Measuring AEO/GEO: How to Track AI-Driven Visibility

Here is the uncomfortable truth about AI search visibility: most hoteliers have absolutely no idea whether they show up in ChatGPT, and the ones who do “know” are usually just remembering that one time a guest mentioned it at check-in.

That is not measurement. That is a vibe.

If you are spending real time or money trying to get your property recommended by AI engines, you need to know whether it is working, and you need a way to prove the number went up. The good news is that you do not need a data science team. You need a repeatable system, a spreadsheet, and roughly two hours a month. Let me show you exactly how to build it.

Why you can’t just “check your ranking” anymore

In old-school SEO, measurement was almost insultingly simple. You typed “boutique hotel savannah” into Google, you saw a ranked list, and you found yourself at position 7 or buried on page 4. Everyone searching that phrase saw more or less the same ten blue links. One query, one answer, one number to track.

AI answers do not work like that, and pretending they do will wreck your reporting.

Three things break the old model:

• Answers are generated, not ranked. Ask Perplexity “where should I stay near the riverfront in Savannah” twice and you can get two different sets of recommendations. There is no fixed position 7 to occupy.
• Answers vary by phrasing and by person. “Quiet boutique hotel” and “romantic boutique hotel” can surface completely different properties. The model is interpreting intent, not matching a string.
• Citations are sparse. A traditional results page shows ten links. An AI answer might name three hotels and cite two sources, or name your hotel with no link at all. Getting mentioned matters even when nobody clicks.

So we stop asking “what’s my rank” and start asking a better question: across a realistic set of traveler questions, how often does my hotel get mentioned, and is that share going up? That shift, from rank to share of mention, is the whole game. If you want the conceptual groundwork on how these disciplines differ, our piece on AEO vs GEO vs SEO for hotels lays out the vocabulary so the rest of this makes sense.

The three signals you actually track

Forget the buzzwords for a second. There are exactly three things worth measuring, and they stack from “easy and noisy” to “hard and trustworthy.”

1. Prompt-panel mentions — does the AI name you when asked the questions your guests actually ask?
2. Citation and source presence — when the AI shows its work, is your website (or a page about you) among the cited sources?
3. Referral signals — are real humans clicking from AI tools onto your site, and do they book?

You want all three because each one covers the others’ blind spots. A mention with no traffic might mean the answer named you but nobody clicked. Traffic with no mentions in your panel might mean your panel is missing the prompts that actually win. Triangulate.

Signal 1: Build a prompt panel (this is the core of everything)

A prompt panel is a fixed list of real traveler questions that you run, the same way, on a schedule. The “fixed” and “same way” parts are the entire point. If you change the questions every month, you have not measured anything, you have just had three separate conversations with a chatbot.

Here is how to build one that is actually useful.

Step 1: Write 15 to 25 questions a real traveler would type. Not keywords, questions. Mix the intent types:

• Discovery: “best boutique hotels in [your city]”
• Constraint-based: “pet friendly boutique hotel near [your neighborhood] with parking”
• Occasion: “romantic small hotel [city] for an anniversary”
• Comparison: “[your hotel] vs [competitor hotel], which is better for families”
• Brand: “is [your hotel name] any good”

That brand-name question matters more than people expect. If the AI cannot describe your own property accurately, you have a content problem before you have a visibility problem, and we wrote a whole diagnostic about that in is your hotel invisible to ChatGPT.

Step 2: Pick your engines. For most independent hotels, run the panel across ChatGPT, Google’s Gemini / AI Overviews, and Perplexity. Add Copilot if you have time. Three is plenty to start.

Step 3: Log it like a scientist. For each question, on each engine, record:

That “competitors named” column quietly becomes one of the most valuable things you own. You will learn which three properties the AI thinks you compete with, and it is frequently not who you think.

Step 4: Turn it into one number. Your headline metric is share of mention: the percentage of panel questions where you got named. Twenty questions across three engines is 60 data points. Mentioned in 9 of them? That is a 15 percent share of mention. Next month it is 22 percent. That is a trend you can put in front of an owner.

Run the exact same panel, on the same day each month, before you change a single thing on your site. That first run is your baseline, and a baseline you skipped is a number you can never get back.

A quick honesty note: AI answers are non-deterministic, so a single run is noisy. Two practical fixes. Run each question fresh (new chat, no memory bleeding in from the last question), and judge the trend across several months rather than panicking over a single dip. One bad month is weather; three is climate.

Signal 2: Citation and source presence

When an engine like Perplexity or AI Overviews shows its sources, you get a second, harder signal: not just “did it mention me” but “did it trust a page about me enough to cite it.” Those cited URLs are gold, because they tell you which of your pages (or which third-party pages) the model is actually leaning on.

Track two things here:

• Are you cited at all for your panel questions, and is that share rising over time.
• What gets cited. Is it your own site, or is it a third party like a listicle, a review aggregator, or worse, an OTA page about your property? If the AI is citing the OTA’s version of you instead of yours, you have just found a leak. That dynamic, where the middlemen end up owning your own story in search, is exactly what we break down in how OTAs steal search.

The fix for thin citation presence is almost always the same: clearer, more factual, more structured content on your own pages, and a clean, accurate Google Business Profile. If your local presence is shaky, the AI has nothing solid to cite, which is why our local SEO and Google Business Profile service is usually the first lever we pull.

Signal 3: Referral signals (the money question)

Mentions are nice. Bookings pay the mortgage. So the third layer is tracking actual humans arriving from AI tools.

The catch: AI referral tracking is genuinely messy right now. Some tools pass a clean referrer, some pass almost nothing, some route through redirects that scrub the source. So you will undercount. Accept that and build the segment anyway, because a rising undercount is still a rising line.

Here is the practical setup in Google Analytics 4:

• Build a custom channel group or a regex filter that catches known AI referrers. Look for hostnames containing chatgpt, openai, perplexity, gemini, copilot, and the Google AI domains. New ones appear constantly, so revisit the list quarterly.
• Watch behavior, not just volume. AI-referred visitors often arrive further down the funnel, they have already been “pre-sold” by the answer, so compare their booking-page reach and conversion against your other channels.
• Tag everything you can control. Any link you place where an AI might read it (your profiles, partner pages) can carry a UTM so it shows up cleanly.

Pair this with your existing distribution analytics. If AI referrals start nudging up your direct bookings, that is a margin story, because every direct booking sidesteps the roughly 15 to 25 percent commission an OTA would have taken. That is not about escaping the OTAs, they are a real channel and you will keep using them. It is about a healthier balance, the kind we map out in building a healthy OTA mix.

Putting it on a calendar so it actually happens

A measurement system you run once is a hobby. The value is in the cadence. Here is a workflow that survives a busy month.

Monthly (about 90 minutes):

1. Run the full prompt panel across your engines, same day, fresh chats.
2. Log every row in your tracking sheet.
3. Calculate share of mention and citation rate.
4. Pull AI referral numbers from GA4.
5. Write two sentences: what moved and your best guess why.

Quarterly (about half a day):

• Refresh your panel questions, retire stale ones, add new traveler phrasings.
• Re-check your AI referrer regex for new tools.
• Review the “competitors named” column for shifts in your perceived comp set.
• Decide the next content or profile fix based on what the data is screaming at you.

Tie it back to channels you already run. Your AI visibility does not live in a vacuum. It rides on the same content, profiles, and structured data that feed your other discovery channels, including metasearch for independent hotels and the brand-defense work in bidding on your own brand in Google Hotel Ads. If your channel manager and rate data are a mess, AI will happily repeat the mess, which is part of why we obsess over channel manager and SEO hygiene. Clean inputs, trustworthy answers.

What “good” looks like (an illustrative example)

Let me paint a picture, and to be crystal clear, these are made-up numbers purely to show the shape of progress, not a real case study.

Imagine a 40-room boutique property runs a 20-question panel across three engines for the first time. Baseline: mentioned in 6 of 60 data points, a 10 percent share of mention, cited by their own site exactly zero times, and the AI keeps recommending a competitor down the street. Over the next quarter they fix their Google Business Profile, rewrite their location and amenities pages to be brutally factual, and add proper structured data.

By month three, the panel shows mentions in 21 of 60, a 35 percent share, with their own site cited on several discovery questions. AI referral sessions in GA4 went from a trickle to a small but real stream that converts above their site average. None of that means they “beat” anyone, it means they clawed back a bigger slice of the conversations travelers were already having, and shifted a few bookings toward direct. That is a win you can measure, and more importantly, a win you can repeat.

The one mistake to avoid

Do not start changing your website the same week you start measuring. Build the baseline first. Run the panel, log the numbers, sit on your hands for a beat. If you change everything at once, you will never know which change moved the needle, and “we did a bunch of stuff and it went up” is not a strategy you can scale. Measure, change one thing, measure again. Boring. Effective.

If staring down a 60-row spreadsheet every month sounds like exactly the kind of thing you would rather hand off, that is precisely the work we do. Our AI visibility AEO/GEO service builds and runs the prompt panel, tracks your share of mention and referrals, and turns the findings into the specific content and profile fixes that move the number, and you can see how that fits together on our pricing page. When you are ready to put a real measurement system behind your AI visibility instead of a vibe, book a call and we will set up your baseline panel together.