By Tessar Napitupulu, Founder & CEO, PT Arfadia Digital Indonesia, GEO pioneer in Indonesia since 2023. More from Tessar.
When a July 2026 study asked Google's AI to plan a four-day Bali trip, it built the answer from a tour operator's blog-style itinerary article, hour by hour, price by price. It did not pull from the same operator's own booking page, even though the booking page almost certainly had the same trip for sale. The difference was not authority, freshness, or even accuracy. It was shape. One page was built like an answer. The other was built like a form.
That distinction, content shaped like an answer versus content shaped like a transaction tool, is the single most actionable lever available to a travel operator trying to earn AI citations in 2026, and it costs nothing but editorial discipline to act on.
What Does "Shaped Like an Answer" Actually Mean?
An AI system assembling a response is not reading a page the way a human does. It is scanning for self-contained, extractable statements it can lift with confidence and attribute to a source. A page that buries its facts inside long unstructured paragraphs, behind an interactive filter, or inside JavaScript that only renders after a click, gives the system nothing clean to extract, regardless of how good the underlying information is.
The winning structure, observed consistently across the July 2026 study's 94 answered queries, follows a repeatable pattern: a short opening paragraph that commits to a direct answer immediately, a labeled list or day-by-day sequence where every item carries a specific, checkable fact, and a short closing section anticipating two or three natural follow-up questions. This is not a stylistic preference. It mirrors, almost exactly, the internal structure a retrieval-augmented generation system uses to decide what to extract and cite.
The Itinerary That Got Cited, Broken Down
The specific example worth studying line by line: a Bali four-day itinerary answer stated "Hari 2, 13.00 tiba di lokasi, 14.30 Pantai Melasti, Rp 5.000 tiket masuk." Every clause in that sentence is independently checkable: a day number, a time, a named place, a price. An AI system can lift that clause whole and attribute it correctly, because there is nothing ambiguous inside it to misinterpret. Compare that to a typical destination-guide sentence like "Melasti Beach offers stunning views and is a must-visit spot," which contains no fact an AI system can extract with confidence at all. Both sentences might sit on the same page. Only one of them is doing citation work.
The Four Parts an AI Actually Extracts
Reconstructed from the consistent structure across 94 answered queries in the July 2026 Indonesian study.
Direct-answer opening
The conclusion stated in the first two sentences, before any scene-setting or marketing language.
Labeled, sequenced facts
Day numbers, times, named places, and prices, each as its own extractable clause.
Checkable specificity
"Rp 5.000 tiket masuk," not "affordable entrance fee." Vague language has nothing for a retrieval system to lift.
Anticipated follow-ups
A short FAQ addressing budget, timing, or party-size variations the AI's own next turn is likely to ask about.
Created by Arfadia • arfadia.com/blog
Why Doesn't the Booking Page Work the Same Way?
A booking page is built to let a human filter live inventory: date pickers, dropdown menus, dynamic pricing that changes on every reload, and a call-to-action button. Every one of those elements is useful for a human ready to transact and actively harmful for an AI system trying to extract a stable fact. A price that changes based on session state is not a citable fact, it is a moving target. A date picker contains no default answer to lift. None of this makes the booking page worse at its actual job. It makes it a fundamentally different kind of page, built for a different moment in the traveler's decision, and citation systems are not shy about routing around it entirely in favor of a page that does have something stable to say.
The operational implication is not to redesign the booking page. It is to make sure a second, separate page exists, the itinerary or planning article, purpose-built to be the citable one, linking through to the booking page at the natural moment a reader is ready to transact.
What Schema Turns This Structure Into Something Machines Can Validate?
Prose structure earns partial credit. Structured data closes the gap between "well written" and "machine-verifiable." TouristTrip schema, with its itinerary property populated as a full ItemList, is the highest-priority markup for any page describing a multi-day plan: each day becomes a discrete, indexable Place or ListItem with position and description intact, rather than a paragraph a machine has to parse and guess at. FAQPage schema should sit on exactly the visible questions and answers already on the page, never on hidden or injected text, since mismatched visible-versus-structured content is treated as a quality signal against the page. Offer, with priceCurrency and availability populated, turns a price mentioned in prose into a validated, machine-readable fact. None of these three guarantee a citation on their own. Google's own guidance is explicit that structured data affects eligibility, not display. What the schema does is remove ambiguity a machine would otherwise have to resolve on its own, and machines that have to guess tend to cite the page that didn't make them.
| Page type | Primary job | Citation eligibility |
|---|---|---|
| Booking page | Let a human filter live inventory and transact | Low. Dynamic content, no stable facts to lift |
| Itinerary article | Answer a specific planning question in depth | High, when structured with named facts and TouristTrip schema |
| Destination guide | Build topical authority and inspire | Moderate, depends heavily on specificity of individual claims |
| Trust and credentials page | Establish licensing and legitimacy | High for trust queries specifically, where government sources are often under-optimized |
Does This Same Structure Help With Traditional SEO Too?
Mostly yes, which is one of the more useful things about this approach. A direct-answer opening, specific named facts, and validated schema are core E-E-A-T signals under Google's own guidelines regardless of whether an AI Overview ever fires. The overlap is real but imperfect: content optimized purely for keyword density without this answer-shaped structure can still rank reasonably in classic organic results while remaining almost invisible to AI citation, since ranking and citation are measured by different mechanisms even when they reward some of the same underlying quality.
Same Trip, Same Domain, Opposite Citation Outcome
Why the booking page lost the citation and the itinerary article won it, on the same underlying product.
Booking page
Itinerary article
Created by Arfadia • arfadia.com/blog
Why Does Consistency Across Platforms Matter as Much as Structure on One Page?
An AI system rarely draws its confidence in a citation from a single page in isolation. It cross-references. If a package is named "Paket Flores 5D4N Explorer" on the website, "Flores Explorer Package" on the Google Business Profile, and "Paket Explore Flores 5 Hari" in an Instagram caption, with three slightly different prices attached across the three, the system has no clean way to resolve which version is authoritative. The practical effect is not that the operator gets penalized, it is that the operator becomes a less confident citation candidate than a smaller competitor whose name, address, and price agree everywhere a system might look.
This is the same entity-resolution principle search engines have used for years in local SEO, extended into a domain where the AI cannot ask a human to resolve the ambiguity, and simply routes around it instead. The fix is unglamorous and almost entirely a discipline problem rather than a technical one: pick one canonical name and price for every package, and update every surface, website, Google Business Profile, Instagram, and any directory listing, on the same day a price or itinerary changes.
A Concrete Before-and-After
Before: "Jelajahi keindahan Flores bersama kami dalam paket wisata terbaik dengan harga terjangkau." Inspirational, but contains zero facts an AI system can extract, verify, or attribute. After: "Paket Flores 5D4N mulai Rp 3.800.000/pax, termasuk open trip Kelimutu, Bajawa, dan Riung 17 Islands, keberangkatan 14 Juli dan 28 Juli 2026." Same product, same operator, same page position in the funnel. Only the second version gives a retrieval system something concrete to lift and attribute correctly, and only the second version is likely to survive being cross-checked against the operator's own Instagram caption for the same trip.
What Does This Look Like as a Repeatable Editorial Checklist?
Five checks, applied to every new itinerary or planning page before it is published: does the opening two sentences commit to a direct, specific answer; does every itinerary line carry a named place, a time, and where relevant a price; is the exact same package name and price used on the website, the Google Business Profile, and the most recent Instagram caption for that package; is TouristTrip schema present, validated, and populated with a full itinerary ItemList; and does a closing FAQ section anticipate the two or three follow-up questions a traveler, or an AI system's own next turn, would naturally ask. A page that passes all five is not guaranteed a citation. A page that fails two or more of them is very unlikely to earn one, regardless of how good the writing otherwise is.
How Should an Operator Restructure Existing Content?
Start with an audit, not a rewrite. Pull the ten highest-traffic planning and itinerary pages on the site and check each against four questions: does it open with a direct answer inside the first two sentences, does every itinerary line carry a specific checkable fact, is TouristTrip schema present and validated, and does a closing FAQ address the natural next questions. Pages that fail two or more of those checks are the highest-value rewrite targets, not because they are poorly written, but because they are the ones most likely to be losing a citation to a thinner competitor that happens to be structured correctly.
New content should be built to this shape from the outset, paired with the long-tail keyword architecture described in our piece on why OTAs own generic search but not the booking, and measured for citation presence the way described in our GEO for travel and tour operators service.
Frequently Asked Questions
Should we delete the marketing language from our itinerary pages entirely?
No, but sequence it correctly. Lead with the direct, factual answer, then let marketing and inspirational language follow. An AI system extracting the page will lift the early, factual content regardless of what comes after it.
Does this mean every page on our site needs TouristTrip schema?
No, only pages actually describing a multi-day trip or itinerary. A single-activity page is better served by different schema entirely, such as an Event or a simple Offer.
Our prices change seasonally. Does that break the schema-based approach?
No, update the Offer schema's price field on the same cycle you update the visible price. The requirement is that structured data and visible content match at any given moment, not that the price stays fixed forever.
How do we know if our current pages are being cited or not?
Run a fixed panel of the actual queries your travelers ask, across ChatGPT, Perplexity, Gemini, and Google AI Overviews, and log whether your domain appears as a citation. Repeat weekly given documented month-to-month citation drift in the 40 to 59% range.
Is this approach only relevant for AI citation, or does it help with anything else?
It measurably overlaps with traditional E-E-A-T signals Google already rewards in classic search, so the same rewrite work tends to improve both outcomes, even though the two are measured through different mechanisms.
How much does it matter if our package name is slightly different on Instagram than on our website?
More than it looks. An AI system cross-references entity details across surfaces to decide how confident it can be in a fact. A mismatched name or price does not get penalized so much as it gets treated as unresolved, and unresolved facts tend to lose the citation to a competitor whose details agree everywhere.
Do we need a developer to implement TouristTrip schema correctly?
Not necessarily a developer specifically, but someone comfortable validating structured data against Google's Rich Results Test. Many CMS platforms support schema through plugins or templates; the harder part is populating it accurately and keeping it in sync with the visible page content, which is an editorial responsibility, not just a technical one.
Sources & References:
- Search.Agency, "Travel SEO and AI Overviews in 2026, a 105-Query Study," 3 to 4 July 2026: the Bali four-day itinerary citation example and answer-shaping pattern across 94 answered queries.
- Google Search Central, structured data general guidelines: schema affects eligibility for enhanced display, not guaranteed display or ranking.
- Schema.org, TouristTrip, FAQPage, and Offer type documentation.
- Princeton University, Georgia Tech, the Allen Institute, and IIT Delhi, "GEO: Generative Engine Optimization" research paper, for the underlying citation-extraction mechanics referenced.