Everything about FoundScore™
and FoundLab™.

ai{Found}RE is an AI-visibility system and education platform built specifically for real estate agents. We audit exactly how often AI systems recommend you, assign you a trackable FoundScore, and deliver a clear roadmap to make you the obvious recommendation across all major AI platforms. You can run a self-serve audit to see your starting position, or join the FoundLab cohort to actually move the score with weekly live coaching from Raf and Nancy.

The search bar is being replaced by the AI prompt. Buyers and sellers are no longer just Googling for agents — they are asking AI platforms like ChatGPT and Gemini who they should hire. Research shows that 67% of buyers are comfortable using ChatGPT to find an agent, and AI systems concentrate their recommendations on the same 2 to 3 agents 80–87% of the time. If your name is not coming up, you are losing deals you never even see. Worse: AI tends to lock in winners. Once a competitor becomes the recommended agent, the inertia compounds against you.

The FoundScore is a proprietary 0–50 metric — like a credit score for your AI discoverability. We calculate it by running 25 tailored prompts across 5 major AI platforms for each of your specific markets (typically 125 prompts total per audit). The score is the weighted percentage of prompts where you appear, normalized to a 0–50 scale. The score gives you an honest answer to the question every agent should be asking: does AI know I exist? It also maps to a Level (1 through 6) so you can quickly see where you stand and how much room there is to move.

Earlier versions of the FoundScore used a 0–100 scale, which made every real-world score feel demoralizing. The strongest agents in the smallest markets topped out in the low 40s. Median scores landed in single digits. The math was correct but the framing was punishing. We recalibrated to 0–50 because that reflects the realistic ceiling of AI visibility. There is always competition, always another agent name surfacing in some prompts. A score of 25 means you are appearing in roughly half of the relevant AI conversations in your market. A score of 41 or higher puts you in the top tier — Level 6 — the rarefied air of agents who genuinely dominate their AI footprint. The 6 Level bands (1: 0–8, 2: 9–16, 3: 17–24, 4: 25–32, 5: 33–40, 6: 41–50) give you a coarse readout of where you stand and how much room there is to grow.

We run 125 total prompts per audit across the five platforms your clients are actually using to find agents: ChatGPT, Gemini, Perplexity, Google AI Overviews, and Claude. We weight these platforms based on current consumer market share. As of 2026, our weights are: ChatGPT (30%), Gemini (25%), Google AI Overviews (20%), Perplexity (15%), and Claude (10%). These weights are reviewed quarterly. When market share shifts — for example, when Claude inevitably gains ground — we re-weight to keep the FoundScore reflective of where buyers actually are. FoundLab members are notified before any re-weighting goes live.

No. Traditional SEO focuses on getting your website to rank on a Google results page. AI does not rank pages — it names experts. AI systems pull from completely different signals than Google search: review sentiment across multiple platforms, structured data, entity authority across LinkedIn and YouTube, directory completeness on Google Business Profile, Bing Places, and Apple Maps, and consistent cross-platform brand identity. Most SEO agencies are doing zero of: structured data same-entity links connecting your profiles, Realtor.com review orchestration, LinkedIn long-form posts optimized for LLM indexing, or the cross-platform consistency checks AI crawlers actually verify. Ask your agency to run a FoundScore audit on you. They cannot. That is our category.

Your existing coach probably teaches you how to sell, how to recruit, how to scale your operations, or how to think about your business. All of that work matters and we are not replacing it. The FoundLab focuses exclusively on AI Engine Optimization — making sure that when an AI platform answers "Who is the best agent in [your market]?", your name is in the answer. This is a different category of work: structured data, review distribution across platforms, entity authority signals, content systems that AI crawlers can actually index. Most coaches are operating on lead-generation playbooks built before AI search was a meaningful channel. We update our methodology quarterly because the platforms themselves change quarterly. If your existing coach is teaching you AI visibility at the depth we cover, you do not need us. We have not yet met one who is.

llms.txt is an emerging file standard — a plain-text document hosted at your site root (yoursite.com/llms.txt) that describes your business in structured-but-human-readable format: identity, markets, services, team, languages, reviews. AI crawlers that honor the spec read it directly and use it to fill gaps the more rigid structured-data format can't. Today, not every AI platform reads it. Hosting it doesn't penalize you anywhere. The platforms supporting it are growing month over month, and within a year it's table stakes. So: not a must-have right now, but get one generated. Host it on your site when your site supports a root-level text file (most do; some restrictive page builders don't). Get ahead while you can.

AI is the distribution; humans are the decision. Plenty of leads arrive having already done deep research: they ask a chatbot, land on your site, and read everything before they ever reach out. We've seen a lead read every agent bio on a team's website and hand-pick their agent before the first call. That's why the content can't be machine-bait — it has to hold up when a careful human reads it. The good news: the same qualities work for both audiences. Specific, concise, locally grounded, and consistent reads as authority to a crawler and as competence to a person.

A wall of five-star reviews that all say 'great to work with' does surprisingly little for your visibility. What moves the needle is detail: the town, the neighborhood, the specific service (relocation, a bidding war, a virtual inspection), and a steady, recent cadence. Those specifics are the keywords and the proof that AI and search engines read as real, current authority. Quality and specificity beat raw volume.

Answer Engine Optimization is a different discipline from social engagement. A typical content agency is measured on views, likes, and follower growth, which uses different keyword logic and content structure than making your business readable and answerable by AI. If you do outsource, hand your team the AEO methodology directly so they produce for AI crawlers, not vanity metrics.

Your raving fans and referral sources first. Then past clients. THEN cold leads. The reason is engagement quality — your Substack open rates and reply rates are signals Substack uses to decide how aggressively to distribute your future Issues. If you front-load your list with cold leads who never open anything, your engagement metrics tank and Substack stops boosting you. If you front-load with people who actually want to hear from you (raving fans, referral sources), you get strong early engagement, Substack rewards that with broader distribution, and the cold leads you add later inherit a healthier sender reputation. Sort your CRM by activity (most recent emails, most engaged), take the top 2,500-5,000, upload them. Add the rest in batches over the following weeks.

Source: Class 4, 27MAY2026.

Domain Authority is a score (developed and measured by Moz, used by every SEO tool) that predicts how well a website ranks in search. AI cares because high-DA domains get cited more confidently — the higher the score, the more willing AI is to surface and quote that source. Real estate agent websites typically score 5-25. Medium is ~95, Substack ~92, LinkedIn ~99. The approach: borrow the authority. Republish to a high-DA platform with a canonical tag pointing back to your blog (Medium does this automatically with Import a Story). Your domain keeps the credit. AI cites the canonical source. The high-DA platforms can’t outrank you for your own content. Backlinks from high-DA platforms (Medium, Substack, LinkedIn, LinkedIn Newsletter) also raise your own DA over time.

There are two ways to grow: by location (a second town) or by niche (luxury, relocation, first-time buyers, a language you speak). Location is the primary driver because almost everyone searches by place — “[town] realtor,” “moving to [town].” A niche rarely beats location on volume, but it's how you stand out in a crowded market and catch what a town name alone misses. Expand one vector at a time — stacking both at once splits your signal instead of compounding it. And give your core market four to six months to take hold before adding the next.

You don't need to be everywhere — you need to be consistent on the surfaces AI assistants read and cite. That's your own website first (the home for everything), then the amplifiers: Google Business Profile, YouTube and its Community tab, a syndication layer like Medium or Substack, and an open-web spot or two like X or BlueSky. The trap is spreading thin across ten platforms and going quiet on all of them. Let one good piece of content fan out across a handful you'll actually keep up — add more only once the habit's locked.

Free pitching platforms (Qwoted, Connectively) plus direct pitches to local outlets like Patch and your town paper are enough to earn periodic press mentions — and one quote in an independent outlet is among the highest-trust citations AI can find about you. Paid press release wires distribute your words, but AI weights what others write about you, not what you paid to syndicate about yourself. If months of free pitching produce nothing, refine the pitch: lead with a specific number or named neighborhood, fit the format the reporter asked for, and respond early in the window. Paid distribution is a last resort, not a starting point.

People moving to your market ask AI about life, not just listings: which towns have an easy train commute, what daycare options exist, why taxes are high and which towns soften the blow, where the beach or trail access is. Answering those questions on your Google Business Profile (and your blog, and the communities) makes AI associate you with living in your area — so when someone asks for an agent there, you're the entity it already connects to the place. Skip the hard call-to-action on these posts. Being the one who clearly knows the area is the call to action.

There's nothing wrong with helping a happy client capture the details that made their experience great: the neighborhood, the strategy, the outcome. Clients read it as a helpful guide, not a restriction. The line is simple: help them remember what actually happened; never put words in their mouth or invent specifics. Specific, true detail is exactly what makes a review valuable for search.

The best reviews come from clients who were set up to write them. Seed the expectation early, while the experience is fresh, and help them remember the specifics worth mentioning. A steady drip of 3 to 10 detailed reviews a month does more for your visibility than a one-time pile of generic five stars, because AI and search reward consistency, recency, and specific, keyword-rich detail.

Every response is another chance to reinforce your authority and your keywords. Reply on the native platform, wait roughly a day so it reads as thoughtful rather than automated, and re-state the specific service in your reply: relocation help, a virtual inspection, a tough negotiation, whatever it was. That restatement is the signal AI and search pick up. Note that some platforms like Realtor.com and Zillow don't allow agent responses, so focus your effort on Google and Yelp.

Paid ads on Facebook and Instagram target people by demographics and interests. Organic AI visibility is about being the answer AI gives when someone asks. They are separate funnels that coexist; running ads does not penalize or cancel your organic AEO work. Expect paid leads to need a longer nurture cycle than the intent-driven leads that find you through AI.

You don't need a finished website to start building AI visibility. Create content directly on high-authority platforms like Medium and Substack now; they get indexed and read by AI quickly. When your own site goes live, you point the canonical link back to it. Your website becomes the canonical home later, the syndication can start this week.

The FoundLab runs 8 weeks end-to-end with a deliberate structure. 7 weekly live classes (Wednesdays 3pm ET) cover one focused topic each — structured data, profiles, reviews, content systems, local authority, platform-specific optimization, and integration. After the 7th class, you get a 2-week implementation break to actually do the work. We re-run your audit during the break and reveal your new score at the wrap-up call. You also get weekly office hours with Raf (Tuesdays 2pm ET) for the questions that do not fit cleanly in a class, the AI Visibility Handbook (Founders-only — future cohorts will pay for it separately), our template library of structured data blocks and review automation and AI-optimized listing copy, a private cohort WhatsApp for peer support, and first-mover access to every tool we ship after the cohort. Your full FoundScore baseline audit is included up front. The re-audit at the end of the program is included. The proof that the work moved your score is the entire point.

The class itself is one hour per week (Wednesday 3pm ET). Office hours are an additional hour (Tuesday 2pm ET) but optional — show up only when you have a question that benefits from live discussion. Implementation work is where the real time goes, and that work belongs to your DOO, marketing director, or operations admin — not you personally as the agent. Plan on roughly 1 hour per day for your implementation owner during weeks 1–3 (heavy setup: claiming profiles, building structured data, restructuring content). After that, the work compresses to maintenance and habit (around 30 minutes daily). Your personal time commitment as the agent is roughly 1 hour per week of recording video content, once the AI tooling is set up to extend that into a daily posting habit. We teach the workflows that make this possible.

No. We audit, measure, and provide the exact roadmap to improve your AI discoverability, but we do not control the AI platforms themselves. AI platform behavior changes frequently. Deals between platforms and review sites shift quarterly. Algorithm updates can move scores in either direction overnight. And the work itself depends on your execution of the action plan we provide. For all of these reasons, we cannot and do not guarantee specific recommendations, lead generation, or business outcomes. What we do guarantee is that we will give you the most accurate possible measurement of where you stand, the most current playbook for moving the score, and direct access to us during the program if you join the FoundLab.

Some quick wins — like claiming missing directory profiles, fixing structured data same-entity links, or rebalancing your reviews across platforms — can move your score measurably within weeks. Building true AI authority across the full set of signals takes compounding effort over months. Every audit comes with a Jump Start one-pager containing your 3 highest-leverage immediate actions. Most agents who simply execute the Jump Start see a measurable bump within 30 days. For real movement, the FoundLab is an 8-week guided program. We re-run your audit during the program break and reveal your new score at the wrap-up — giving you a delta report showing exactly what moved, by how much, and where to focus next. We then continue to follow you for a full year with quarterly re-audits to track sustained growth.

AI assistants surface the names they see tied to a place often enough to predict the connection — so visibility tracks consistency, not a one-time push. Agents who publish steadily (a blog, a video, the platform-native pieces that follow) usually see early movement within a few weeks and meaningful pull within a few months. Treat it like a habit: make the one hard piece each week and let the associations stack. There's no overnight switch — anyone promising one is selling something.

Community answers compound slowly and then durably. One team posting one or two genuinely useful answers a week in a single local subreddit saw the first direct Reddit lead at roughly the three-month mark — and a consistent flow after that. The reason it keeps working: those answers never expire. They surface in Google's results and in AI-generated answers years after posting, so every answer you write is permanent inventory. Pick one active thread in your market, plant your flag, and hold the quality bar — one great answer beats ten rushed ones.

AI rarely closes the deal by itself; it starts it. A chatbot mentions you, the person searches your name, lands on your site, reads your reviews, and then reaches out. That means AI visibility and review velocity are not separate initiatives, they compound into one funnel. Strengthening either one lifts the whole path from first mention to signed client.

Attribution breaks if you ask too early; at the intake call leads just want the appointment. The right catch point is the agent-led consult, when the client already trusts you. Make 'so what made you reach out?' a standard beat in the script and log whether it was an organic search, social, or an AI recommendation. That is how you back into real attribution for AI-visibility work.

Capture the source at the BUYER CONSULTATION, not the intake call — and dig deep when you do. Don't settle for "I found you online" as the answer. Ask specifically: what exact search query did you type? Which AI did you use — ChatGPT, Gemini, Perplexity, Claude? What was the exact prompt you asked? The specificity matters because the common path right now is invisible upstream: AI mentions you in a chatbot answer → lead Googles your name → lands on your site → reads your reviews → calls. If you only ask "how did you find us," you'll get "Google" and miss the AI step. If you ask "what was the exact search," the lead might remember they asked ChatGPT for a buyer's agent in your market — and that's the source signal you want. If an ISA, VA, or admin is screening inbound, you'll lose this entirely because the lead just wants the appointment. Build the dig-deep question into your consultation script as a standard beat after they've sat down with you.

Source: Office Hours 3, 26MAY2026.

Keller Williams' Command platform is a managed website environment, which means KWRI deliberately limits how much structured data and custom HTML you can add. This is partly a tech-support cost decision on their side. The downside is that some of the most important AI visibility levers — structured data same-entity links, structured listings markup, llms.txt files — are difficult or impossible to implement on Command. The FoundLab covers everything you can do inside KW Command, including profile completeness, content publishing through the available channels, and review consistency. We will not pretend it is unlimited. For most agents past about 50 units of annual production, the right next move is to migrate to a platform with full structured data control. We help you scope that move during the program if it makes sense for your business.

Google allows up to 1,500 characters per post, but the sweet spot is roughly 300: long enough to say one specific thing, short enough for Google's AI to lift whole and serve as an answer in search or Ask Maps. Concise also wins with the humans — some leads still read everything on your profile before reaching out. One idea per post, a real local specific, a link if you have one, and out.

It's a tempting idea: park a copy of every video and post in your own subreddit so AI crawlers have one more surface to read. We tested it — and Reddit banned the subreddit, because a community used solely for one brand's links meets their definition of spam regardless of intent, and the risk extends to the posting account. The Reddit play that works and lasts is the one that takes effort: answering real questions in the communities where your buyers actually ask them, under your real name, with no links and no pitch. Those answers get indexed, cited by AI, and never expire. For an owned archive that machines can crawl, use surfaces you actually own: your blog, your Google Business Profile feed, your YouTube channel. Park content on property you own; participate on property you don't.

Next Door content lives behind a login wall, so AI crawlers never see it. Time spent there builds zero AI visibility — unlike Reddit and Quora, where answers are publicly indexed, surface in Google results, and get scraped into AI training data for years. If you personally generate business on Next Door, keep it as a lead-gen channel; just don't count it toward your AI visibility work. Put the content time where the machines can read it.

How you post a video changes who sees it. Native uploads, where you drag and drop the actual file, get favored by every platform's algorithm and let AI crawlers read the title, captions, and contact info burned into the video itself. Posting a link out to YouTube or your site instead usually gets the post deprioritized. Upload native, burn in captions and a soft call to action, and save links for the description or a later comment where the platform allows it.

LinkedIn tends to show posts with external links to fewer people because it wants to keep users on-platform. But if you are optimizing for AI visibility and protecting your domain authority, the link matters more than the engagement dip. Unless LinkedIn engagement itself is your goal, keep the link in.

Yes — for many real estate agents, Substack is a complete replacement. Every Substack Issue goes out as an email automatically to your subscriber list. You can upload your CRM emails directly (in batches — usually around 5,000 per upload, come back the next day for the next 5,000). No TCPA hoops if you already have a relationship with these contacts. Once they're in your Substack subscriber list, every future Issue fires automatically — same as your old newsletter tool, but free. Plus you get the engagement layer (Substack Notes) and the discovery-within-Substack-network upside. The newsletter is the main job; Substack does it cleanly without the monthly subscription. Cancel Mailchimp once the migration's clean.

Source: Class 4, 27MAY2026.

Yes, leave the old site up. Information on the internet should stay live as long as it's still correct — don't delete old URLs that AI has indexed. The move: put your new content on BOTH sites during the migration period (same blog posts, same structured data, just copy-paste-imported into the new platform), then add a clear link from the old site pointing to the new one. The old URLs stay in AI's index, and the canonical authority migrates over time as AI re-crawls and discovers the new site. Taking the old site down breaks the chain — every backlink, every AI citation, every reference to the old URL goes 404. Leave it up.

Varies per platform — each one has its own freshness-pattern rules for comments fired after publish. About 30 minutes is a safe directional answer on YouTube — pinned comments and first comments add indexable text to your video page, but YouTube has some weak pattern-detection for comments that fire instantly after publish (looks bot-ish). Other platforms (Instagram, LinkedIn, Facebook) have their own rules. That said, if you forget and drop the pinned comment immediately after hitting publish, it's not going to hurt you measurably on most platforms. The much higher-impact rule is making sure you DO drop a pinned comment within the first day — period. Most agents skip it entirely, which leaves a free indexable text block on the table. Volume of correctly-pinned-once-per-video beats perfect timing on any one.

Source: Office Hours 3, 26MAY2026.

Title and the first 150 characters of the description, primarily — those are the highest-weight surfaces AI scrapes. The first 50 characters of the title carry disproportionate weight, so front-load your primary keyword there. After title and description: tags, transcript, pinned comment, chapters, thumbnail text overlay, and channel-level metadata (description, keywords, linked accounts, playlists). Captioning helps but isn't the whole game. What moves the needle most is auto-generating titles and descriptions through AI prompts that bake in the target keywords — front-load your primary keyword in the first 50 characters of the title, write the first two lines of your description as a direct answer to a buyer's search query (the AEO snippet zone). Most agents stop at decent captions and never optimize the title/description, which leaves 80% of the indexable surface on the table.

Source: Office Hours 3, 26MAY2026.

It depends on your state realtor association and local office requirements. The brokerage-name rules that apply to yard signs and listing platforms vary state-by-state for channel names — some states are strict, some are not. YouTube isn't a universally monitored brokerage-compliance surface the way yard signs are. Check your specific state association and local office policy before you decide. The strategic reason we deliberately leave the brokerage off our own channel name (just Nancy Chu Homes — no Keller Williams): portability. What if we leave the brokerage? The channel doesn't have to be renamed, the back-catalog of videos doesn't have to be re-tagged, the entity model AI has built around our name stays intact. If your state association requires the brokerage in your channel name, abbreviations like KW Realty [City] are typically fine — common practice.

Source: Office Hours 3, 26MAY2026.

At least 3. Each playlist is a separate search index reference — when someone searches for a topic on YouTube, your video has three different chances to surface based on which playlist matched the query. The same video can live in a geographic playlist (e.g. Bloomfield Real Estate), a topical playlist (First-Time Buyer Resources), and a buyer-type playlist (Relocation Buyers) — it's not in a place, it's index-referenced from multiple places. Don't constrain yourself to just one home for a video. Three is the floor; five or six is fine if the categories are distinct.

Source: Office Hours 3, 26MAY2026.

Shoot horizontal — but compose your shots with the vertical crop in mind. Horizontal shooting gives you the most flexibility: you can reframe to 1:1 (square) or 9:16 (vertical) in your editor for short-form distribution, but you can't go the other direction. The composition trick: frame loose. Keep your subject (you) roughly center with breathing room on all sides — left, right, top, bottom — so the 9:16 reframe captures you cleanly without cutting off your head or losing the background context that gave the shot meaning. Tools like Opus Clip can auto-reframe with subject-tracking, but they work better when you've already left room for them to crop. If you're shooting something exclusively for short-form (Town Minute that's never going to YouTube long-form, one-off Reel, TikTok-native idea), shoot vertical natively.

Source: Office Hours 3, 26MAY2026.

Use a video editor that has caption-burn-in as a feature — Opus Clip, Descript, CapCut, Final Cut, Adobe Premiere, most modern tools support it. Burning captions in means baking the caption text into the video file itself — they become part of the visual frame, not a separate caption track. Do this BEFORE you export, even if you're not using auto-clipping features. Why it matters: when a short cross-posts from YouTube Shorts to TikTok, Instagram Reels, Facebook Reels, or LinkedIn video, each platform uses its own caption track — if you rely on YouTube's auto-CC, those captions don't follow the file across platforms. Burned-in captions persist everywhere. They're also readable by multimodal AI crawlers from the frames themselves, which is the AEO play.

Source: Office Hours 3, 26MAY2026.

Practice. Don't film the hook in one take with the rest of the video. Record it separately — three, four, five candidate hook lines, each 15 words or less, each shot fresh and direct to camera. Then during editing, slap the best one onto the front. The hook is in a weird space between a question and an answer — it has to create a question in the viewer's head fast enough to make them not scroll — and that delivery rarely lands on the same take as your narrative-style talking. Separate the production of the hook from the production of the body. Edit them together at the end. This gets easier with reps — by your 10th video you'll be writing the hook in your head while you're filming the rest.

Source: Office Hours 3, 26MAY2026.

Google is phasing out the traditional user-managed Q&A section on GBP. The rollout is uneven by region, but the direction is clear: that surface is going away. The replacement is "Ask Maps," an AI-driven Q&A that generates answers by scraping your website, reviews, GBP profile, and other indexed sources.

Don't chase the old Q&A surface. Treat it as gone. The play is simpler:

POST YOUR Q&A PAIRS AS GBP POSTS (the Updates section)

Each Q&A pair becomes its own Post:

1. Sign in to your GBP dashboard

2. Click "Add update" or go to the Updates section

3. Question as the opener: "Q: Best Bloomfield neighborhoods for first-time buyers under $700k?"

4. Answer as the body (200-700 chars). Named neighborhoods, named numbers, your take, your signature framings.

5. CTA button (Learn More) linking to a fuller version on your website's FAQ page

6. Publish

Why this works as well as the dedicated Q&A section did, possibly better:

- GBP Posts get indexed by Google immediately

- Same content feeds Ask Maps as a scrape source

- Five-minute-per-week habit

- One Q&A pair per week = 50 indexed Q&A Posts per year per market

Seed 5 to 10 high-intent pairs now (real questions buyers and sellers in your market actually ask), then add one new pair per week as part of your GBP cadence.

The cleaner mental model: GBP Posts is the surface. Your website FAQ page is the depth. Ask Maps is what scrapes both. You don't need to manage a separate Q&A section anymore.

This is one of the cleaner wins from Google's transition: the surface most agents ignored (Posts) is now the right home for Q&A content, and the surface most agents couldn't find (the dedicated Q&A) is going away anyway. The five-minute-per-week habit beats trying to seed a vanishing surface.

Real estate brokerages and agent teams often have a brokerage HQ that's different from where you actually work day-to-day. Realtor.com will frequently show your brokerage's main office because that's what MLS feeds. Zillow and your own site might show your agent office. AI sees both and tries to reconcile. The fix is to be transparent: your structured data can declare a primary location AND additional locations together, so AI reads them as the same person operating from both. Pick the address you can control on most platforms and align everything to it. Declare any secondary offices in the structured data block, and the outlier becomes a confirmed secondary signal, not contradicting drift.

Many agents go by a different name on their license than what they use professionally — maiden names, married names, anglicized first names, professional shortenings. AI scrapes every platform and sees the variations. If you don't tell it those variants are all you, it either interprets the differences as multiple people or as drift it can't trust. Structured data has name fields for exactly this case: declare the license name, the common name, the married or maiden version, any professional alternate you use. Being upfront makes it easier for AI to confidently connect all the records back to one person — you. Hiding the mismatch doesn't help. It just creates more drift.

Phone numbers are one of the easiest pieces of NAP drift to clean up because the format is entirely under your control. But the format matters — to an AI parser, (555) 555-1212, 555-555-1212, and +1 555 555 1212 are three different strings. Pick one format and use it on every platform where the field is open. Our recommendation: (999) 999-9999. Parens around area code, dash separator, no country code. Most common US format, reads as default. Where a platform forces its own format (some auto-add the +1, some strip parens), let it — that's the platform's normalization, not your drift. Where the field is open: pick and stick.

A canonical URL is the official, identifier-stripped link to a resource — the version search engines and AI treat as 'the real one.' Most URLs you copy from a logged-in browser session aren't canonical: they carry tracking parameters (the long ?utm_source=...&gclid=... tail) that platforms append to log where the click came from. Those tracking-wrapped URLs aren't your profile — they're a session of your profile. The clean process: open a private/incognito window, don't log in to anything, search for yourself on the platform (Zillow, Realtor.com, Yelp, Google Business), click into your profile as a stranger would, and grab the URL from the address bar. That's the canonical version. If you're unsure, paste it into ChatGPT or Claude and ask: 'Is this a canonical URL for [your name] on [platform]?' AI will tell you yes or no and clean it if needed.

Perfect NAP consistency is the goal, but it isn't always possible — Zillow and Google assign their own tracking numbers, call-tracking systems give every platform and campaign its own line, and some teams carry both a "Group" and a "Team" name. Don't burn months trying to kill every variant. List all of them in your master intake document and reconcile them in your site's structured data so the machine understands that every identifier points to the same business. IBM doesn't have one phone number either; what matters is that AI can connect yours. One caveat: if a rename is coming anyway (some states are pushing teams away from "Group" naming), start migrating now — every month you wait adds cleanup.

Password sharing is a security risk and a single point of failure. Google Business Profiles support roles: add a team member as a Manager so they can respond to reviews and post updates without ever touching your password or account recovery. Reserve Owner for yourself — full control and full responsibility. The same principle applies across your platforms: delegate by role, not by password.

No, the exact text doesn't matter as much as whether the accounts are connected in your structured data. Use whatever variation you want — your personal name vs your team brand, the short form vs the long form — as long as your website's machine-readable profile links them as the same entity. AI follows those connections: if your structured data says you're part of a specific team brand and your Medium account uses just your personal name, the link still resolves. Consistency across surfaces helps, but the difference between a short personal byline and a longer team-brand byline is negligible as long as both are linked as the same entity. Pick the variant that reads cleanest as a personal byline and keep moving.

Consistency is what matters more than the specific format. Pick the format you want (with the plus sign or without, with a comma or without) and use it everywhere — your blog, your bio, your structured data, your social profiles, your LinkedIn About. AI looks for matching strings across your entire entity graph to confirm the same fact is being claimed in the same way. Mixed formats (22+, 22 years, twenty-two years, over 20 years) read as four different claims even though you mean one. If you've already published with inconsistent formatting, go back and align. Five minutes of cleanup, real entity-graph payoff.

Validators return three signal levels: pass, warning, and error. Errors are real — fix them before deployment. Warnings are notes saying 'you're using a field that isn't strictly on the supported list for your declared type.' Sometimes that's fine (you're being more specific than the standard suggests), sometimes it's a typo or a deprecated field. The fastest way to triage is to copy the entire warning report and paste it back into the AI that generated your structured data. Ask it to interpret each warning and return a corrected, complete block. Re-run the validator. Zero warnings is nice; explained warnings you chose to keep are fine.

For most websites, structured data goes in one place — the global header (also called code injection, custom header, or site-wide head). That single declaration cascades to every page on your site automatically. You don't paste it 50 times for 50 pages. Some site builders are wired differently — KW Command, certain template systems, custom-built sites with per-page templates. If your website doesn't have a global-header injection point, you may need to add it to a master template or every page template (not every page individually). The one place it can't go: IDX-served pages (your property search, individual listing pages pulled from MLS) — you don't control those headers. That's fine. Everywhere else you control: same block, one declaration, cascades to every page.

Most websites already have some structured data in the header — sometimes from your site builder, sometimes from a Realtor.com badge, sometimes from an SEO plugin you forgot installed it. Don't just paste your new block on top — two competing declarations send mixed signals to AI. Instead: copy the existing block, paste both blocks into ChatGPT or Claude, and ask it to merge them into one canonical block, prioritizing your new content but keeping any unique fields from the existing block you'd want to preserve. AI is much better at structured-data reconciliation than manual stitching. Take its output, validate it, deploy. (Keep your tracking pixels — Google Analytics, Meta Pixel — those live alongside structured data, not replaced.)

Your structured data is a snapshot of your business at the moment you generated it. If something material changes — a new agent joins your team, you expand into a new market, a major award lands, a language capability gets added — re-run it. The change won't be reflected in AI's view of you until you regenerate and redeploy. For routine updates (one more closed transaction, one more 5-star review, minor tweaks to the bio), batch them and re-run quarterly. Once a quarter is a sane minimum cadence. Once a year is the floor. More often if your audit shows a drop in AI visibility.

Your machine-readable business data is not set-and-forget. Refresh it on a regular cadence so AI systems re-crawl and update their understanding of your business as your listings, reviews, and focus change. A quarterly refresh keeps the signals current without becoming a burden.

The slug is the part of your blog URL after your domain — for example, in nancychuhomes.com/blog/afford-home-north-jersey-stairstep, the slug is afford-home-north-jersey-stairstep. AI reads slugs as additional context about what's on the page. Keyword-rich slugs (afford-home-north-jersey-stairstep) outperform generic ones (blog-post-47 or untitled-12) in AI citation likelihood. Most CMS platforms let you override the auto-generated slug — find that field, replace it with a slug that includes your primary keyword + your market. Ask your AI for slug recommendations when generating the blog post: it'll suggest 2-3 options based on the post topic and your target keyword. Two-second move, real entity-graph value.

Source: Class 4, 27MAY2026.

Both work fine. AI reads HTML directly — that's actually its preferred format. The deciding factor isn't WordPress vs HTML; it's whether you can edit the site's header to add your structured-data signals. WordPress makes that easier through plugins or the theme header. Other platforms (including newer ones like Go High Level and Squarespace) typically expose a header-editing option too. Both end up in the same place: your machine-readable signals live in the header so AI doesn't have to read and parse your entire site to figure out who you are — they tell it directly.

A common worry is that too much information dilutes the signal. In practice, the opposite is true. AI models are built to filter — they read your full knowledge document, identify patterns, and pull the specific details they need for each piece of content. The agents who score highest spend hours on these documents because granularity pays off across every downstream piece of content the AI generates. Specific neighborhood names, transaction counts in specific markets, distinctive client-fit criteria, banned phrases you'd never use — every detail compounds. The diminishing-returns line exists somewhere, but in our experience no agent has actually hit it. Better to over-share and let the AI pick the golden nuggets than to under-share and watch the output go generic.

Your AI intake document is the source-of-truth your AI tools read whenever they generate content for you. New press appearances, new awards, expanded markets, team additions, language capabilities — they all need to land in the document for the AI to know they exist. The minimum cadence is once a year, treated like an annual planning exercise. The recommended cadence is incremental: when something new happens, drop it at the bottom of the document and let the AI reorganize from there during your next content generation. If your monthly FoundScore™ audit shows a dip, that's a signal to revisit your document, look for what's missing, and update before the next audit cycle.

A common temptation when claiming platforms is to add the link to your AI intake document immediately, before the profile has any content. Don't. AI engines crawl the linked profile when they evaluate your authority, and an empty Substack or zero-post LinkedIn newsletter sends a contradicting signal — it implies you're not actually active there, which downgrades the trust your other (filled) profiles built up. The fix: claim the profile, build out at least 3-5 pieces of real content on it, then add the link. The exception is identity-only platforms (your professional license registry, your brokerage profile, your headshot directory) where the profile itself is the signal — those go in immediately.

When agents hear "voice," they often think audible voice — and worry that AI tools can't capture it. They can't, and that's fine. What AI can capture is your written voice: the phrases you use repeatedly, the rhythm of your sentences, the words you reach for and the words you'd never use, the specific way you describe a market or a client. The most effective way to feed this to the AI: pick the video or audio that's most representative of how you actually talk, run the transcript through a tool like Turboscribe, drop the text into your AI project, and tag it explicitly as a voice sample. The AI will then mimic that voice across every piece of content it generates for you — text messages, blog posts, scripts, email replies. Audible voice belongs to avatar tools (HeyGen, 11 Labs). Written voice belongs in your intake document.

Each AI tool builds a memory of you over time — your communication style, your business, your preferences, your past projects. When you switch tools or add a second one, you don't have to start from scratch. The migration prompt is straightforward: in the tool you're leaving, ask "Output everything you know about me and how I like to work, formatted so I can paste it into a new AI tool." Copy the response. In the new tool, paste the response, then drop in your AI intake document. Within a few exchanges the new tool will be operating at the same level of familiarity as the one you left. We recommend keeping multiple AI tools in rotation — different models surface different angles on the same prompts, and the migration step costs you nothing.

CRITWE is the five-part discipline we use and teach for writing prompts that actually produce useful output. Context, Role, Interview, Task, With Example. Context — the background the AI needs about your situation. Role — who the AI should be (persona, expertise, perspective). Interview — force the AI to ask clarifying questions before producing. This is the step most prompting frameworks (RTF, CRAFT, CO-STAR) skip. Don't skip it — the cost of one clarifying question is far less than the cost of generating 500 wrong words. Task — the actual ask, specific and scoped. With Example — show the output format you want; one concrete example beats five sentences of description. Applied consistently, CRITWE is what separates 'AI generated something' from 'AI generated something useful.' The canned prompts in our program are CRITWE-structured under the hood — when you go beyond the canned set, write your own using CRITWE and you'll get the same quality.

You can build a complete AI content stack for a fraction of what most agents assume. HeyGen handles avatar video, ElevenLabs gives you natural-sounding voice, and Opus Clip auto-cuts long videos into shorts, all with inexpensive starter tiers. Start on the basic plans, learn the workflow, and only upgrade when a token or credit limit actually stops you. Run this way, the entire toolset stays under roughly $2,000 a year.

No. They're different artifacts. Your video should be filmed from bullet points, not a word-for-word script — reading from a script makes you sound robotic on camera. AI generates the blog post separately, from your video's transcript (the actual words you said). There will be drift between any script you started with and the transcript that gets generated — that's expected. You can also add more context to the blog that you didn't say in the video. Throw the original script into your AI prompt as additional context if you want, alongside the transcript. The AI synthesizes both into the blog. The transcript is the source of truth for what's in the video; the blog can expand beyond it.

Source: Class 4, 27MAY2026.

Habit-stacking. Nancy spent a good deal of time stacking videos onto existing routines in the beginning because she also didn't enjoy it. Now it's second nature — she records without makeup, looking like she just rolled out of bed, and doesn't care anymore. The content is what matters, not the production polish. Build the recording into something you already do every week (Monday morning coffee, Friday afternoon market review, whatever) so the trigger fires automatically. The first 3-4 videos are the worst. Past that, it gets easier. Don't quit at video 2.

Source: Class 4, 27MAY2026.

Both, for different jobs. TubeBuddy or VidIQ or whatever keyword research tool you prefer — they're engagement-focused. They rank keywords by how likely they are to drive audience views, surface in YouTube's own search, and grow your subscriber count. Right tool if your priority is human discovery on YouTube specifically. Claude (or any capable LLM) is the right tool for pure AI-driven SEO/AEO purposes — being cited and recommended by ChatGPT, Gemini, Perplexity, Claude when someone asks best real estate agent in [market]. Claude understands AI crawlers better because it IS an AI. If you can only pick one, pick Claude — the AI-citation game compounds longer than the YouTube-view game. If you can afford both, run them in parallel: Claude writes the title/description/tags optimized for AI crawlers; TubeBuddy or VidIQ validates the keywords for human-search demand.

Source: Office Hours 3, 26MAY2026.