A partner circle with shared facts

How it works

1. Booking inA row goes into your queue sheet: dates, party, product, and send_date = arrival minus 30 days.
2. Every morningThe automation picks rows that are queued and due: send_date today or earlier.
3. AI draftsFrom the shared fact sheet and the booking fields. Nothing else reaches the AI step.
4. Gmail draftThe kit is saved as a draft. Nothing is sent.
5. A person approvesOne person checks the draft, sends it, and records it in the sheet.
6. Monthly countPartners count the codes and agree the numbers.

Your week

1. Day 1: agree the circle with your partners (Step 1). Pick the businesses, the codes and who approves kits. About 2 hours together.
2. Day 2: fill the shared fact sheet (Step 2). Each partner fills or approves its own block. About 30 minutes per partner.
3. Day 3: set up the queue sheet and paste in the five fictional test bookings (Step 3). About 1 hour.
4. Day 4: dry-run the AI step by hand and check the draft (Step 4), then build the automation in Zapier or Make (Step 5). 3 to 5 hours.
5. Day 5: run the five fictional bookings end to end, drafts only (Step 6). Fix what you find, then connect real bookings.
6. Every month: reconcile the codes (Step 7), note the attach rate, and re-check any block older than 60 days.

Times are estimates for a small operator who already has a 1.a fact sheet and has used 3.a by hand.

Step 1. Agree the circle

Agree the rules before you build anything. The referral codes are the measurement, so agree them first. The agreement template is a starting checklist, not legal advice. Each line says what kind of obligation it is.

• Recommendation Three to five businesses that complete each other's trips: a lodge, a cruise, a kayak outfitter, a restaurant. Not two businesses selling the same hour to the same guest.
• Policy choice Approvals. One named person at the business that sends the kit approves each kit before it goes out. Each partner approves the wording of its own block before it is used.
• Policy choice Fact owners. Each block has one named owner who re-checks it at least every 60 days and before the season opens, and tells the circle about a change within two business days.
• Contractual Referral codes. One unique code for each pair of businesses (the one that sends the guest, the one that honors the code), what the guest gets with it, and any referral fee, written into your signed partner agreement. Check each booking platform's terms on promo codes.
• Policy choice Monthly reconciliation. Each business sends its code counts by the 5th of the month, and the circle settles differences within a week (Step 7).
• Policy choice Data. The AI step receives dates, party size and ages, and the product booked. Nothing else: no names, emails, phone numbers, payment details or medical notes.
• Contractual Tools. Check the terms and data processing agreements of Zapier or Make, your AI provider, Google Workspace and your booking platform before you connect them.
• Legal mandate Guide licensing. A partner that guides big game hunting needs its license under AS 08.54; a sport fishing guide and business register with ADF&G each year. That license stays with that partner. It is not delegable to the circle, the coordinator or an AI tool.

Download the agreement checklist (.md)

A .md file is plain text. Open it in Notepad, TextEdit, Word or Google Docs, fill the brackets, and have every partner review it. Ask your attorney before you sign anything that involves money.

Step 2. Build the shared fact sheet

One block per business, including yours. Each partner fills its own block, or approves what you typed. The sheet goes into the AI step for every booking, so use published, approved facts only.

Business facts only. No guest names, emails, phone numbers or booking details in this sheet. Everything stays in this browser.

A block is flagged when a required field is empty or it was last checked more than 60 days ago. Empty fields become [FILL] in the AI step, so the AI asks instead of guessing. Import reads shared-facts-template.csv, a CSV saved here, or the partner list saved from 3.a.

The fictional example has July only. For any other month the AI step writes [FILL: month note] instead of describing weather.

Where it lives: make a tab called facts in your queue spreadsheet, with the headers id, shared_fact_sheet, month_notes in row 1. In row 2, type current in A2, paste the fact sheet text into B2 and the month notes into C2. The automation reads row 2 every morning, so an update reaches the next kit. Send the Markdown version to partners when they review their blocks.

Step 3. Set up the booking queue

The queue is a sheet with one row per booking, in Google Sheets or Excel online. The automation never holds a booking inside itself; it reads the sheet every morning. Start from kit-queue-template.csv: it has the columns and five fictional test bookings.

The queue holds dates, ages and products only. Keep guest names, emails and phone numbers in your booking system, not in this sheet.

The five fictional test bookings

When would the daily run draft each kit? The rule is "queued, and send_date today or earlier", so a late booking is picked up the morning after it arrives.

Swipe the table sideways to see every column.

Try the rule on a date

• Changed dates: update arrival and send_date in the row. If a kit was already drafted, delete the draft and set status back to queued.
• Canceled: set status to canceled, and delete any draft. The daily run skips the row.
• A missed morning: nothing is lost. The next run picks up every row that is still queued and due.
• Booked less than 3 days before arrival: Policy choice write to the guest yourself. Delete the automated draft if one appears, and set the row to sent with a note.

Step 4. The AI step

This is the Prompt Kit 3.b prompt with six additions: partners are introduced as partners, age and time clashes are left out, safety calls stay with each business, missing facts become a question list, internal fields (fact owners, check dates, code counting) stay out of the guest's kit, and the draft is marked at both ends.

Map each placeholder

In Zapier or Make, delete each {{placeholder}} and insert the matching field with the field picker. Typing the braces does nothing.

Never map a guest name, email, phone, booking notes or payment field into the AI step. The booking_ref stays out too.

Dry run: fill it by hand with a fictional booking

Before you build anything, run the prompt once by hand. Pick a test booking, copy the filled prompt, paste it into Claude, ChatGPT or Gemini, then paste the answer into the check below.

Fictional or de-identified only: dates, party size, ages and product. No guest names, emails, phone numbers, card numbers, IDs or medical details.

Open: Claude ChatGPT Gemini (internet required; check each tool's data-use setting)

Check the draft

Paste the AI's answer. The page checks it in this browser for the DRAFT line, length, referral codes, numbers that are not in your facts, promise words, age clashes, guest details and open [FILL] items. It checks against the booking above.

Drafts from fictional bookings only. If a draft has a real guest's details, remove them before pasting.

Step 5. Build the automation in Zapier or Make

Two automations: one adds each new booking to the queue, one runs every morning and drafts the kits that are due. Both products offer built-in AI steps for Claude, OpenAI and Gemini. Pick the one you already pay for.

Step names change often. The names below are generic; the grey line under each step says roughly what to look for. Check current names in each product's help (Zapier help, Make help). No ready-made template files are shipped, because they are tied to each account's connections.

Running cost is an estimate: roughly $20 to $30 a month for an automation plan that allows multi-step automations and a daily schedule, plus cents per kit for the AI step. Each kit uses a few tasks (Zapier) or operations (Make). Check current plans before you choose.

Step 6. Test with five fictional bookings, then approve every kit

Run the whole chain on the five fictional rows before any real booking touches it.

1. Copy the five TEST rows from kit-queue-template.csv into your queue tab. Change send_date on TEST-001 to TEST-004 to today's date. Leave TEST-005 as canceled.
2. Put the fact sheet text and month notes in the facts tab (the fictional ones are fine for the test).
3. Run the daily automation once by hand. Zapier: test it step by step, or turn it on and run it. Make: Run once.
4. Expect four Gmail drafts and nothing in Sent. TEST-005 is untouched. Four rows now say drafted, with a date and a draft id.
5. Run it again. Expect no new drafts: those rows are no longer queued.
6. Open each draft. Paste it into Check the draft and go through the approval checklist below. Expect TEST-001 to leave out the kayak (minimum age 12), and the June and August kits to ask for a month note.
7. Delete the four test drafts and the TEST rows. Only then point the booking trigger at real bookings. Keep drafts only.

The approval checklist: one person, every kit

KIT APPROVAL CHECK: one person, every kit
[ ] Booked product, date and time match the booking system.
[ ] Every partner option fits the party's ages (minimum age in the fact sheet).
[ ] No partner option overlaps the booking once drive time is added.
[ ] Every fact matches the shared fact sheet. Nothing added, nothing rounded.
[ ] Referral codes are exactly as in the sheet.
[ ] If you earn referral fees from these partners, the kit says so.
[ ] No promises about availability, prices, weather or wildlife.
[ ] Safety wording (weather, water, glacier, wildlife) is each business's own wording.
[ ] Every [FILL] is answered or its line deleted. QUESTIONS FOR THE OWNER is answered, then removed.
[ ] The guest's address goes in the To field, from the booking system. No other guest details in the body.
[ ] Both DRAFT - PENDING OWNER REVIEW lines deleted, last.
Then send it. In the queue sheet: your initials in approved_by, the date in sent_date, status sent.

Policy choice One named person approves each kit, and the automation only ever creates drafts. If you ever consider automatic sending, decide it with your partners after a full season of approved drafts, and write it into the agreement.

Step 7. Reconcile the referral codes each month

By the 5th of each month, each business sends the codes it counted last month and how many. Add one count per business below. The page lines them up by code, shows what does not match, and catches codes that are not in the shared fact sheet, which are usually typos.

Codes and numbers only. No guest names, booking numbers, emails or phone numbers.

Swipe the table sideways to see every column.

Settle each mismatch by comparing booking dates for that code, then write the agreed number and who agreed it in the saved CSV. The format matches referral-reconciliation.csv.

Downloads

• shared-facts-template.csv The fact-sheet fields plus fact_owner and last_checked, with the three fictional blocks. Replace the rows, then import it in Step 2.
• kit-queue-template.csv The queue columns and five fictional test bookings (Step 3 and Step 6).
• referral-reconciliation.csv The monthly reconciliation format, with a fictional July.
• partner-agreement-template.md Approvals, fact owners, monthly code reconciliation, what reaches the AI step. A starting checklist, not legal advice.
• zapier-recipe.md Both Zapier automations, step by step, with the test plan.
• make-recipe.md Both Make scenarios, step by step, with the test plan.
• Your fact sheet, prompt and reconciliation Use the Save buttons above. Files are made in your browser.