precision_manufacturing Zone · The Lab

AI Discoverability & Presence

calendar_today Wednesday 3 June

schedule 40 minutes

groups Two facilitators

Isabel Mosk

Destination Marketing Strategist
Sherpa's Stories

Lili-Sheryl Tchepelova

Marketing & Insights Executive
Digital Tourism Think Tank

1 / 7

visibility The shift in how travellers find destinations

Search and discovery are changing.

Travellers no longer ask Google "things to do in Brussels." They ask AI for a weekend that suits how they travel. The destinations that get cited in that answer win the trip. The ones that don't, vanish from the consideration set entirely.

Then · the keyword era

search things to do brussels

Visit Brussels , Top 10 Things

TripAdvisor , 50 Best in Brussels

Time Out Brussels

Now · the answer era

smart_toy "I'm planning a long weekend in a European capital with serious food, good design culture and rail-friendly day trips. Somewhere that's not Paris or Amsterdam. Where would you recommend?"

AI answer

For that brief, I'd recommend Antwerp or Ghent , both have a strong design and food culture, and Antwerp in particular has rail-friendly day trips to Bruges or the coast. Brussels would also fit but it gets less attention in this kind of brief , mentioned for completeness.

Cited: cntraveler.com, theguardian.com, visitflanders.com

1 in 3

searches now return an AI Overview before any blue links

70%

of AI citations come from sources outside the SEO top 10

58%

of Google searches now end without a single click to any website

What you're about to do

Pick one of three lenses on the next slide. Each lens audits a different layer of your destination's AI presence , the technical foundation, the competitive position, or the experience an AI agent has when it actually tries to plan a trip on your site. You leave with a structured audit you can take back to your team.

2 / 7

layers Three lenses · pick where to start

Pick the lens that matches where you are.

Each lens is a complete audit on its own. Most participants work through one in 20 minutes. The Foundation Audit is the recommended starting point. The Competitive lens deepens what you have learned. Pushing It Further is the one that produces uncomfortable findings.

layers

Lens A · Recommended start

Foundation Audit

Foundational

Technical and content together. Schema, crawler access and content structure in one connected flow. Codex generates the fixes you find at the end.

Schema check Crawler access Content structure Codex fixes

For: any team starting their AI presence work. Requires only a browser and the free Rich Results Test.

compare_arrows

Lens B · For confident teams

Competitive Presence

Intermediate

Reverse prompt engineering, citation gap analysis and direct content comparison. Ask AI not just what it says about your destination, but what the sources it cites have that yours does not.

Reverse prompting Citation analysis Content benchmarking

For: content and marketing leads with the basics in place. Needs Claude or ChatGPT in a browser tab.

rocket_launch

Lens C · Most challenging

Pushing It Further

Advanced

Autonomous AI agents attempting a full planning journey on your site. Not asking AI about your destination, asking AI to complete the booking research. Exposes where your content actually breaks the experience.

Claude in Chrome ChatGPT Agent Full journey test Failure analysis

For: digital leads, strategy teams, anyone who already audits regularly. Requires Claude in Chrome or ChatGPT Plus with agent mode.

3 / 7

layers

Lens A · Foundation Audit

Technical and content. Together, not separately.

Run Google's free Rich Results Test on three of your pages.

Schema markup is the hidden code that tells AI and search engines exactly what is on a page, that this is a TouristAttraction, that the opening hours are these, that the location is here. Without it, AI guesses. And it often guesses wrong, or cites a competitor with better markup.

Crawler access is the second half of the foundation. AI training crawlers like GPTBot, ClaudeBot and Google-Extended need permission to read your site. If your robots.txt blocks them, you are invisible to those AI systems entirely.

build

Free tool · open this in a new tab

Rich Results Test

search.google.com/test/rich-results arrow_outward

Paste your homepage URL · note the schema types found and any errors
Paste an attraction or experience page URL · same check
Paste an events or itinerary page URL · same check
Open your-domain.com/robots.txt in a browser · check whether GPTBot, ClaudeBot and Google-Extended are allowed

Capture · what did you find?

Strong schema on weak content is the wrong cited source.

If you fix the markup but the content underneath is shallow, AI cites you for things you do not want to be cited for. Content structure means: specific named places, verifiable details (dates, prices, distances), clear hierarchy, scannable answers to real visitor questions.

Check three of your key pages against five structural signals. Most destination sites score honestly at two or three out of five. That is normal, and it is also actionable.

Specific named places throughout the page Named venues, named routes, named towns. Not "various attractions" or "the local area."

Verifiable details, dates, prices, opening times, distances Things AI can extract as facts. Not "all year round", actual months. Not "affordable", actual prices.

FAQ-style answers to real visitor questions Direct answers to specific questions. Not paragraphs of prose where the answer is buried.

Itineraries with named, sequenced stops "Day 1 morning, start at X. Day 1 afternoon, drive 20 minutes to Y." Not generic suggestions.

Clear headings and a logical page hierarchy H1, H2, H3 used semantically. AI uses heading structure to understand what each section is about.

Hand your findings to Codex or ChatGPT and get the corrected code back.

Codex is OpenAI's coding agent, you paste your audit findings, it generates the corrected schema JSON-LD and the corrected robots.txt that your web team can implement directly. You can also use ChatGPT or Claude directly with the same prompt.

You do not need to write any code. The output is ready-to-implement. Your web team takes it as-is.

The prompt · copy and run with your findings Codex · ChatGPT · Claude

I have run a foundation audit on my destination website. These are my findings across three pages and my robots.txt: [PASTE YOUR FINDINGS FROM THE PREVIOUS TAB] Based on these findings, generate: 1. Corrected JSON-LD schema for each page type that had errors or no schema. Use the most appropriate schema.org types for a destination management organisation, typically TouristAttraction, Event, FAQPage, ItemList for itineraries. Include the required and recommended properties for each. 2. A corrected robots.txt file that explicitly allows GPTBot, ClaudeBot, Google-Extended, PerplexityBot and CCBot, while keeping any existing rules for traditional search crawlers intact. 3. A two-line comment above each block of generated code that explains what changed and why. Output ready-to-implement code blocks only. No surrounding explanation needed.

Capture · what did Codex give you?

Or switch lens:

4 / 7

compare_arrows

Lens B · Competitive Presence

What the sources cited instead of you actually have.

Ask AI the questions your visitors are asking, and see what it cites.

Reverse prompt engineering means starting from the traveller's question rather than from your content. You run real conversational queries through Claude or ChatGPT and observe what gets cited. The destinations and sources that appear instead of you are your real competitive landscape in the AI answer space, not the SEO competitors you have been tracking.

Run both prompts below. The first tests whether AI discovers your destination unprompted. The second tests what it says when your destination is named directly.

Prompt 1 · Dreaming query Claude · ChatGPT

Tests whether AI surfaces your destination unprompted. Do not name your destination. If it does not appear, that is your first finding.

I am planning a long weekend in [YOUR REGION OR COUNTRY] in spring. I want somewhere that feels genuinely rural, offers good walking and has a strong sense of seasonal character, something that changes noticeably through the year. I am not interested in a city break. What destinations would you recommend and why? Cite your sources.

Prompt 2 · Planning query Claude · ChatGPT

Tests what AI says when your destination is named. Watch which sources it cites, those are who is competing with you for the citation.

What are the best things to do in [YOUR DESTINATION] if I only have two days? I enjoy walking, local food and finding places that are not overrun with tourists. Give me a specific plan with named places and cite every source you draw from.

Capture · which sources got cited?

Read what AI cited, and ask AI exactly what it has that you do not.

Most audits stop at "we were not cited." This one goes further. You take the URLs of the sources that were cited and feed them into Claude or ChatGPT alongside your own destination URL, then ask AI to do the comparison directly. The output is a structured content gap brief, not a list of misses.

This step is where the audit becomes adversarial. AI will tell you uncomfortably specific things about what the cited sources have that you do not.

The comparison prompt Claude (best) · ChatGPT

Paste up to three cited source URLs and your own URL. Claude handles this better than ChatGPT because it reads URLs more reliably.

Compare the content at these cited source URLs: [CITED URL 1] [CITED URL 2] [CITED URL 3] …with the content at my destination URL: [YOUR DESTINATION URL] Tell me specifically what the cited sources have that mine is missing. Be precise about content structure, depth of information, types of detail and editorial format. Do not be polite, be specific. Output as a content gap brief with five gaps. For each gap: (1) the missing element, (2) where the cited source handles it well, (3) a concrete recommendation for my destination, (4) the rough effort level, quick fix, content sprint, or strategic rebuild.

Capture · the five gaps AI found

Turn the gaps into a brief your team can act on next week.

The findings from the previous step are diagnostic. This step turns them into a working brief. Three quick-fix items you can do in the workshop, three content sprint items for the next month, one strategic question for leadership.

The structure matters. A list of 20 gaps gets ignored. A brief with seven prioritised items in three time horizons gets actioned.

The brief prompt Claude · ChatGPT

Based on the five content gaps below, produce a structured content brief for a destination marketing team. Format it as three time horizons: QUICK FIXES (this week, no budget needed), three items, each with the page to edit, the change to make, and the sentence-level rewrite if it is short enough to include. CONTENT SPRINTS (this month, may need a writer), three items, each with the new page or section to commission, a working title, and the structure outline. STRATEGIC QUESTIONS (next quarter, needs leadership), one or two items, each with the question and what the answer unlocks. The five gaps: [PASTE THE FIVE GAPS FROM THE PREVIOUS STEP] Do not include explanation or commentary in the output. Format as a clean brief ready to share with the team.

Capture · the brief that came back

Or switch lens:

5 / 7

rocket_launch

Lens C · Pushing It Further

Let an AI agent try to plan a trip from your site.

Two tools. Both browse the web autonomously on your behalf.

This lens uses AI agents that can act on the web, not just answer questions about it. They open tabs, click links, fill forms, scroll pages and read what they find. You give them a goal, you watch them work and you capture where they failed.

Before you pick a mission, get familiar with what each tool actually does and what you need to run them.

explore

Claude in Chrome

Browser extension · operates your actual browser

Claude in Chrome takes over a Chrome window and browses on your behalf. You see exactly what it does, tabs open, pages load, the cursor moves. Useful for tasks that need a real browser session: logged-in dashboards, multi-step forms, sites that block automated traffic.

How to access: install the Claude extension from the Chrome Web Store. Requires Claude Pro or Max.
What to watch: Claude narrates each step. You can pause or take over at any time.
Best for: when you need the agent to behave like a real visitor on your real site.
Watch out: slower than the alternative because it actually moves through pages. Budget 10 minutes per mission.

terminal

ChatGPT Agent mode

Cloud-based agent · runs in a virtual environment

ChatGPT's agent mode runs in OpenAI's cloud. It has its own browser, can write and run code, can produce files. Faster than Claude in Chrome because it is not constrained to your screen, but you are watching a recorded session rather than the live one.

How to access: chat.openai.com · select Agent in the model picker. Requires ChatGPT Plus or Pro.
What to watch: the agent streams a live log of what it is doing. You can interrupt.
Best for: when speed matters and the site does not block automated browsers.
Watch out: some sites block cloud agents at the firewall. If your site does, switch to Claude in Chrome.

Three missions. Each one is built to surface a different kind of gap.

Each mission gives the agent a complete planning task, not a question, a task. The agent will attempt to do what a real visitor would do, on your site, and will tell you where it had to give up or go elsewhere. Pick the mission that would expose the most uncomfortable findings for your destination.

Mission 1

The accessibility planner

The hardest test most destination sites fail. The agent plans a four-day trip for a couple where one person uses a wheelchair, using only your site.

Claude in Chrome · ChatGPT Agent

Act as a travel planner for a couple where one person uses a wheelchair. They want a four-day trip to [YOUR DESTINATION] in late April. Browse [YOUR DESTINATION URL] autonomously and produce: 1. A day-by-day itinerary with accessible routes only 2. Accommodation with verified step-free access 3. Transport options that do not require driving 4. Specific opening times and prices for each attraction 5. A list of every question you could not answer from the site alone Be honest about what the site failed to provide. Do not fill gaps from your training data, tell me where my site fell short.

Mission 2

The seasonal authority test

Tests whether your site demonstrates genuine seasonal expertise. The agent is briefed as a travel writer producing an off-season feature, and asked which competitor sites it had to use to fill the gaps.

Claude in Chrome · ChatGPT Agent

Act as a travel writer being commissioned to produce a "what to do in [YOUR DESTINATION] in November" guide. Browse [YOUR DESTINATION URL] autonomously and tell me: 1. What specific November content exists on the site 2. What is genuinely unique to this destination at this time of year 3. What content gaps would force you to draw from competitor sites 4. Which competitor sites would you go to instead, name them Be specific about which pages on the destination site you used and where you had to look elsewhere.

Mission 3

The booking pathway test

Where does the site lose the high-intent visitor? The agent plays a ready-to-book traveller and maps every friction point on the journey from inspiration to booking.

Claude in Chrome · ChatGPT Agent

Act as a high-intent visitor ready to book a three-night trip to [YOUR DESTINATION] for two adults in May. Browse [YOUR DESTINATION URL] autonomously and attempt to complete the full booking journey: choose accommodation, identify experiences, plan transport. Track every point where you had to leave the site to make a decision, every dead end, every moment a competing source provided the answer instead. Output a friction map of the journey with every drop-off point named.

Run the mission. Watch what happens. Capture where it broke.

Paste the mission prompt into your chosen tool. While it runs, watch which pages it visits, when it pauses, where it goes off-site. Most missions take five to ten minutes to complete. The interesting parts are not the success steps, they are the moments the agent had to give up, guess, or rely on a competitor.

While the agent is running, watch for:

visibility

The questions it asks itself out loud The agent narrates. "I cannot find pricing information." "The page does not say if this is accessible." Those questions are your content gaps.

redo

The pages it visits twice If the agent revisits a page, something on it was confusing. If it backtracks across the site, your navigation is broken for AI just like it is for humans.

launch

The external sources it falls back to Every time the agent leaves your site, a competitor wins. Note which domains it goes to. Those are who is filling your gaps for you.

cancel

The questions it admits it cannot answer A good agent will explicitly say "I could not find X." That is your content roadmap.

Capture · what the agent reported

Mission outcome:

Turn agent failures into a structured roadmap.

Agent failures fall into three categories. Knowing which category a failure belongs to determines who fixes it and how. This is what makes the agent test useful, not the failures themselves, but the structured read on them.

Category 1

Content gap

The content does not exist. Should be written. The web team or an agency writes it.

Category 2

Structure gap

The content exists but the agent could not find it or use it. Information architecture problem. The digital team restructures.

Category 3

Strategic gap

A capability the destination needs to invest in. Booking integration, accessibility verification, a real-time data feed. Leadership decision.

The analysis prompt Claude · ChatGPT

Paste the failures you captured. The prompt asks AI to sort each one into one of the three categories and propose a specific next step.

An autonomous AI agent attempted [WHICH MISSION] on my destination site and these are the failures it reported: [PASTE FAILURES FROM THE PREVIOUS TAB] For each failure, categorise it as one of: 1. Content gap, content that should exist but does not. Action: write it. 2. Structure gap, content that exists but the agent could not find or use. Action: restructure or surface it. 3. Strategic gap, a capability the destination needs to invest in. Action: a leadership decision. For each failure, give: the category, a specific next step, and an owner (web team, content team, digital lead, or leadership). Format as a numbered list. Do not add commentary outside the list.

Capture · the categorised roadmap

Or switch lens:

6 / 7

workspaces Your output

Take it back to your team.

A clean summary of what you captured, plus what to do with it in the next 14 days. AI presence is a moving target, the destinations that stay visible are the ones that audit again, not the ones that audit once.

Your AI Presence Audit

From your XDW 2026 Lab session.

Run again quarterly

AI citation patterns shift fast.

A·1Schema & crawler access

Not completed.

A·2Content structure signals

Not completed.

A·3Generated fixes (Codex)

Not completed.

B·1Reverse prompt findings

Not completed.

B·2Content gap brief

Not completed.

B·3Prioritised content brief

Not completed.

C·3Agent mission findings

Not completed.

C·4Categorised roadmap

Not completed.

Next 14 days

Three things to do before this fades.

01

Share the findings with your team this week.

Screenshot or PDF this summary and bring it to your next team meeting. The longer you wait, the more it feels like a workshop artefact rather than an action list.

02

Pick the one thing your team can ship in the next two weeks.

Across all your captures, identify one quick fix. Schema correction, an FAQ section, a piece of restructured content. Ship it before the urgency fades.

03

Schedule the next audit for three months from today.

AI citation patterns move. The audit is a quarterly habit, not a one-off. Put it in the calendar now, the lens you skipped today is the one to run next.

open_in_new

The AI Transparency Framework

The wider DTTT framework that contextualises this work.

7 / 7