Three pains, nine AI solutions:
Prompt Kit

1.a Rewrite for the short list

BASIC · Free tier · about an hour · no code · Pain 1 · Feeds 1.b, 1.c and 2.b

Also fits: any listing on any channel. The point is one set of checkable facts published identically everywhere, because the AI tools that write the traveler's short list reward specific, consistent facts over "family friendly".

Tool: Claude, ChatGPT or Gemini, free tier, one chat. No files needed.

Step 1: adjectives out, facts in

Here is our current listing:
[PASTE CURRENT LISTING, or use this fictional one: "Unforgettable scenery, friendly crew, perfect for families. Alaska's best half-day cruise!"]

Approved facts:
[PASTE APPROVED FACT SHEET, or use the fictional teaching facts]

Delete every adjective or claim a competitor could also make. Replace each with a fact a traveler could check: capacity, duration, minimum age, what is included and what is not, what happens in rain, cabin heating, restrooms, walking distance from the cruise ship dock, cancellation terms. Use only the approved facts. Where a fact is not supplied, write [FILL] rather than guessing. Output a numbered fact list first, then one paragraph under 120 words that uses only those facts. Mark DRAFT - PENDING OWNER REVIEW.

Step 2: the same facts on every channel

Using only the numbered fact list I approved above, write three versions: (a) ten bullets for Google Business Profile, (b) 150 words for TripAdvisor, (c) three lines for a Viator description. Do not add, drop or change a fact between versions. After the three versions, list any fact that appears in one version and not the others, and any [FILL] still open.

Then: fill the blanks from the owner's records, publish all three the same day, and fix the website to match. Mismatches between channels are what review 6 in the teaching facts is about.

Measure: rerun the live-audit prompt from the session in three AI tools and score the Appear and Described checks (Number 1, AI visibility score). Expect no change for weeks; the tools recrawl on their own schedule.

1.b Your own curated ten

INTERMEDIATE · Paid tier, about $20 a month · a week of evenings · Pain 1 · Needs 1.a from the business and two partners · Feeds 1.c and 3.b

Also fits: a flightseeing operator plus a kennel and a riverboat; a lodge plus its three most-referred activities. In the survey, 63% preferred a platform showing ten curated options over 500 with filters.

Tool: a Claude Project, a ChatGPT Project or Custom GPT, or a Gemini Gem. Upload the fact sheets from 1.a for the business and two partners, plus the last 50 reviews. The paid tier is for file limits and reliability, not for a different model.

Attached are approved fact sheets for our business and two partners, plus our last 50 reviews, de-identified.
[ATTACH FILES, or paste the fictional teaching facts]

Traveler types we actually see:
[LIST YOURS, or use: family with children under 10; serious photographer; cruise ship day visitor with 5 hours ashore; couple on a 10-day road trip]

For each traveler type, choose up to three trips from the attached inventory only, and give a two-sentence reason that cites a specific fact from a sheet (age minimum, duration, distance, what is included). If fewer than three fit, say so and do not pad. Then draft a three-question picker (who is traveling, how much time, what matters most) that routes to the right list. Then draft the page copy under 300 words with the reasons visible. Do not invent prices, availability, reviews, awards or partner facts. List questions for missing facts. Mark DRAFT - PENDING OWNER REVIEW.

Then: paste into the website builder (Squarespace, Wix or similar) as a "Which trip fits you?" page. Keep the "why it made the cut" line visible under each pick. Each partner approves its own lines before publishing.

Measure: saves, inquiries and bookings that start on that page; direct share of bookings (Number 2). The page is also structured content the AI tools can quote, which is the bridge to 1.c.

1.c AI answer monitoring and structured data

ADVANCED · Developer or vendor · 2 to 4 weeks · model API fees · Pain 1 · Automates the live audit · Needs 1.a published

Also fits: ATIA or a regional group running it once for every member. Monitoring cost scales with the number of prompts, not the number of businesses, because one answer is parsed for every watched name. That makes the association-scale version cost about the same as the single-operator version.

Tool: Claude Code, OpenAI Codex or Gemini CLI writes the script from the brief below in roughly 1 to 3 days (80% interval: 1 to 5 days). The script calls the three model APIs with live web access: the Claude web search tool, the OpenAI Responses API web_search tool, and Gemini grounding with Google Search. API cost at 50 prompts, three models, weekly: roughly $10 to $100 a month (80% interval; check current tool pricing). Commercial AI-visibility services exist; compare them on prompt count, model coverage and whether the raw answers are kept.

Build brief: the weekly monitor

Build a weekly AI answer monitor.

Inputs: prompts.csv with 20 to 50 traveler questions, for example "best half-day glacier cruise from Seward for a family with an 8-year-old" and "what to do in Seward on a rainy day with 5 hours ashore"; names.csv with the business names to watch (ours, and every member if run for an association).

Each week, send every prompt to three models with live web access: Claude through the Anthropic API web search tool, OpenAI through the Responses API web_search tool, and Gemini through the Gemini API with Google Search grounding. For each answer record: date, model, prompt, whether each watched name appears, its position in the answer, the URLs cited, and the full answer text. Append to a Google Sheet or a CSV. Produce a weekly diff: names that gained or lost mentions, and sources that appeared for the first time.

Keep API keys in environment variables. Add a dry-run mode that uses two prompts. After the first full run, report the actual monthly API cost from token and tool-call counts; do not estimate from assumptions. Store no traveler data; prompts are generic. Write a README a non-developer can follow to add a prompt or a name.

Build brief: structured data for the website

Generate JSON-LD for our website using schema.org types: TouristAttraction or LodgingBusiness as applicable for the business, TouristTrip for each product, Offer with the price and availability pointers I supply, and amenityFeature for the checkable facts in the approved fact sheet.
[PASTE APPROVED FACT SHEET AND PAGE LIST]

Use only the approved facts. Output one script block per page and say where to paste it. Flag any fact in the markup that is not on the visible page, because the two must match. I will check every block in Google's Rich Results Test before publishing.

Caution: one week's answers are noise. Report trends over a quarter. Do not infer ranking or booking impact from one answer.

Measure: share of AI answers naming the business, by prompt and model, quarterly (Number 1).

2.a The honest picture, from your reviews

BASIC · Free tier · about 2 hours · a person approves every reply · Pain 2 · Feeds 2.b and 3.a

Also fits: a B&B, a lodge, a charter. Accor's A/B tests found that genuine reviews, including bad ones, built more confidence than curated brand copy. Expectations set before booking also lower in-window cancellations (Number 4).

Tool: Claude, ChatGPT or Gemini, free tier. Copy reviews from Google, TripAdvisor or the booking platform; remove guest names.

Step 1: sort the reviews and draft the section

Here are our last 50 reviews, de-identified:
[PASTE REVIEWS, or use the six fictional reviews in the teaching facts]

Sort them into four groups with counts: (1) what guests loved, (2) what guests wished they had known before booking, (3) complaints we could fix, (4) complaints inherent to the experience (weather, sea conditions, walking distance). Quote the review phrase that supports each item. Then, from groups 2 and 4 only, draft a "What to expect, honestly" section under 150 words in plain language: no marketing adjectives, no promises, no reassurance that is not a fact. Mark DRAFT - PENDING OWNER REVIEW.

Step 2: reply to every review

Draft replies to these unanswered reviews:
[PASTE UP TO 5 REVIEWS, or use fictional reviews 2, 3, 4 and 6]

Under 60 words each. Name the specific issue the guest raised. Do not use a template apology, do not offer refunds, credits or free trips, and do not dispute the guest's experience. State that a fix has been made only if it appears in this list of confirmed fixes: [LIST CONFIRMED FIXES, or "none"]. Mark each DRAFT - PENDING OWNER REVIEW.

Then: publish the section on the website and on every booking page that allows it; a person edits and posts the replies. Keep the section current: refresh it when the review mix changes.

Measure: review reply rate; inquiry-to-booking rate; in-window cancellations once a season of data exists (Number 4).

2.b A grounded answer assistant

INTERMEDIATE · Paid tier · 1 day to set up · monthly error check · Pain 2 · Needs 1.a and 2.a · Feeds 2.c and 3.b

Also fits: any business whose inbox repeats the same 20 questions. Staff use it first; nothing guest-facing until a month of graded answers says it is safe. This is the approach a large hotel group has described publicly: answer only from verified first-party facts, and say "ask us" otherwise.

Tool: NotebookLM (free, answers only from uploaded sources and cites them) for the staff-only version; or a Claude Project, a Custom GPT or a Gemini Gem with the instruction below. Files to load: fact sheet, cancellation policy, packing list, the 2.a section, the reviews, the FAQ. A guest-facing widget comes later through a chat vendor, with the same instruction.

Instruction for the assistant

You answer questions for staff of [BUSINESS NAME, or Bay Light Day Cruises]. Use only the attached files. For every answer, quote the sentence you used and name the file it came from. If the answer is not in the files, reply exactly: "Not in our facts. Ask the office." Never state a price, date, availability, weather, access or policy that is not in the files. Never offer a refund, credit, change or exception. If a question involves safety, medical, legal or accessibility matters, reply "Ask the office" and stop. Mark every reply DRAFT - PENDING STAFF REVIEW.

Monthly error check

Here are 20 real guest questions from last month, de-identified:
[PASTE QUESTIONS, or use: Is lunch included? Can my 3-year-old come? How far is the harbor from the cruise ship? What if it rains? Do you offer a group discount? Can we bring our dog?]

Answer each one following your instructions. I will grade every answer as correct, needed correction, or correctly declined, against the source files.

Then: run it staff-only for 30 days. Record the correction rate. Add missing facts to the files rather than widening the instruction. Only after the correction rate is near zero for a month consider a guest-facing widget, and keep a person on every refund, change or complaint.

Measure: reply time; share of answers needing correction; repeat questions per booking (Number 3).

2.c Post-booking retention engine

ADVANCED · Developer plus booking-system API · 4 to 8 weeks · Pain 2 · Needs 2.a and 2.b, and a booking platform with an API

Also fits: a multi-day tour company or a 20-room lodge with 1,500 or more bookings a season on FareHarbor, Peek, Checkfront, Cloudbeds or similar. What the survey says would most raise confidence: the ability to review and change before confirming (52%), a track record (30%), and a human who can step in (20%). The offer is confidence first, discount last.

Tool: a developer with Claude Code, Codex or Gemini CLI, the booking platform's API, and a model API (Claude, OpenAI or Gemini) for drafting only. Exposure scoring is a readable rule set, not a model.

Design and build a post-booking retention pilot on [BOOKING PLATFORM].

Nightly: pull open bookings through the platform API (booking id, lead time, rate type, channel, party size, add-ons, price-alert flag if available). No guest names, emails or payment data leave the platform.

Score exposure with a transparent rule set I can read and change, for example: +2 lead time over 60 days, +2 flexible rate, +1 OTA channel, +1 no add-ons. For the top 10 percent by score, call a model API to draft one of three approved messages and nothing else: (A) a confidence note stating what to expect and the weather plan B, (B) one relevant add-on from the approved list, (C) a lock-in offer to convert to non-refundable on terms I supply. The model must use only the approved facts file.

Write drafts to an approval queue (a Google Sheet or email drafts). Nothing sends without a person. Randomly assign 20 percent of eligible bookings to a holdout that receives nothing, and store the assignment. Report monthly: in-window changes and cancellations, treatment versus holdout, as counts and rates, with flexible and non-refundable bookings shown separately. Do not match competitor prices and do not chase every cancellable booking.

Sample size, stated plainly: detecting a drop from 15% to 12% in-window cancellations at conventional significance needs roughly 2,000 bookings per arm (two-proportion test, alpha 0.05, power 0.8). Most members will get a directional read, not proof, unless data is pooled through ATIA or the test runs across seasons. A directional read is still worth having; it is not evidence to scale on.

Measure: in-window changes and cancellations versus holdout; revenue kept (Number 4).

3.a The AI trip kit

BASIC · Free tier · 1 hour to build, then minutes per booking · Pain 3 · Needs 1.a and a partner list · Feeds 3.b and 3.c

Also fits: a cabin near Denali, a lodge on the Kenai, any operator whose guests have open days. Alaska trips are multi-business by default, so this pain is probably larger here than the national 39% suggests, though nobody has measured it.

Tool: Claude, ChatGPT or Gemini, free tier. Save the prompt and the facts in a reusable place (a Project, a Gem or a text file). Per booking, paste dates, party and what was booked only; strip names, emails and payment details, and check the tool's data-use setting.

Act as our pre-arrival drafting assistant.

Approved facts, partner list and drive times:
[PASTE, or use the fictional teaching facts]
Month and typical conditions:
[PASTE, or use the fictional July note]
Booking, de-identified:
[DATES, PARTY SIZE AND AGES, WHAT WAS BOOKED, or use: July 14 to 16, two adults and two children aged 6 and 9, the 9:00 cruise on July 15]

Draft a one-page trip kit under 400 words: day by day with our booking anchored; for each open half-day, up to two partner options with drive time from our meeting point and a one-line reason that cites a fact (age minimum, duration, runs in rain); what to wear this month; a weather plan B for every outdoor item; three things first-time visitors get wrong here (daylight hours, fuel stops, phone coverage). Do not promise partner availability, prices or weather. Put every missing fact in a list of questions for the owner. Mark DRAFT - PENDING OWNER REVIEW.

Then: a person checks it against the partner list, exports to PDF and sends it with the confirmation. The kit is the first draft of the trip record in 3.c.

Measure: partner referrals and add-ons per booking; repeat questions per booking (Number 3).

3.b A partner circle with shared facts

INTERMEDIATE · Paid AI tier plus a no-code tool · about a week · partner agreement · Pain 3 · Needs 3.a and 1.a from every partner · Feeds 3.c

Also fits: three to five operators who complete each other's trips: a lodge, a charter, a kayak outfitter and a restaurant. The referral codes are the measurement, so agree on them first.

Tool: one shared fact sheet in a Google Doc; Zapier or Make for the automation (both offer built-in steps for Claude, OpenAI and Gemini); the booking platform as the trigger. Running cost: automation plan roughly $20 to $30 a month plus cents per kit.

Shared fact sheet template (one block per partner)

PARTNER: [name]
Offer and duration: [what, how long]
Departures and season: [times, dates]
Minimum age and group limits: [ages, capacity]
Runs in rain / cancels when: [conditions]
Drive time from each partner's meeting point: [minutes]
What is included / not included: [list]
Cancellation terms: [exact wording]
Referral code and how it is honored: [code, what the guest gets, how it is counted]
Fact owner and last checked date: [person, date]
Not supplied (ask the owner): [list]

Automation: the AI step

Trigger: new booking. Delay: until 30 days before arrival. AI step: the prompt below, with fields filled by the automation. Approval: the operator. Send: email to the guest with the codes.

You draft the trip kit for a partner circle of [N] businesses.
Shared fact sheet: {{shared_fact_sheet}}
Month note: {{month_note}}
Booking fields: arrival {{arrival_date}}, departure {{departure_date}}, party {{party_size}}, ages {{ages}}, booked {{product_booked}}.

Draft a one-page kit under 400 words: day by day with the booked product anchored; for each open half-day, up to two partner options with drive time and a one-line reason citing a fact from the sheet; what to wear this month; a weather plan B for every outdoor item. Include each partner's referral code exactly as written in the sheet. Do not promise availability, prices or weather. Do not include any fact that is not in the sheet. Output plain text. Mark DRAFT - PENDING OWNER REVIEW.

Then: the partner agreement covers who approves kits, who owns each fact, how codes are reconciled monthly, and what booking data reaches the AI step (dates, party and product only). Test with five bookings before switching on auto-send, if auto-send is ever switched on.

Measure: cross-referrals by code; attach rate per booking.

3.c An agent-ready trip record

ADVANCED · Platform API plus a developer, or ATIA-wide · months · Pain 3 · Needs 3.a and 3.b · The strategic rung

Also fits: for a six-person operator this rung is one question to the booking platform, not a build. For a larger operator or for ATIA it is a build, and a board-level question: the checkout layer AI agents will use is being built now, and whether a supplier is reachable through it is being decided now, possibly by default. The reversible hold is the whole point: an agent without holds will not place a high-consideration Alaska trip.

Tool: the booking platform's API; an MCP server (the open protocol Claude, ChatGPT and Gemini agents use to call external tools); Claude Code, Codex or Gemini CLI to scaffold it. The months are in platform integration, hold-expiry rules and policy, not in the code.

The question for the booking platform

Does your platform expose (1) live availability through an API, (2) time-limited, reversible holds that expire automatically, and (3) confirmation and cancellation through the same API? Are you participating in any agent checkout protocol, such as Google's Universal Commerce Protocol? If yes, what do I have to switch on, and what does it cost? If no, is it on your roadmap, and when? Please answer in writing so I can share it with our association.

Build brief, for a larger operator or ATIA

Design a minimum viable trip record and an MCP server over the [BOOKING PLATFORM] API.

Trip record schema (JSON): party (count and age bands, no names), dates, budget band, preferences, shortlisted options, rejected options with the reason, open questions. No names, contact details or payment data in the record.

MCP tools: check_availability(product, date, party); place_hold(product, date, party) returning a hold id that expires in 24 hours; confirm_hold(hold id); release_hold(hold id). Every hold must expire server-side even if the caller disappears. Log every tool call. Rate-limit holds per caller so an agent cannot create phantom inventory. Produce a test harness that runs a scripted agent through search, hold, review, change and confirm. Document what a Claude, ChatGPT or Gemini agent needs to connect, and what a person at the business sees when a hold is placed.

Policy before code: who may place a hold, how long it lasts, what happens to inventory while it is held, and how a person can step in. The survey's confidence drivers (review and change before confirming, a track record, a human available) are the design spec.

Measure: bookings through AI channels; hold-to-book rate.