How I Got a Luxury Tour Operator Cited by ChatGPT and Gemini

To get this tour operator cited by ChatGPT, Gemini and Claude, I did four things in order: pages that answer a traveler's question in the first lines, with specific data and sources; schema and open access for AI crawlers; clusters of pages built around trip occasions; and press coverage and links that give the engines a reason to trust the brand. About three months in, the first quote request arrived from ChatGPT.

The real-world case of a boutique bespoke travel brand in Italy: from depending almost entirely on brand-name traffic to generating qualified quote requests directly from AI engines. Full method, data, and results — the custom consulting work I did, and the know-how I later codified into CiteProof.co.

Transparency note: this is a case study of my custom consulting work, not of the CiteProof.co SaaS product. CiteProof is the software I built by codifying the automatable part of this know-how — created for small and medium businesses that can't afford six months of consulting but still want to track and improve their AI visibility. I explain that bridge at the end. This June 2026 update extends the originally tracked two AI leads (February–March) to the additional four that came in through May. Every number about the client is internal data (Search Console, Semrush, referrer tracking and the form provider); the external context figures link to a source at the bottom, re-checked on September 29, 2026.

Starting point and audit

The client is a boutique bespoke travel operator, active since 2004, designing luxury itineraries in Italy for affluent American travelers. Excellent product, real access to exclusive experiences, an Italian founder with territorial knowledge hard to replicate. The problem wasn't the offering. It was visibility.

The initial audit, built by cross-referencing Search Console, Semrush, and a full technical scan, painted a clear picture:

• Organic traffic came almost exclusively from brand queries. Commercial keywords — the ones that bring buyers ready to spend — generated no clicks.
• Phrases like luxury italy tours and bespoke vacations italy had dropped 15 to 32 positions in the months before I took over.
• Strategic pages were effectively invisible: the tours page collected one click, the Rome destination page zero.
• Technically, dozens of pages with redirects, eleven 404 errors, and indexing issues were slowing down crawling.
• The backlink profile was significantly weaker than direct competitors in the same tier.

The percentage growth shown in previous reports was real only on paper. Growing 5,000% from 7 clicks per month means nothing. The site relied on organic and ads, with no other channel acting as a safety net.

Why traditional SEO alone wasn't enough

In 2025, the way people search for travel changed. A growing share of the audience doesn't open ten Google tabs anymore: they ask ChatGPT, Perplexity, or Gemini to organize a honeymoon in Tuscany and read the synthesized answer, with the brands the model decides to cite. In June 2025, AI platforms sent more than 1.13 billion visits to the world's top 1,000 websites, 357% more than in June 2024, and ChatGPT alone accounts for more than 80% of those visits ^[1].

That changes the rules. Ranking isn't enough: you need to become citable. The research backs this with precise numbers:

• 87% of ChatGPT search (SearchGPT) citations match Bing's top organic results, against 56% for Google's: optimizing for Bing becomes a direct proxy for ChatGPT visibility ^[2].
• In Princeton's GEO study, adding source citations raised visibility by 115.1% for websites ranked fifth; citations, quotations and statistics were the three best-performing methods, with a relative improvement of 30–40% ^[3].
• Reddit accounts for 46.7% of Perplexity's top-10 cited sources, according to Profound's analysis of 680 million citations between August 2024 and June 2025 ^[4] (more on why Reddit shows up so often in AI citations).
• In July 2025, 76% of the pages cited by Google AI Overviews ranked in the organic top 10 for the same query ^[5]; when Ahrefs repeated the study in March 2026, the share had fallen to 38% ^[6]. Classic SEO remains the foundation, but on its own it covers less and less.

Source: Seer Interactive, 500+ citations across 100 queries, February 2025 [2]

So the strategy wasn't "do more SEO." It was building content that ranks well on Google AND is engineered to be cited by generative engines. Two goals, one system, the one described in the guide to answer engine optimization and how it differs from SEO.

The AI leads started in month three, once the commercial clusters were live and the pages were readable by AI crawlers.

The content system that gets pages cited by AI (GEO/AEO)

Every page and every article follows a repeatable structure, designed so both a human reader and a language model find the answer immediately. The pillars:

The last three pillars are not a stylistic preference: in Princeton's GEO study, source citations, quotations and statistics are exactly the three interventions that moved visibility in generative answers the most ^[3].

On top of this structure, I applied a structured prompting framework: Actor, Input, Mission to define intent, and Memory, Assets, Prompt to anchor each text to real data instead of the model's memory. The tone is codified on a consistent profile across the site, with high openness and conscientiousness and low neuroticism: warm but authoritative, never empty promotion. No generic claims, always concrete detail over vagueness.

Production: clusters and pages

I organized production into thematic clusters, each with a hub page, service pages, and supporting articles, connected by disciplined internal linking (one dofollow link per destination URL, the rest nofollow) and closed by a commercial page that collects the inquiry.

In the February–March period alone, eight pages were published or optimized. In parallel, I built two three-step request forms (one generic, one dedicated to corporate retreats) to separate and qualify incoming leads.

The AI detection obstacle

Producing content at volume with AI assistance creates a real risk: detectors recognize it, and content perceived as serially generated loses credibility both with editors and models. The threshold I set myself was staying under 30% AI detection before publication.

The path wasn't linear, and this is the most useful point of the case study. No stylistic trick changes the probabilistic distribution of tokens: some richer versions went from 52% to 100% AI score, because more text gave the model more signal to work with. The breakthrough came from a precise combination, later codified into a reusable voice document:

• First-person founder voice, with autobiographical anecdotes impossible to generate.
• Named characters with verifiable details (a boat captain, a couple met at a pharmacy) that break the patterns.
• Domain-specific Italian vocabulary (40–60 terms per article) to raise reading complexity without forcing the English register.
• Systematic elimination of AI's statistical signatures: too-regular rhythm, parallel constructions, em dash overuse, missing contractions.

The strategic lesson: useful, quality AI-assisted content isn't penalized by Google, which evaluates E-E-A-T and not detector scores. But real human input, even on just a portion of the text, is the only thing that reliably lowers the score and raises perceived quality. Publishing and iterating beat searching for the perfect version.

Technical work: schema, crawlers, indexing

The best content doesn't get cited if AI engines can't read it or don't trust the source. The technical work covered three fronts:

• Structured data. TouristTrip schema for itineraries, FAQPage for question sections, and article schema with citation, about, and mentions properties to declare sources and entities.
• Crawler access. Verification that the bots that fetch pages for AI answers could access content: OAI-SearchBot and ChatGPT-User (the ones that matter for ChatGPT citations), PerplexityBot and Bingbot, with server-side rendering where needed, because AI crawlers don't execute complex scripts. Training-only crawlers such as GPTBot don't decide citations.
• Technical hygiene. Cleaning up redirects, 404s, and indexing issues from the audit, because every wasted page is crawl budget lost.

Authority: media and link building

E-E-A-T isn't a slogan: it's the trust threshold that decides whether an AI engine cites a source. I worked this front in two directions.

Earned media. In twelve months the brand collected fifteen verified press mentions, with two Tier 1 placements on The Telegraph and one on Forbes. This coverage was turned into an SEO signal with a dedicated press page and recap articles citing the publications via blockquotes and citation properties in the schema. This should be distinguished from press release syndication pickups, which exist on a different and lesser plane.

Link building. Backlink profile growth focused on quality over volume, with links from high-Domain-Rating industry publications.

Results in numbers

Visibility growth was steady. But the data point that changes the conversation with a client isn't average position: it's quote requests arriving directly from AI engines across four months. Every number in this section is internal data: Search Console, Semrush and the form provider's referrer tracking.

Internal data: referrer tracking and budgets declared in the client's request form. Total about $209,000 across 6 leads.

The six leads from AI engines

Referrer tracking showed the full path of each user. The story it tells is precise: those arriving from AI are already qualified, know the brand, have read a recommendation, and in most cases fill out the form quickly — a signal of high intent. That is not a quirk of this client: in Semrush's study of 500+ high-value topics, a visitor coming from an AI engine converts 4.4 times more often than one from traditional organic search ^[7]. The size of that gap varies a lot from study to study and from sector to sector; the article on the conversion rate of AI referral traffic compares them.

For privacy reasons I've anonymized the leads. The data is verifiable in the client's form provider.

Internal data: referrer tracking, budgets declared by the leads in the quote request form, February–May 2026.

Patterns observed across all six leads:

• High intent: nearly all filled out the quote request form within minutes of the initial click. These aren't curious clickers — they're prospects who've already decided to request a quote.
• Mid-to-high budgets: average declared budget around $35,000 per trip (range $16K–57K). AI-sourced leads for this luxury brand are in the high tier of the segment.
• Three different AI engines: the fact that leads are distributed across ChatGPT, Gemini, and Claude.ai confirms that AEO work produces a cross-engine effect — it's not a single algorithmic accident.
• Distribution over 4 months: leads aren't concentrated in a single period. There's continuity — a signal that the system built is producing recurring returns, not an isolated spike.
• Trip occasions: 3 anniversaries or milestone birthdays, 1 honeymoon, 1 friends trip and 1 multi-city couple trip. Vertical specialization by trip occasion proved effective.

These aren't anonymous clicks. They're people who asked an AI to recommend a trip to Italy, received the client's brand as the answer, and arrived ready to talk. That's exactly the goal the system was designed for.

What I take away from this project

• In luxury travel, AI engine citability is now a real acquisition channel, not a future bet. Six leads in four months on a boutique brand is not an isolated case.
• The three major AI engines (ChatGPT, Gemini, Claude) produce qualified leads if the site is built around AEO pillars. You don't need to be "Google's #1 result" — you need to be citable: on average across the AI assistants Ahrefs studied, only 12% of the cited URLs rank in Google's top 10 for the same question ^[8].
• Content wins when it answers the question immediately, brings specific data, and demonstrates direct experience. Vagueness and generic claims don't get cited.
• Technical structure matters as much as text: without schema and crawler access, even the best content stays invisible to AI.
• External authority — earned media and quality backlinks — is the trust threshold that unlocks citation.
• Publishing and iterating beats searching for perfection. Every week out of the market is rankings and leads lost.

How this know-how became CiteProof.co

Six months of custom work on a single luxury client generated 6 tracked AI leads for over $209,000 in declared total budget. Great for that client. But the model has a scaling problem: work like this costs $1,500–5,000 per month in consulting, and that's simply out of scale for most small and medium businesses. A dentist in Brooklyn, a law firm in Austin, a boutique hotel in California can't afford six months of consulting for a channel that today is still a small share of their revenue — but that by 2027 could mean far more.

Yet the problem is here now. AI platforms already sent 1.13 billion visits to the top 1,000 websites in a single month, growing 357% in a year ^[1], and the pages those assistants cite are mostly not the ones at the top of Google: on average across the AI assistants Ahrefs studied, only 12% of them rank in the top 10 for the same question ^[8]. When the AI answers, it does so with 2–3 names, not 10 links. If you're not in those 2–3, for those customers you don't exist — even if your Google SEO is perfect.

I built CiteProof.co to answer exactly this market fracture. It's the SaaS I created by codifying an automatable part of the work you just read about: it asks ChatGPT, Perplexity, Gemini, Claude and Google AI the questions your customers would ask, measures the answers in a monthly audit (with an on-demand re-measure), checks schema and AI crawler access, and verifies each fix before counting it.

The signal that traced the six leads in this case study, a visitor arriving from a link in a ChatGPT, Gemini or Claude answer, is the same one CiteProof counts, next to the pages the assistants open while they write an answer.

The guide to the AI Bots page explains each card. To see what the first measurement looks like on a different business, read a real check of which AI engines cite a site and who they cite instead.

It doesn't replace custom work. It democratizes it. For that boutique hotel that can't spend $4,500/month but can invest $29/month in an AI agent watching its citability, it's the difference between being visible and not. For that dentist, same.

CiteProof starts at $29/month for the first agent, with a free scan (no signup) for anyone who wants to see whether the problem exists for them in the first place.

Want AI engines to recommend your brand?

If you're a luxury travel or hospitality brand with budget for a 4–6 month custom engagement, we can work directly on your project. If you're an SMB looking to monitor your AI visibility without dedicated consulting, CiteProof is the right path.

Either way: hello@citeproof.co

Frequently asked questions about this case study

How do you know a lead came from ChatGPT?

From the referrer recorded by the form provider and the site's analytics: someone who clicks a link inside a ChatGPT, Gemini or Claude.ai answer arrives with a recognizable referrer (for example chatgpt.com or gemini.google.com). Every lead in this case study was traced back to its engine that way, with the date of the request.

How long did it take from the start of the work to the first AI lead?

About three months. The first tracked lead from ChatGPT is dated February 24, 2026, after the commercial clusters (honeymoon, anniversaries, family) were live, the pages were readable by AI crawlers and the press coverage had started to build authority.

Do you need to rank first on Google to be cited by AI?

No. On average across the AI assistants Ahrefs studied, only 12% of the cited URLs rank in Google's top 10 for the same question, while 87% of ChatGPT search citations match Bing's top results. SEO remains the foundation, but citability depends on structure, specific data and sources the engine can verify.

How do I get my business cited by ChatGPT?

The order that worked here: pages that answer the customer's question in the first lines, with specific data and sources; question-format headings and FAQ schema; AI crawlers allowed in and no broken pages; clusters of pages around the occasions people buy for; and press coverage and links that give the engines a reason to trust the brand. Then check every month whether the engines name you, engine by engine.

Can a small travel business get recommended by ChatGPT?

This one did: a founder-led boutique operator received three quote requests from ChatGPT, two from Gemini and one from Claude.ai between February and May 2026. Size mattered less than being the clearest, best-sourced answer for specific trips such as anniversaries and honeymoons.

Sources

1. [1] AI referrals to top websites were up 357% year-over-year in June, reaching 1.13B · TechCrunch on Similarweb data · July 2025 · accessed 2026-09-29
2. [2] 87% of SearchGPT Citations Match Bing's Top Results · Seer Interactive · February 2025 · accessed 2026-09-29
3. [3] GEO: Generative Engine Optimization · Aggarwal et al., Princeton University / IIT Delhi (arXiv, KDD 2024) · November 2023, v3 2024 · accessed 2026-09-29
4. [4] AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information · Profound · June 2025 · accessed 2026-09-29
5. [5] 76% of AI Overview Citations Pull From the Top 10 · Ahrefs · July 2025 · accessed 2026-09-29
6. [6] Update: 38% of AI Overview Citations Pull From The Top 10 · Ahrefs · March 2026 · accessed 2026-09-29
7. [7] We Studied the Impact of AI Search on SEO Traffic | Our Findings · Semrush · June 2025 · accessed 2026-09-29
8. [8] Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt · Ahrefs · 2025 · accessed 2026-09-29