How to hand AI visibility to your team: six jobs, in order.

The hard part of AI visibility is almost never deciding what to do. The playbook already sets the order: read the board for your market, scan your own property, then work outward from what you control. The hard part is the next morning, when a general manager who now knows exactly what is wrong has to turn it into something a person other than themselves can pick up.

That is where this work usually stalls, and it stalls for a boring reason. Every job on the list sounds like somebody else's job. "Structured data" sounds like the developer. "Listing copy" sounds like whoever has the Booking.com password, and nobody is quite sure who that is any more. So the list sits in a tab, and the answer a traveler gets when they ask ChatGPT where to stay in your city — a ranked list of about nine hotels, across the 18 US markets on our AI Hotel Ranking — stays exactly as it was.

So here is the same work, cut into six jobs you can hand out. Each one names who it belongs to, the exact ask to send them, and the single condition that decides whether it is finished. None of them requires you to understand how a language model works. All of them require you to be specific about what you want back.

The rule that makes this safe to delegate

Before any of the six, there is one rule, and it is the only part of this article you genuinely have to enforce. Whoever does the work — an agency, your web developer, your marketing coordinator, or an AI assistant — sends three things back. Not one of them is optional, and the first one happens before anything is edited.

Paste this at the bottom of every request below. It is the same three lines every time, which is the point: your team learns one standard, not six.

Before you change anything, send me:

1. THE PLAN — which pages you would touch, and what you would
   write. One message, then stop. Nothing changes until I say go.

2. THE GAPS — every fact you could not verify on our own website,
   written as [VERIFY: what is needed]. Never guess a number.

3. THE REPORT — afterwards: what changed and where, what you
   could not do, and what you checked it with.

Each line exists because of a specific way this goes wrong. Without the plan, you find out what was changed by looking at your live site. Without the gaps, a missing ballroom capacity quietly becomes a plausible ballroom capacity, and your front desk is the one that has to explain it at check-in. Without the report, you have no idea which half of the job is still open. These are three of our seven working rules — the page is public, needs no account, and is written to be read by an assistant as easily as by a person.

The six jobs

They are ordered by how much of each one you control, not by how badly it is broken. That is deliberate: the first three you can finish this month without asking anyone's permission, and the last one you cannot finish at all — you can only put it in motion. If a scan flags something urgent further down the list, do that first; otherwise work top to bottom.

1. Your own website

What it is. The pages a traveler reads — and the only pages an assistant reads with your name attached. AI answers from text it can fetch, which means a fact that lives only in a PDF brochure, inside an image, or behind a click does not exist. If your site never states the pet policy, no assistant can tell a guest what it is, and one of them will guess.

Hand it to. Whoever edits the website — in-house marketing, your web agency, or the one person who knows the CMS password.

The ask. "Add a short section to [page] that answers this question in its first sentence: '[the exact question a guest asks]'. Use only facts already published on our own site, and mark anything you cannot verify. Match the heading levels, the markup and the tone of the sections already on that page — it should look like it was always there."

Done when. The section is live, a stranger could not pick out which part is new, and every fact in it traces back to something you published. Full detail for whoever does the work: the website content playbook.

2. Site schema

What it is. The machine-readable version of facts your page already states in prose — room types, amenities, rates, policies, location. It is the one place where an assistant does not have to interpret you, so it is the cheapest accuracy you can buy.

Hand it to. Whoever can edit the site's code. You do not need to understand any of it. You need to ask for one specific thing and refuse the alternative.

The ask. "Extend the schema.org markup already on [page] with [our amenities / our room types / our meeting space]. Do not add a second Hotel block — extend the one that is there. Every value has to come from what the page already says. Run the result through a schema validator and send me exactly what it returned."

Done when. One block, not two, and you are holding the validator output. The "do not add a second" clause is the whole job: two descriptions of the same hotel force every reader, human or machine, to decide which is true, and they may not pick yours. Detail: the structured data playbook.

3. Maps and local

What it is. Your Google Business Profile and the local listings around it — hours, address, category, amenities, photos. Usually years out of date, usually contradicting the website in two or three places, and usually fixable in an afternoon.

Hand it to. Whoever manages the Google Business Profile. If nobody does, that is the finding.

The ask. "Go through our Google Business Profile field by field against the website. Send me a list of every field where the two disagree, with the value you propose for each. Change nothing until I have seen the list."

Done when. Nothing in the profile contradicts the site. Worth knowing before you delegate this to software: no AI assistant can edit a Google Business Profile. What it can do is write the exact value for each field so that updating it becomes a paste.

4. OTA listings

What it is. Booking.com, Expedia, Hotels.com, Tripadvisor. Assistants read these as third-party descriptions of you, which is precisely why the facts in them have to match your own site to the letter.

Hand it to. Whoever has the extranet logins. This is the job where the honest limit matters most: no assistant can log into an Expedia or Booking.com extranet, and one that claims it can is about to invent the result. What it will do is write the exact text for every field, so the update becomes a paste rather than a writing job.

The ask. "For [platform], rewrite [the description / the room descriptions / the amenity fields]. Lead each field with what a guest is trying to find out. Replace every generic phrase with a specific — 'a short walk from the convention centre' is unusable, 'a nine-minute walk from the convention centre's south entrance' is not. Keep every fact identical to our own website. Give me each field separately, with a character count."

Done when. The facts match the website exactly, field by field. A contradiction between the two costs more than a gap in either — it makes both unreliable. Detail: the OTA listing playbook.

5. Guest reviews

What it is. What your guests say about you and what you say back. Review text is among the most-quoted material about any hotel, and a management reply is the only part of it you write.

Hand it to. Guest relations, or whoever already answers reviews.

The ask. "Draft replies to the last [n] reviews that mention [the theme the scan flagged]. Our voice, specific, and no promise we have not checked is true today. I will post them."

Done when. The replies are posted, and any complaint that showed up more than twice has been routed to the person who can actually fix the thing being complained about. An assistant cannot post to a review platform; everything here is a draft that you send.

6. Editorial and PR

What it is. Guides, "best of" lists, local press. The most quoted of the six, the slowest, and the only one you cannot change directly — which is exactly why it is last and not first.

Hand it to. Marketing or PR.

The ask. "Draft a pitch for [the angle] and a shortlist of ten outlets or writers who actually cover it. I will send them."

Done when. The pitches are sent. That is the whole measure, and it is deliberately not "we got the placement" — a placement is not something you can schedule. You can change your own page this afternoon; you cannot change a press mention at all.

If you only do three of these

Do the first three. They are the ones where the work lands entirely inside your own control, they are the cheapest, and they are the source material every other channel ends up quoting. Jobs four through six are worth doing; none of them can be finished by you alone, which is why they wait.

If the team you are delegating to is an assistant

Most hotels do not have six people for six jobs. The good news is that the writing half of all six is work an AI assistant is genuinely good at, and you already have one — ChatGPT, Claude, Gemini, whichever is open.

Doing it by hand takes three ingredients: the finding (which questions you are missing, on which engines), the page or listing it concerns, and a link to sigtrip.com/playbooks/ so the assistant can fetch our working rules itself. Then the three-line standard above, verbatim. Ask for the plan before it writes anything.

If you are on a paid plan, the dashboard assembles that for you: AI Visibility groups each finding by where the work lands — the same six channels, in the same closest-to-your-control order — and every action carries a Fix with AI agent button. Each finding there is specific to your property — it comes out of your own scan results, and it already accounts for what you have marked done or dismissed before, so what you are handing over is never a generic checklist. What lands on your clipboard is that finding, the concrete steps, the rules, and the link to the matching playbook, in one block you paste into whichever assistant you use. It never publishes anything, and it is not meant to: review is the part that stays yours.

What comes next

Two things, in this order. First, re-measure — and give it time. Ask an AI the same question twice and you can get different hotels back, so reading your numbers the morning after an edit means reading noise. Give it a few scans and expect movement in position rather than a guarantee of one. Ticking the job off is not the verdict; the next measurement is.

Second, and only once the visibility work is holding: the booking path. Being recommended and then handed to an OTA is half the job done for you and all of it for someone else. AI Direct Bookings closes that — the assistant that recommends you can also book you, and the reservation lands in your own system. After the list, not instead of it.

You do not need six people, and you rarely need all six jobs. What you need is to know which of them your own results point at, and why — that judgement is the one part no assistant can make for you. Once you have it, an AI agent can carry the writing half of every job on the list, and your own part shrinks to three lines: show me the plan, mark what you could not verify, tell me what you changed.