

    Real Estate AI Search Statistics

    Third-party statistics on AI use in the homebuying journey, each attributed to its original publisher with what was actually measured.

    Market:
    United States
    Sample:
    Stated per statistic where published
    Updated:
    September 2026
    Read the report

Research in progress

Explore the research hub

06Request a market analysis

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

Who This Is For

Built for teams with something worth discovering.

Good fit

  • Established real estate teams
  • High-producing agents
  • Market leaders in their city
  • Specialists with a strong transaction record
  • Teams with a measurable authority/recommendation gap
  • Teams investing in long-term brand authority

This works best for established teams and agents with real expertise, real results and a reputation worth representing accurately.

Not built for

  • Guaranteed rankings
  • Mass-generated content
  • Shortcuts designed to manipulate AI systems

Probably not a fit if you’re looking for guaranteed rankings, mass-generated content or shortcuts designed to manipulate AI systems.

Market Exclusivity

One client. One market.

We don’t retain directly competing teams in the same defined market. Your competitive intelligence stays yours.

Exclusivity defined by

  • geography
  • service category
  • client segment
  • active engagement
  • agreed contract terms

Market definitions and exclusivity are established during qualification and reflected in the engagement agreement.

FAQ

Questions sophisticated teams ask.

Request a market analysis

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

Where you appear, who is recommended instead, and which gaps are worth investigating in your market.

  • How often AI names you, and where you rank
  • The competitors AI names instead
  • The questions you win and the ones you lose
  • The websites AI keeps quoting
  • What you're missing, ranked by what's worth fixing

A short note reaches me directly. The form on the right requests a full analysis.

Evidence first. Recommendations second. Requests are reviewed before analysis — we don’t sell submitted information or use it to manufacture ranking claims.

Market assessment

Step 1 / 6

By submitting, you agree we may use these details to prepare and discuss your assessment. We never sell your information. See our privacy policy.

Request a market analysis