Every bad AI output you've gotten was a briefing problem.
You asked for a renewal email and got something addressed to "Valued Resident" at "The Property." You asked for ad copy and got a sentence about "luxury living redefined," which is what a model writes when it knows nothing about your asset except that it has walls. The model wasn't being lazy. You gave it nothing, and it filled the gap with the average of every apartment website ever scraped.
The fix is one document. You write it once, you paste it at the top of every session or drop it into a project's custom instructions, and every output after that knows your unit mix, your comp set, your concession policy and the fact that your brand voice does not use the word "sanctuary."
Call it a context file. It's the highest-leverage hour you'll spend this quarter, and it's mostly copy-paste from documents you already have.
What actually goes in it
Six sections. Keep it under about 1,500 words per property group, because context that's too long gets skimmed by the model the same way your 40-slide deck gets skimmed by asset management.
1. Portfolio roster. One line per community: name, city and submarket, unit count, year built, class, unit mix with square footage ranges, current occupancy and current pre-leased or trend number. Add a one-clause identity note. Not marketing language. Facts a leasing person would tell you.
> Ashline Row, Charlotte NC (South End), 248 units, 2019, A-, studio 520-580 / 1BR 690-810 / 2BR 1,050-1,190. 93.1% occupied, 95.4% trend. Podium building, no washer-dryer in studios, structured parking $125.
2. Brand and voice rules. Two lists. Words you use, words you don't. Then two or three sentences of actual approved copy pulled from something you've already published, because showing beats describing. If your brand is warm and plainspoken, paste a paragraph that is warm and plainspoken. The model pattern-matches on the sample far better than on the adjective.
Also write the banned list honestly. Luxury. Sanctuary. Oasis. Nestled. Boasts. Whatever your team keeps quietly deleting from vendor drafts.
3. Pricing and concession posture. Current effective rent ranges by floorplan, what concession is live at each property, expiration date, and the rule about who can quote what. This is the section that saves you the most rework, because AI-drafted campaign copy that promises six weeks free when you're offering four is a problem you find out about from a resident.
4. The channel stack. Which ILS contracts you hold at each property and what you pay. Your CRM and PMS. Your website vendor. Your call tracking. Your review platforms and current ratings. Your chatbot or leasing AI, if you run one. When you later ask for a lead-source analysis or a budget reallocation memo, the model needs to know you're on Entrata and not RealPage, and that Zillow Rentals is only live at two of your assets.
5. Audience and demand drivers, described as demand. This is where you write down the employers, universities, hospitals and transit lines that generate traffic. Write it as demand geography, not demographics: "two miles from the medical center campus, 15 minutes to the airport, light rail stop at the corner." Not who you imagine living there.
6. Reporting context. Who you report to, what they ask about, what the standing metrics are. Cost per lease, lead-to-tour, tour-to-lease, days on market, exposure. If asset management wants a Monday number, name it. Outputs get sharper when the model knows the memo lands on a specific desk.
The fair housing line, and it's a real one
Section 5 has teeth. Ask a model for ad copy and it will cheerfully produce "perfect for young professionals" or "ideal for a growing family," because that phrasing is everywhere in its training data. Both describe the resident. Under the Fair Housing Act, advertising that indicates a preference based on familial status, or that steers by implied demographic, creates exposure regardless of whether you meant it.
Write the rule into your context file in plain language: describe the unit and the location, never the person. "Two bedrooms, 1,050 square feet, in-unit washer-dryer, four blocks from the light rail" is compliant and, frankly, converts better because it answers a question a renter actually has. Include your instruction verbatim so it applies to every output: Describe the apartment, the building and the neighborhood. Never describe or imply a target resident, their age, family status, occupation, national origin, or any protected class. Keep your normal review process. The file reduces bad drafts, it does not replace the person who approves copy.
Where to put it so you stop pasting it
If you're on ChatGPT, build a Project per portfolio or per lease-up and paste the file into the project instructions. Every chat inside inherits it. Claude Projects works the same way, with a knowledge base you can load actual PDFs into: your brand guide, your last three market surveys, your comp set. Gemini uses Gems. Copilot, if that's what IT approved, will read a pinned file in the same SharePoint folder.
Then break it into a master file plus per-property addenda. Master holds brand, banned words, fair housing rule, channel stack and reporting norms. Each property gets a short addendum with its own mix, pricing, concession and demand drivers. Working a lease-up? Master plus that one addendum. Nothing else.
What it feels like on Tuesday
You need eight renewal emails for a property with a 4.2% increase, four different resident tenures, and a competitor two blocks away offering one month free. Without a context file, that's a twenty-minute briefing before you get a first draft worth reading. With one, it's a sentence, and the draft comes back knowing the increase, the unit mix, the comp, and that your brand doesn't say "we value you as a resident."
Same for review responses, ILS spend memos, floorplan descriptions, tour follow-up sequences, the Monday occupancy narrative.
Update it monthly. Pricing and concessions move, and a stale concession in your context file is worse than no context file, because now the model is confidently wrong and you're the one who taught it. Put a "last updated" date at the top. Twenty minutes on the first of the month.
One hour today. Then you never explain your portfolio again.