How Do Real Estate Teams Track AI-Influenced Consultations?

What Should You Count First?

Start with the question a team owner actually needs answered: are people finding us through AI and then asking to talk? A website visit is useful context, but it is not the same as a request for a conversation. Choose a reporting period and identify the real requests received during it. Keep traffic totals elsewhere so the two numbers cannot be confused.

Write down what qualifies as a request before you count. For this proposed method, a person asks to speak with the team about a real estate matter. Remove spam, tests and duplicates using the same rule each period. Keep vendor pitches, job inquiries and support contacts outside the residential buyer and seller headline count. Explain exclusions so another team member can reproduce the total.

Fictional illustration, not a client result: someone submits a form asking to talk about buying a home. They voluntarily select Buyer and AI assistant. Create one buyer-request entry and record that they reported AI discovery. If no usable referral information is available, mark that information unknown. The honest report says one requested buyer conversation with self-reported AI discovery. It does not say one AI-generated sale.

If the same fictional person selects Not sure instead of Buyer, put that one request in the both-or-not-sure group. Do not count it as a buyer and a seller. If the person later schedules a meeting, update the meeting stage without creating a second person. If they never attend, the requested conversation remains in the request total but does not enter the held total.

Give each entry an internal reference so duplicates and corrections can be checked. That reference does not make personal information anonymous. Follow the applicable correction, deletion and retention rules. Add dated corrections when keeping a history is permitted; do not keep personal data indefinitely or silently rewrite an earlier source answer after a later sale.

Evidence: ev-ga4-credit-limitev-pii-minimizationev-cppa-current-versions

How Should You Separate Buyer and Seller Requests?

Evidence: ev-fair-housing-scopeev-pii-minimization

What Should the Discovery Question Ask?

Ask, How did you first hear about our team? Make the question optional. Offer AI assistant, search engine, personal referral, social media, another source, not sure, and prefer not to answer. An optional follow-up can ask which assistant. A person must still be able to request a conversation and receive a response without answering these discovery questions.

Ask about the reason for the conversation separately. Someone may report AI discovery but want help as a buyer, seller, renter or referral partner. How they found you does not tell you what service they want. Keeping the two answers separate lets you explain both the inquiry source and the person's stated need without guessing either.

Collect only what is necessary for this measurement. Do not request raw AI prompts, transcripts, account names, device identifiers, exact addresses, financial details or information about another person just to identify a source. Inspect the form and its destinations before launch. A harmless-looking free-text field can still send private information into analytics or a vendor system.

A referral link may provide a source category such as ChatGPT. Remove private details before retaining only the information your collection rules permit. Save this separately from the person's answer. If the answer says AI assistant and the link has no usable source information, report the answer and leave the link source unknown. If the two disagree, show the disagreement.

Check the exact question and answer choices used during each reporting period. A change from first heard about us to last clicked before contacting us asks something different. Save the question's wording and change date so the team does not compare unlike answers. A later change in branded search or a public AI-answer test cannot fill an individual person's missing response.

Evidence: ev-openai-referral-limitev-pii-minimizationev-ga4-credit-limit

What Should You Save About the Inquiry Source?

The person chooses AI assistant

Save their voluntary answer, answer date, question wording or version, and optional assistant name. Label it self-reported; it is not verified evidence of the answer they saw.

A referral link identifies chatgpt.com

Save only the permitted source category and landing-page category after removing private details. The link does not reveal a prompt, answer text or reason for contact.

The answer and link source disagree

Keep both and mark them as conflicting. Do not pick whichever gives AI more credit.

Neither source answer nor usable link information exists

Say unknown. Do not substitute direct traffic, household data, a later sale or a public prompt test.

The team tests whether AI answers mention it

Keep the query, date, platform, available settings and cited sources in a separate research log. Do not connect that test to a person's contact record without a separately reviewed legitimate basis.

Evidence: ev-openai-referral-limitev-bing-visibility-limitev-pii-minimization

How Should Requested, Offered, Scheduled, and Held Be Counted?

1. Requested

   Count the person when they ask for a conversation. Record their voluntarily stated purpose and the available discovery information. Do not require an appointment to include a valid request.

2. Offered

   Record when the team offers a scheduling option or another neutral response through its established process. This describes the team's response, not the person's value or likelihood of buying.

3. Scheduled

   Count a scheduled appointment only when the required booking evidence exists. Show canceled, rescheduled, unreachable, declined and unresolved requests separately rather than treating every booking as attendance.

4. Held

   Mark held only when the defined conversation occurred. Keep appointment evidence separate from the original source answer. Representation, lending, listing and transaction records belong in their own business records and cannot establish the cause of discovery.

Evidence: ev-ga4-credit-limitev-california-agency-timingev-credit-application-conduct

Who Should Respond to the Request?

1. Use what the person asked for

   Start with their stated purpose and requested service or territory. Do not replace that information with a guessed identity, financial profile, property status or predicted value.

2. Check the activity and responsibility

   The broker or other responsible reviewer should check the requested activity, credentials or endorsements, entity relationships, supervision, brokerage policy and jurisdiction. A public license lookup alone does not establish authority for every activity.

3. Follow the brokerage's assignment process

   Share only necessary information with the person assigned to respond. Recording an assignment does not certify expertise, permission for every activity, service availability or a likely outcome.

4. Check what the report says

   Describe the actual service request, response and available source information. Keep unknowns visible. Do not turn a lead report into a statement that an operator is legally compliant.

Evidence: ev-dre-supervisionev-california-agency-timing

Which Data Must Stay Outside Marketing Measurement?

A field reveals or could suggest a protected characteristic

Keep it out of marketing attribution, eligibility decisions, priority, automated assignment, conversion scoring, suppression and public reporting. Do not use substitute clues, such as neighborhood or household data, to make the same inference.

Someone needs an accommodation or communication assistance

Do not score that need as a marketing signal. Use the separate process responsible for lawful accommodation, access, requested language help, notices, translation or required monitoring. Excluding it from a marketing report must not erase the operational duty.

Address, browsing, ownership or precise location data is available

Do not infer purpose, wealth, motivation, eligibility, protected characteristics or suitability. Collect a necessary service-area fact only through the team's reviewed service process.

The team wants to compare performance

Compare voluntarily stated buyer, seller and both-or-not-sure requests using identical definitions. Do not create marketing comparison groups based on demographics, neighborhood, family, disability, language or substitute personal clues.

Evidence: ev-fair-housing-scopeev-access-and-language-boundaryev-pii-minimization

What Do Platform and Analytics Signals Actually Show?

Evidence: ev-openai-referral-limitev-bing-visibility-limitev-ga4-credit-limit

What Should the Report Show?

Evidence: ev-ga4-credit-limitev-pii-minimization

What Should You Check Before Collecting Real Data?

• Explain why each source, purpose, scheduling, assignment and reporting field is necessary before collecting it.
• Make source questions optional and keep public AI-answer research separate from personal contact records.
• Keep raw prompts, transcripts, full query strings, account details, borrower data, precise locations, protected characteristics and third-party information out of AI attribution. Any separately necessary use needs its own responsible review and lawful process.
• Remove private information from URLs, page titles, campaign fields, custom dimensions and user-entered text before sending them to analytics. An internal reference is not automatically anonymous.
• Limit access to people who need it. Inspect forms, customer relationship management software, call tracking, enrichment, analytics and AI vendors before real information passes between them.
• Follow applicable correction, deletion, retention, access and security rules. Keeping a dated history does not justify keeping personal information forever.
• Withhold small or identifying groups from public reports. Do not export consultation records into unapproved advertising audiences, training data or other systems.

How Do You Keep Later Results Separate?

Finish this proposed consultation report at the held-conversation stage. Track later business milestones in their own records with their own definitions. A seller consultation is not a listing, and a buyer consultation is not a representation agreement. Keeping separate totals lets the team see progress without calling one event another.

Separate does not mean later in law. Agency disclosures, agreements, property access, financing discussions, applications, valuations, offers and contracts may carry duties before an attended meeting. The report cannot postpone those requirements. The responsible broker, lender or legal reviewer must check the actual activity and timing.

For an independently reviewed legitimate business purpose, a team may compare aggregate later milestones with consultation totals. It must not change the first discovery answer or turn a referral link into proof that AI produced revenue. A commission entry shows a commission recorded under its own rules, not the cause of the original contact.

Before sharing a consultation report, state the reporting period, question wording, counting rules and change dates. Show valid requests, missing sources, conflicting answers, corrections, exclusions and meeting stages. Say how small samples or withheld identifying groups limit the interpretation. Use self-reported AI discovery when that is the evidence, or recorded ChatGPT referral when only the link is known.

Use the report to ask a focused next question: which public pages and intake steps should we inspect? A page may receive AI-referred visits but no recorded conversation requests; that is a reason to review the page and form, not proof that the page failed. Check missing answers and collection changes before assigning a cause. Two different teams should be able to reach the same count from the same rules.

Evidence: ev-california-agency-timingev-credit-application-conductev-ga4-credit-limit

How Do You Put the Reporting Method Into Practice?

1. Confirm who can do the work

   Have the responsible broker or other reviewer check activity, entity role, credentials or endorsements, supervision and jurisdiction. For example, California DRE limits what unlicensed assistants may do and requires adequate supervision; this is not a nationwide certification.

2. Review housing and access requirements

   Check questions, substitute personal clues, assignment rules, report groups and vendor uses against applicable housing and consumer requirements. Include accommodation, communication, language and monitoring obligations where they apply. A published checklist is not proof that a particular team has completed this review.

3. Review where personal data goes

   Confirm applicable notices, rights requests, necessary collection, retention, deletion, security, vendor sharing and system connections. Check actual coverage before applying California privacy rules; they do not automatically apply to every team.

4. Write the counting instructions

   Define a valid request, purpose groups, source answers, referral information, unknowns, disagreements, exclusions, duplicates, corrections and reporting dates. Test the fictional example with a team member. If two people count it differently, clarify the rule before using a trend.

5. Check the current sources and public wording

   Recheck changing platform documentation and applicable public-law sources. Confirm that public pages, related links, evidence references, branding and reports say only what the supporting records show. AE's audit can help identify confusing public claims and gaps between traffic tracking and conversation-request reporting.

Evidence: ev-dre-supervisionev-fair-housing-scopeev-cppa-current-versionsev-openai-referral-limit