Complete Guide to Building Your Realtor Daily Brief with Claude (2026)

Why Every Realtor Needs a Daily Briefing in 2026

The default state of a real estate agent’s morning is reactive. Open email, scan for fires, respond to whatever shouts loudest. The texts and calls that came in overnight set the tone. Whoever was most demanding gets the first hour. By the time the agent thinks about the deals that actually move the business forward, half the morning is gone.

A daily brief flips that order. Instead of starting the day inside the inbox, the agent starts with a one-page synthesis: today’s calendar with conflicts flagged, this week’s pipeline with the deals most at risk, the three contacts most likely to convert if touched today, the listings that changed status overnight on the watchlist, and a one-line market pulse. The inbox still gets opened. It just gets opened second.

The compounding advantage shows up in three places. First, response time on hot leads. If the brief catches a lead who toured a property at 8 p.m. last night and never got a follow-up, the agent calls before the competition does. Second, deal protection. A pipeline view that flags “Buyer X has been in inspection for 9 days, average is 5” gives the agent a chance to call the inspector before the option period runs out. Third, opportunity discovery. New listings that match an active buyer’s criteria, a price reduction on a watched comp, a closed sale that just reset the comps for an active seller listing. None of this is exotic information. All of it sits in tools the agent already pays for. The brief just stops it from getting buried.

The Beginner’s Guide to AI for real estate agents covers the broader case for why agents who adopt AI early in 2026 will outpace those who wait. For deeper context on Claude as the specific platform, the post on why Claude beats ChatGPT for real estate workflows walks through the reasoning. The short version: Claude’s MCP support, longer context window, and willingness to follow detailed instructions without drifting make it the right engine for a structured daily report.

What Belongs in a Realtor Daily Brief

A good daily brief is short. The temptation is to cram everything in, which produces a wall of text the agent skims and stops opening within a week. Aim for 200 to 350 words on the page, organized into clear sections. The agent should be able to read the whole thing in 90 seconds and still come away with three concrete actions for the day.

The sections that pay off in practice, ranked roughly by how often agents end up using them:

The “Top 3 priorities” section is where the brief earns its keep. It is the only section Claude generates with judgment instead of just data. Everything else is a fielded readout. Priorities ask Claude to look across all the sections and rank what matters most for revenue this week. Done well, this is the line the agent reads twice.

Two sections that look useful but rarely deliver: full email summaries (the inbox is its own discipline, and a brief is not the place to triage 47 newsletters), and detailed market data (anything beyond a single line is noise unless the agent’s day involves negotiating today). Cut both unless you have a specific reason to keep them.

Tools You’ll Need (and What’s Optional)

The minimum viable stack for a daily brief is small. Three things, plus a CRM the agent already uses.

Required. A Claude paid plan (Pro at $20/month, Max at $100 or $200/month, or Team/Enterprise). The free tier does not include MCP connectors or projects, which are the two features that make this whole approach work. Claude Desktop installed on the computer the agent uses each morning. A Follow Up Boss account or equivalent CRM with API access. Most modern real estate CRMs qualify; if a CRM has no public API, the brief can still work, but pipeline data has to be pulled manually or skipped.

Strongly recommended. A Google Workspace account (Gmail and Google Calendar) for the calendar and inbox connectors. An MLS subscription or VOW feed if the agent wants automated listing watchlists. The MCP standard is the connector layer Anthropic built that lets all of these talk to Claude using a shared protocol; the official specification lives at modelcontextprotocol.io.

Optional, advanced. Claude Code, the CLI version of Claude, for agents comfortable in a terminal. Cloud Routines, Anthropic’s hosted scheduled execution environment, for agents who want the brief delivered at 6:30 a.m. whether or not their laptop is open. Both are covered later in this guide. Neither is required to get started.

Claude Plan Comparison for Daily Brief Use

Most solo agents can stay on Pro for the first 60 days, run the brief manually each morning by typing “good morning” into the project, and only consider Max or Team once they have a workflow that earns its keep. Do not start by paying for the most expensive plan. Start by proving the brief is worth opening.

Setting It Up in Claude Desktop

Claude Desktop is the application Anthropic ships for macOS and Windows. It looks like a chat window. Inside that window, two features matter for building a daily brief: Projects and Connectors. A project is a saved workspace with its own system prompt, file uploads, and conversation history. A connector is an MCP server attached to Claude that gives it tool access to live data.

The setup, in plain steps:

1. Install Claude Desktop from claude.ai/download if not already installed.
2. Sign in with the paid Claude account.
3. Click the “Projects” tab in the left sidebar. Click “New Project.”
4. Name it something the agent will recognize, “Daily Brief” works fine.
5. Add a project description and a system prompt. The system prompt is the durable instruction Claude reads every time the agent opens this project. The master prompt template in the next section goes here.
6. Click the “+” button at the bottom of the chat box and select “Connectors.” Add the connectors the brief needs: Google Calendar, Gmail, Follow Up Boss MCP. Each will walk through its own OAuth flow.
7. Close and reopen the project. Type “good morning” and watch what comes back.

The first run will be ugly. That is expected. The system prompt has not been tuned to the agent’s market, the connectors may not have the right scopes, and Claude does not yet know what the agent considers a “hot lead.” Do not fix all of this on day one. Get something running end-to-end first. Tune later.

The official Anthropic documentation on local MCP servers is at the Claude Help Center. Read it once before adding the first connector; the OAuth flows can be confusing the first time through.

The Master Prompt Template

This is a starter prompt. Copy it, paste it into the project’s system prompt field, and modify every section to match the agent’s market, pipeline, and priorities. The exact wording matters less than the structure. Claude does best with explicit instructions, named sections, and clear formatting requirements.

You are my real estate daily brief assistant. Every morning I will say
"good morning" or similar. When I do, produce a brief with the exact
sections below, in this order, with these headings.

Total length target: 200 to 350 words.
Tone: direct, scannable, no filler. Use bullets where appropriate.
If a section has no signal worth surfacing today, write "Nothing flagged."
and move on. Do not pad.

Sections:

1. TODAY (calendar)
List today's events in chronological order. Flag any back-to-back
appointments with less than 30 minutes drive time as a CONFLICT.
Include prep notes for any showing or listing appointment.

2. TOP 3 PRIORITIES
Rank by revenue impact this week. Each priority is one sentence and
references a specific contact, deal, or listing by name. The first
priority should be the single highest-leverage thing I can do today.

3. PIPELINE AT RISK
From Follow Up Boss, list any deal that has been in its current stage
longer than 5 business days. For each, name the contact, the stage,
and the days stuck. Maximum 5 items.

4. HOT LEADS TO TOUCH
From Follow Up Boss, list contacts tagged as hot or with last activity
in the past 7 days who have not been contacted by me in the past 3
days. Maximum 5 items. For each, give a one-line reason.

5. NEW LISTINGS (BUYER MATCH)
List any new listings hitting the market in the past 24 hours that
match active buyer criteria. Skip if no buyer matches.

6. STATUS CHANGES (WATCHLIST)
Price changes, pendings, withdrawals on listings I'm watching.
Skip if nothing changed.

7. MARKET PULSE
One sentence. Median price, days on market, or rates if relevant.
Source the number when you can.

8. INDUSTRY SIGNAL
Up to 3 items. Only items that affect my market, my buyers, or my
service area. No general AI hype unless it ties to real estate.

Rules I always want followed:
- Do not draft client emails. I write those myself.
- Do not commit to actions on my behalf in any external system.
- If you are unsure about a fact, say so and ask.
- Today is [INSERT YOUR DATE FORMAT PREFERENCE].
- I serve [INSERT YOUR MARKET, e.g. "the Austin metro and Texas Hill
  Country"].
- My current revenue priorities are [INSERT 1-2 SENTENCES, e.g.
  "list-side activity in Lakeway and Bee Cave; buyers in the
  $500k-$900k range across the western suburbs"].

Three things to notice. First, the prompt names every section by an ordinal number and an all-caps heading, which makes Claude’s output predictable and easy to scan each morning. Second, the prompt explicitly forbids drafting client emails, a guardrail covered in the privacy section below. Third, the bottom three lines (date format, market, priorities) are the parts the agent has to fill in personally. Without those, the brief will produce generic output. With them, the brief becomes specific to the practice in a way no off-the-shelf tool can replicate.

Plugging In the Follow Up Boss MCP

For agents on Follow Up Boss, the FUB MCP server is the single highest-value connector. It exposes contacts, deals, stages, notes, calls, texts, and emails as tools Claude can call. Once installed, the agent can ask questions like “which contacts haven’t been touched in 14 days but viewed a listing in the last 7?” and Claude will answer using live data.

Two FUB MCP implementations are publicly available. The first, an open-source 157-tool server that covers 100% of the official FUB API plus convenience tools for common workflows, was released by our founder, Ed Neuhaus, and is documented on the FUB MCP server page. The second is a community-maintained 34-tool server. Either works for a daily brief; the larger one is more flexible if the agent wants to extend the brief later into automated tagging, list cleanup, or pipeline reports.

Setup, at a high level, requires a FUB API key (generated from the FUB account settings under “Integrations”), a Claude Desktop install, and a configuration file edit to register the MCP server. The full step-by-step process is covered in AI Real Estate CRM: How to Make Follow Up Boss Actually Work for You. Plan on 30 to 60 minutes for the first install. Plan on zero minutes after that, ever.

Sample questions worth running once the connector is live (these are the kinds of prompts that flush out problems in the prompt, the data, or the agent’s own pipeline hygiene):

• “List every contact in stage ‘Nurture’ with last activity older than 30 days. Group by source.”
• “Which deals have been in ‘Under Contract’ longer than 14 days? Show contract date, close date, and days remaining.”
• “Show me hot leads tagged ‘buyer’ who have not received a touch from me in the past 5 business days.”
• “Pull every contact added in the last 90 days where ‘lead source’ is empty. List them.”
• “Which deals are scheduled to close this month? Who has not signed all required documents yet?”

The first time an agent runs queries like these, the output is uncomfortable. There are always contacts who fell through the cracks, deals stuck in stages they should not be stuck in, and lead-source fields that nobody filled in. That discomfort is the point. The brief surfaces it. The agent fixes it. By the third week, the questions return cleaner answers because the underlying CRM is cleaner.

Plugging In Calendar and Gmail

The Google Calendar MCP server is one of the simplest connectors to install. Anthropic has an official remote server hosted at https://calendarmcp.googleapis.com/mcp/v1, configured through Claude’s connector UI with an OAuth client ID. Claude Pro, Max, Team, or Enterprise plans are required. Once authenticated, Claude can read the agent’s calendar, list today’s events, identify conflicts, and pull event details into the morning brief.

Gmail integration follows a similar pattern. Several community-maintained Gmail MCP servers exist; pick one with active maintenance and good documentation. The most useful Gmail capabilities for a brief are:

• List unread messages since 5 p.m. yesterday from any sender flagged as a client, lender, title rep, or co-agent.
• Surface any email with words like “offer,” “counter,” “inspection,” “closing,” or “extension” that came in overnight.
• Ignore newsletters, marketing blasts, and platform notifications. Tell Claude explicitly which senders to ignore and which folders to skip.

The brief is not the place to triage the entire inbox. It is the place to surface signals that affect today’s decisions. A useful Gmail prompt addition for the system prompt:

For Gmail integration:
- Only surface messages from real client, lender, title, or transaction
  threads.
- Ignore promotional, marketing, listing alert, and platform notification
  emails.
- Never quote more than one line from any email in the brief. The brief
  is a pointer, not a substitute for reading the message.
- Do not draft replies. Surface, do not write.

Some agents prefer to skip Gmail integration entirely and let the inbox stay separate from the brief. That is a reasonable choice, especially for agents whose email volume is low or whose CRM already captures the important client communications. The brief works without Gmail. It just works better with it for high-volume practices.

Plugging In MLS and Market Data

This is the section where most agents hit a wall, and it deserves an honest answer. As of 2026, almost no MLS in the United States exposes a public MCP server. ABoR (the Austin Board of Realtors) and a handful of others provide VOW (Virtual Office Website) data feeds to brokerages, and a brokerage with engineering capacity can build an MCP server on top of that VOW feed. Neuhaus Realty Group runs one such server at mls.neuhausre.com for the Austin Central Texas market, which exposes the same listing data displayed on neuhausre.com programmatically through MCP. That is the architectural pattern: the data is not for sale and is not being sold; the same data the brokerage already publishes on its public site is delivered through a programmatic interface for the agents and clients with permission to access it.

For agents whose MLS does not have an MCP yet, three workarounds are working in 2026:

1. Email-based watchlists. Most MLSs let agents save searches and send daily emails. Configure the search, route the email into a labeled Gmail folder, and let the Gmail connector summarize the folder each morning. Coarse, but functional.
2. RSS or webhook feeds. Some VOW providers expose feeds that can be polled. A small Python script (or a Zapier workflow) can normalize the feed into a JSON file Claude reads from local disk through the file-system MCP.
3. Wait. The MLS MCP problem will be solved within 12 to 24 months. Several brokerages, vendors, and individual developers are working on it. Agents who set up the rest of the brief now will be ready to plug in MLS data the moment it becomes available in their market.

For Austin-area agents and clients, the existing MCP at mls.neuhausre.com is documented at the Austin MLS MCP page. Beyond Austin, the post on why every real estate agent should care about MCP covers the broader landscape and what to watch for as the standard rolls out.

Scheduling It: From Manual to Autonomous

The simplest way to run a daily brief is manually. Open Claude Desktop, click into the project, type “good morning,” read what comes back. This works. Many agents will never need anything more. The system prompt and connectors do all the work. The agent provides the trigger.

For agents who want the brief delivered without a manual trigger, three levels of automation exist as of 2026, each with tradeoffs.

Level 1: Claude Code /loop (Session-Scoped)

Claude Code, the command-line version of Claude, supports a /loop command that re-runs a prompt on an interval inside an open session. Useful for short-running checks (“every 30 minutes, check if the deployment finished and tell me what happened”) but not designed for daily briefs. The loop dies when the session closes. Skip this for the morning brief use case unless the agent already lives in Claude Code.

Level 2: Desktop Scheduled Tasks

Desktop scheduled tasks, available in Claude Code v2.1.72 and later, run locally on the agent’s machine. Each scheduled run fires a fresh session at the time and frequency configured, with full access to files, MCP servers, and connectors. For a daily brief, set the schedule for 6:30 a.m. weekdays, point it at the brief project, and have the output written to a markdown file the agent opens with their first cup of coffee.

The catch: the laptop has to be on and awake at the scheduled time. For agents whose machines sleep overnight, this means setting a “wake on schedule” preference in macOS or Windows power settings. Not difficult, but worth confirming the first morning.

Level 3: Cloud Routines (Hosted)

Cloud Routines, launched by Anthropic on April 14, 2026 as a research preview, run on Anthropic’s cloud infrastructure. The laptop can be closed, asleep, on a plane, or replaced; the routine keeps running. Pro accounts get 5 routine runs per day, Max gets 15, Team and Enterprise get 25. A morning brief uses one run per weekday, well within all paid tiers.

Setup is documented in the Claude Code scheduled tasks documentation. Routines support the same MCP connectors as Desktop, with the caveat that any local-only connector (one that talks to a database on the agent’s laptop, for example) will not work in the cloud. Most real estate use cases (FUB, Calendar, Gmail) work fine because all the data lives in cloud APIs anyway.

Routines auto-expire 7 days after creation unless renewed, which prevents forgotten loops from running forever. For a daily brief that the agent intends to keep running, set a calendar reminder to re-create the routine weekly until Anthropic ships permanent recurring tasks.

Output Delivery: Where the Brief Lands

An autonomous run is only useful if the brief lands somewhere the agent will see it at 6:45 a.m. Options:

• Email to self. Have the routine call the Gmail connector and send the brief as an email to the agent’s primary inbox. Simple, durable, works on the phone before the agent even sits down.
• Local markdown file. Have the routine write the brief to a file on the desktop. The agent opens the file with morning coffee. Works for Desktop scheduled tasks but not Cloud Routines (no shared file system).
• Slack or Microsoft Teams DM. If the agent or team uses one of these, an MCP connector can post the brief into a private channel or DM. Useful for teams who want the brief shared with an assistant or transaction coordinator.
• Push to a static URL. Advanced. Some agents host their brief on a private webpage they bookmark on their phone. Overkill for most.

Email-to-self is the right starting point. It works on every device, requires no extra infrastructure, and is impossible to forget about because it lands in the same place every other notification does.

Iterating: The First Three Weeks

The brief will be wrong before it is right. This is normal, predictable, and the most important part of the build. Plan for three weeks of tuning before the brief becomes something the agent looks forward to opening.

Week 1: Too Noisy

Day one’s brief will surface everything. Every contact who looks vaguely relevant. Every deal in any stage. Every email from anyone. The brief will be 800 words, and the agent will read maybe 200 of them. That is fine. Spend the week noting which sections delivered value and which were skim-and-skip. Cut the dead sections from the system prompt. Tighten the criteria for “hot” and “at risk” by raising thresholds. If “hot leads” returned 18 names on day one, change the criteria so it returns 5.

Week 2: Missing Signal

Once the noise is cut, week two reveals the opposite problem. The brief misses something the agent expected. A lead the agent considers hot doesn’t show up. A listing change the agent cared about gets ignored. A calendar event without a description gets flagged as a conflict. Each miss is a prompt edit. Write the missing scenario into the system prompt as a rule. Most weeks two end with the system prompt doubling in length and getting more useful in the process.

Week 3: Dialed In

By week three, the brief opens to the same length most days. The agent reads it in 90 seconds and has three things to act on by 7 a.m. The system prompt has stopped changing daily. The connectors are stable. The brief becomes the kind of background tool the agent stops noticing, until one morning it surfaces something genuinely valuable and earns its keep for the quarter in a single line.

Beyond week three, the iteration cycle becomes monthly rather than daily. Things that change a brief over time: market conditions (different priorities in a buyer’s market versus a seller’s market), business model shifts (an agent moving from buyer-side to listing-side will tune for different signals), new tools coming online (an MLS MCP becoming available, a new connector launching), and personal workflow changes (the agent who hires an assistant will route different sections of the brief to that assistant).

Privacy, Ethics, and Compliance

The brief touches client data. That changes what the agent can do with it, where it can run, and what controls the agent owes the people whose information is in it. The rules below are not optional. They protect the agent’s license, the brokerage’s reputation, and the trust the clients placed in the agent when they signed an agreement.

Do not put client PII in shared or team projects. Claude projects can be private (visible only to the agent) or shared with a team. Never put a project that pulls live FUB data, Gmail content, or calendar details with client names into a shared project unless every member of that team is bound by the same fiduciary duties as the agent. For a solo agent, this is automatic. For a small team, it requires explicit thought about who has access to what.

API keys and OAuth tokens are secrets. The FUB API key, the Gmail OAuth token, and the Calendar credentials all live in configuration files on the agent’s machine or in Anthropic’s cloud (for Cloud Routines). Treat them like the password to the brokerage’s bank account. Do not paste them into a chat. Do not email them. Do not commit them to a public GitHub repository. If a key is accidentally exposed, rotate it immediately from the source system (FUB account settings, Google account security page) before doing anything else.

Do not auto-send AI-drafted client communication. The master prompt template above explicitly forbids Claude from drafting client emails. This is intentional. Real estate is a relationship business, and the agent’s voice, judgment, and timing are what the client is paying for. Claude is excellent at surfacing signal. Claude is mediocre at writing the kind of email that holds a deal together when the buyer’s lender just delayed funding by four days. The agent writes those emails. Claude can help the agent prepare for them. Anything more invites a compliance failure and a damaged relationship at the same time. The post on the first casualty of AI in real estate covers the broader argument.

Disclose AI use per TREC and firm policy. The Texas Real Estate Commission and most state regulators are still working through how AI use should be disclosed in agent practice. The conservative position, and the one most firms are adopting, is to disclose that AI tools assist the agent’s workflow without replacing the agent’s professional judgment. Neuhaus Realty Group’s full AI Disclosure and Communication Policy is published on the public site and offers a usable template for solo agents and small teams who need their own policy. Adapt the language; do not copy it without thinking.

Document the architecture once. Write down which connectors are installed, which keys are where, and what the system prompt looks like. Store this in the agent’s secure password manager or a document the agent’s broker has access to. The brief should not be a system that breaks the day the agent loses their laptop.

Sample Brief Output

Here is a sanitized example of what a polished daily brief looks like in week three. Names, deals, and numbers are fictional. Format and length are representative.

Notice the length: about 290 words. The agent reads it in 60 to 90 seconds. By 6:50 a.m. the agent knows that Marcus Hoffman is the first call, the Bee Cave inspection needs scope confirmation today, and the 3:30 showing has zero buffer if the title coffee runs long. None of this is rocket science. All of it would otherwise have been discovered in pieces over the first three hours of the day, with at least one item slipping until the afternoon.

Common Mistakes to Avoid

Most failed brief implementations fall into a small number of patterns. The mistakes below are the ones agents repeat most often.

Building the brief before cleaning the CRM. The brief is only as good as the data underneath it. If FUB has 2,400 contacts with no tags, no stages, and no last-activity dates, the brief will surface noise. Spend a week tagging hot leads, assigning stages to active deals, and deleting dead contacts before plugging the FUB MCP into Claude. The brief will reward the cleanup tenfold.

Trying to make the brief draft emails. Already covered above. Worth repeating. The brief surfaces. The agent writes. The moment Claude starts drafting client communication, the agent’s voice goes flat and the relationship erodes. Resist this even when it would save 10 minutes a morning.

Setting and forgetting the system prompt. The system prompt is a living document. Markets change. Pipelines change. Priorities change. An agent who is doing $300K in production today should not be using the same brief prompt 12 months from now. Edit it monthly at minimum.

Not deciding what to ignore. A useful brief is more about what it leaves out than what it includes. Newsletters, social media, generic AI hype, market data older than 7 days, contacts who clearly did not engage, listings outside the agent’s service area. All of this is signal pollution. Tell Claude explicitly what to ignore.

Skipping the privacy review. The brief reads client data. That data has to be protected at the same standard the brokerage protects every other piece of confidential information. Do the review once, write down the architecture, lock the credentials.

Treating the brief as a finished product. The brief is a workflow tool, not a deliverable. It is supposed to evolve. Agents who build a perfect brief in week one and never touch it again will watch it slowly become irrelevant as their practice changes around it.

What Comes Next

A daily brief is a foundation, not a ceiling. Once the brief is stable, the same architecture extends to other workflows. Weekly pipeline reviews. Monthly farming reports. Listing performance dashboards. Buyer search briefs sent directly to clients (with permission). Each of these is a variant of the same pattern: a Claude project, a system prompt, a few MCP connectors, and a scheduled trigger.

Two specific extensions worth thinking about as the brief matures:

The end-of-day debrief. A 5 p.m. counterpart to the morning brief that reviews what got done, what slipped, and what needs to roll over to tomorrow. Pulls from the same connectors plus the day’s sent email and call log. Useful for agents who want to close out the day cleanly and start tomorrow’s brief with cleaner inputs.

The weekly pipeline review. A Monday morning roll-up that goes deeper than the daily brief: full pipeline by stage, conversion rates, deals lost in the past 30 days with reasons, lead source ROI. This is the brief that gets used in a brokerage one-on-one or a team standup. Same architecture, different schedule, different output format.

The agents who will outpace their peers in 2026 are not the ones who buy the most expensive AI tools. They are the ones who build a small set of personalized workflows like the daily brief, iterate on them weekly, and stop wasting the first hour of the day on inbox triage. The tooling has stabilized. The connectors exist. The protocols are open. The path from “no daily brief” to “brief opens at 6:45 every weekday” is roughly two afternoons of setup and three weeks of iteration. That is a small price for an extra hour of focus per day for the next decade.

Building Yours

The agents who win the next five years of real estate are not the ones with the flashiest tech stack. They are the ones who put 30 minutes into something like this in May 2026 and have it running for them every weekday for the rest of the decade. The brief is small, durable, and compounds. It does not generate listings on its own. It does not replace cold calls. It does not write contracts. It just makes sure the agent starts the day looking at the right three things instead of the loudest three things.

This guide showed the architecture and the patterns. The exact tools, prompts, and schedules an agent picks are personal. So is the iteration. Build a first version this week. Read it for three weeks. Edit the prompt every Friday. By the end of the month, the brief will not look like the example above. It will look like something tailored to one specific practice, in one specific market, focused on one specific set of priorities. That is the right outcome. The brief is not a product anyone sells. It is a habit, scaffolded by tools that finally caught up to what real estate work actually requires.

For more on the broader case for AI in real estate, see The Beginner’s Guide to AI for Real Estate Agents in 2026, the post on how to run a real estate business every day with AI, and the deeper dive on seven ways to use Claude beyond the chat box. For the full library of practical guides, the Complete Austin Real Estate Guides hub collects every long-form resource Neuhaus Realty Group has published. For the firm’s policy on AI use, the AI Disclosure and Communication Policy covers what gets disclosed and why. Build yours. Iterate. Then come back in 90 days and read this guide again with the brief running. Most of it will read differently the second time.