Attribution is the word of the year in hotel marketing.

Everyone is asking about it when it comes to AI Search. Very few people have defined what it should actually look like for this new way of discovering and choosing hotels.

Attribution in traditional SEO is largely settled. Attribution in local search is largely settled. Attribution in AI Search is genuinely unproven, and that is a problem, because AI Search is where a growing share of your guests are now forming their shortlist.

This article is the method we use to solve it. Not a summary of it. The actual recipe, step by step, so you can run it on your own property.

Fair warning before you start. It is not clever. It is not a tool you can buy. It is a lot of unglamorous work, and that is precisely why almost nobody has done it.

You cannot attribute anything without the data. Not some of the data.

All of the data. The more you can get your hands on, the better. Every dataset you skip is a hole that somebody at the ownership meeting will drive an objection straight through.

There are seven steps to gathering it. Do them in this order, because the order is what turns a pile of numbers into an argument. Context and market first, then results, then behaviour, then revenue.

Step One: Context and the Measurement Period

Before a single number, settle two things.

Context. What was running before, and what did AI Search do to it? There are three answers, and they are not equally strong.

A standing start. Nothing was running, and AI Search was the first thing switched on. This is the cleanest test there is, because there is no other programme in the window to argue about. It is also the rarest, and it carries its own weakness: with no prior activity there is often no meaningful comparison period, and a neglected property will lift somewhat simply because someone finally paid attention to it. If this is your situation, say so, and be careful not to claim more than the absence of alternatives gives you.

A replacement. Budget moved out of one activity and into AI Search. Less clean than a standing start, because the old programme leaves foundations behind that decay slowly rather than stopping on the day you switch it off. But it is the most commercially honest test, because AI Search has to do the job the old programme was doing, and if it fails you will see it in the same numbers that used to justify the old spend. A replacement also produces something no other context gives you: a declared cost. You gave something up. Publish what it was.

An addition. AI Search went on top of activity that was already running. Hardest to isolate, and the most common by a distance. It is not the dead end people assume it is. AI Search has its own metrics and its own timeline: mentions, share of voice, citations, prompt level positions, sentiment, sessions arriving from AI platforms and any AI assistant channel. These provide a distinct AI visibility layer that should be measured separately, even where the underlying activity overlaps with SEO. Track them month on month and year on year, against the months before AI Search existed on the property.

What you cannot do in an addition is claim every dollar of revenue growth for AI Search alone. What you can do is show exactly what AI Search did, when it did it, and what followed. Those are two different claims, and the discipline is in not presenting the second as the first.

Measurement period. When did it start? When did it end? Twelve months, three months, ninety days? And what are you comparing it against? The same period the year before, the immediately preceding period, or nothing at all?

If a programme was already running before AI Search was added, those earlier months are your baseline. You are not comparing AI Search against nothing. You are comparing the combined programme against the trajectory the old one was already on, which is a far stronger comparison and one most hotels already have the data for.

Write it down and fix it. Every number that follows has to sit inside the same window, and every claim you make later needs that window attached to it.

Step Two: Market Trends

Your results did not happen in a vacuum. Before you claim credit for anything, establish what the market was doing over the same window.

Was demand rising, flat, or falling? Was your compression coming from an event calendar rather than from anything you did?

This is what allows you to tell the real story. Direct revenue up 15% in a booming market is a very different claim from direct revenue up 15% in a market that fell. Without this step you cannot make either claim honestly.

Where to get it: Google Trends for interest over time, and STR or CoStar reports if you can access them. Your revenue manager or your regional office will often have the industry data already. National tourism and arrivals statistics are free and are frequently the most persuasive evidence in the whole document, because nobody can accuse you of having produced them yourself.

Step Three: AI Results Over the Measurement Period

Now the visibility layer. What actually changed in the AI answers themselves?

Collect share of voice against your competitive set, the number of citations and where they came from, prompt level rankings and positions, and sentiment. Look at branded and non-branded queries separately, because the non-branded ones are where new demand is actually captured.

Where to get it: your search agency, or an independent tracking platform such as SEMrush. Prefer independent measurement where you can get it. A number produced by the agency that is being judged by the number is weaker evidence, and your commercial director will spot it.

Step Four: Google Search Console

Impressions, clicks, click-through rate, and the split between branded and non-branded terms.

Expect something that looks alarming here, and do not panic when you see it. It is very common for impressions to rise sharply while click-through rate falls, because AI answers and rich results increasingly satisfy the guest without a click. Seen far more, clicked proportionally less, is the signature of this shift rather than evidence of failure. Record it accurately either way.

Also look for queries that did not exist a year ago. New non-brand content pathways appearing in your data are a discovery story that no revenue report will ever show you.

Step Five: Google Analytics

Go deeper than the summary screens. You want traffic acquisition, referral sources, and events.

In referrals, look specifically for AI platform sources and any AI assistant channel. Volumes are usually small in absolute terms. On many properties, engagement quality is disproportionately high relative to the small session volumes.

In events, you want every intent signal you track: book now clicks, booking engine handoffs, enquiry and group form submissions, calls, and clicks through to the brand site. Intent signals are the bridge between visibility and money, and they often move well before revenue does.

Step Six: Google Business Profile

Impressions, clicks, calls and direction requests.

This is the most frequently ignored dataset in hotel marketing and one of the most useful, because it captures the guest who found you, decided, and then acted somewhere your website never sees.

Step Seven: Revenue Data

This is the most important step by a distance, and it has three parts of its own.

Part One: Revenue recorded in Google Analytics. Whatever your tracking captured. Treat this as the floor, not the truth. It is almost always the smallest of the three numbers, and it is where the gap described at the top of this article comes from.

Part Two: Revenue from your brand or system reporting. The central reporting tool your chain or your booking engine provides. Better, still incomplete, and often measuring only one property of a shared website.

Part Three: Revenue from your revenue manager. This is the one that matters, and almost nobody in marketing ever asks for it.

At minimum, request:

• Total occupancy
• Average daily rate
• Average length of stay
• Total direct against total indirect revenue
• Direct share of room nights and of room revenue
• MICE, conference and events revenue
• Month by month splits across the whole window, and the comparison period

Direct share is the single most valuable number in the list. Occupancy can hold steady while the channel mix underneath it changes completely, and that shift, business moving off the OTAs and onto your own channels, is the commercial outcome that actually pays for your programme.

You will have to ask for most of this. Repeatedly. From people who do not report to you, who did not choose this project, and who do not see why it matters.

It is hard. It takes time. It is unglamorous.

Ask lots of questions and keep digging. Ask for the raw exports rather than a summary slide, because the summary has already made decisions on your behalf. Ask what changed in the hotel that year: the refurbishment, the new general manager, the concert calendar, the tracking that broke for three weeks in January. That context is not in any platform and it is the difference between a report and an explanation.

And be prepared to explain, more than once, that you are not auditing anyone. You are building the evidence that the money the hotel spent is working.

Gathering is step one of the method, not the whole of it. Once the data is in front of you, the attribution itself comes from four disciplines.

Sequence the evidence. Visibility should move first, then behaviour, then intent, then revenue. That shape is what makes a causal story credible. If revenue moves first and visibility follows, something else caused it, and you need to say so.

Rule out the alternatives in public. Whatever else was running in that window, paid social, a brand campaign, a refurbishment, name it and test it against the same data. An explanation you have not tried to disprove is not evidence.

Declare the cost. If the programme cost something, ranking positions, keywords, visibility on terms you used to own, publish the number. A case study that contains only good news is not a case study, and every experienced commercial reader knows it.

State what you still cannot see. Group and MICE enquiries that never reach a revenue line. Short measurement windows. Foundations built by work that came before. Saying these things out loud is not a weakness in the argument. It is usually the most persuasive part of it.

Then attach a source and a measurement window to every single claim. Where possible, make those sources ones owned by Google or by the hotel rather than by your agency, so that anyone who doubts the result can go and check it themselves.

There is no button for this. There is no dashboard that will hand it to you.

There is a defined measurement period, seven datasets, three layers of revenue, someone willing to ask the same question five times, and someone who knows what actually happened inside that hotel over the period being measured.

Data does not join itself. That is the whole recipe, and now you have it.

With 25+ years in hotel digital marketing, Michael has driven success for Accor, IHG, and Marriott worldwide.