Instagram beats every OTA in Indonesia's travel AI Overviews.

Google's AI cited Instagram 104 times across the Indonesian travel answers we recorded in early July. Traveloka, the market's biggest OTA, collected 72 citations, tiket.com took 33, and no other travel platform reached 20. We expected the big platforms to dominate before we started, and that expectation is exactly why we are publishing the full numbers instead of a hot take.

How the study ran

On July 3 and 4 we ran 105 buying-intent travel queries in Bahasa Indonesia through Google Search and recorded what the AI Overview did with each one. The set covers flights and OTA apps, hotels, tour packages and open trips, umroh, travel insurance and visas, local transport, and destinations, phrased the way real buyers type them, so "aplikasi booking hotel termurah" and "travel umroh murah tapi amanah" rather than anything a marketer would write. For every query we saved the full answer text, classified its structure, and logged every source it cited in order. 94 of the 105 queries produced an AI answer, and those answers carried 722 citations across 298 unique domains.

That last number deserves a pause. 298 different domains earned at least one citation in a single two-day snapshot of one vertical. The answer layer is not a closed club of ten big sites, it is a wide-open field where a single well-matched page from almost anywhere can surface, which cuts both ways for every brand in the market.

Trigger rates were not even across categories.

Two caveats belong next to every number here. The study ran from one logged-in profile in the Jakarta metro area, and the answers are personalized to a degree that surprised us, one umroh answer recommended agencies near the searcher's own neighborhood with drive times attached, and a year-end holiday answer ranked destinations by distance from "area Anda". Trigger rates and structural patterns are the stable findings, while the exact brands occupying each slot will differ per user and per week. The raw dataset, including full answer text for all 105 queries, is available if you ask us for it.

Instagram works as citation infrastructure

Instagram's 104 citations were not profile links padding out the answers. Vendor reels and captions were quoted as evidence for prices, package contents and recommendations, and 69 of the 94 answers cited at least one social or UGC source. On the cheapest-transport-to-Ubud query, the AI's lead source for the Rp 4.400 public bus fare was a TikTok video, ahead of Trip.com's own guide. On the sharia-financing umroh query, three of the seven citations were Instagram reels, mostly from one travel agency's own account.

Google's AI treats an information-dense caption from a tour operator the same way it treats a blog post, as a page with facts it can lift. For a travel brand the implication is direct. The Instagram account is no longer just a feed that markets to followers, it is a source the AI reads and quotes, so captions carrying real prices, named packages, dates and locations give the AI something to cite. The operators winning citations today appear to be doing it by accident, which means the ones who do it on purpose start with an advantage.

The AI answer switches off where Google sells inventory

Hotels was the weakest category at 11 of 18, and the misses follow one pattern. Queries naming a city and asking for bookable rooms got the Google Hotels module with sponsored listings and auto-applied filters instead of an AI answer, and named-route flight queries got Google Flights the same way. One tour-package miss went further, replaced entirely by a local operator's business profile with bookable packages. Where Google has its own inventory or ad product to sell, the AI steps aside.

This splits the work for a travel business in a useful way. Advice, comparison, planning and trust queries get AI answers nine times out of ten, and that is where citations are won or lost. Bookable-inventory queries still run on classic SEO, Google Hotels and Flights presence, and business profiles, so the older playbook keeps its job while the new one gets added next to it. Treating GEO as a replacement for SEO would mean abandoning exactly the queries where money changes hands.

The AI picks winners, but not consistently

Asked whether Traveloka or tiket.com is cheaper, the AI refused to commit, explaining that base prices compete closely and the outcome depends on daily promos and timing, then it described each platform's strengths, Traveloka for bundles and price alerts, tiket.com for flash sales and points. Asked the same question about Agoda and Booking.com, it declared a winner outright, saying Agoda often has cheaper base rates for Indonesian hotels because of its Asia focus, and then noted both belong to Booking Holdings.

Elsewhere it named Citilink the best low-cost carrier and Garuda the most generous with baggage, so the hedging is selective rather than a policy. We could not find a structural reason in the data for when the AI commits and when it hedges, which is itself useful to know. A brand cannot rely on the AI staying neutral about it, and the sources feeding these verdicts were mostly UGC, the Agoda verdict rested on a personal travel blog, a points-hunting blog, a Reddit thread, a Lemon8 post and an Instagram reel. Verdicts about your brand are being assembled from content you do not control and probably have never read.

Trust queries feed a loop nobody is guarding

Queries containing terpercaya, amanah or tidak menipu produced the most consequential answers in the set. Asked for a trustworthy umroh agency for 2026, the AI named four specific companies, Dream Tour, Alhijaz Indowisata, Rabbani Tour and Arminareka Perdana, quoted a market price band of Rp 23 to 40 million, invoked official Kementerian Agama licensing as the trust criterion, and then cited the agencies' own marketing pages and an Instagram reel as the evidence. The regulator it name-dropped was not among the citations.

The pattern held even where government sources should dominate. On the query about checking an agency's official license, the AI correctly described the Kemenag SIMPU portal and the Satu Haji app, and cited siskopatuh.haji.go.id once, yet the primary citations were still two travel agencies' blog posts explaining the verification process, plus two Instagram posts. Agencies are winning visibility on a pure government-procedure query by publishing the explainer the government never optimized.

Nothing in the data suggests the winning operators engineered any of this. The loop is open, and it is not subtle. A structured page that answers the trust question directly, with the PPIU or PIHK license number, the "5 Pasti Umroh" checkpoints, verifiable claims and real reviews, is currently enough to become the AI's evidence for your own trustworthiness. In a category where the scam risk is real and searchers know it, owning that answer is worth more than any listicle placement.

Your industry does not gate the citation

Some of the most reliable citation winners in the data are not travel companies at all. Bank blogs from CIMB, OCBC and blu by BCA earned citations on budget and destination queries, Telkomsel and Axis articles surfaced on trip-planning answers, insurer content appeared well beyond insurance queries, and one Toyota dealer's East Java travel listicle dominated an entire nature-destinations answer. On the sharia-financing query, the lead citation was a BCA Syariah product page sitting above every travel agency in the answer.

The AI rewarded whichever page matched the question and never appeared to check what the domain sells. That should worry travel brands, because the competition for citations now includes every content team in the country, and a bank's destination guide can take the slot a tour operator assumed was theirs. It should also encourage them, since the same rule means a travel operator with one well-built page can beat domains far larger than itself inside the answer layer.

The pages that get lifted share a shape

The dominant answer looks the same across categories. A short intro paragraph that commits to a direct answer, then a labeled list where each item carries its own citation, closing with two or three follow-up questions about budget, dates or party size. Those closing questions appeared on nearly every answer we recorded, Google's AI Overview is behaving like the opening turn of a chat, not a static snippet, and it is fishing for the refinement that its AI Mode can answer next.

Itinerary queries produced the richest version of the shape. The Bali four-day query returned an hour-by-hour plan, arrival at 13.00, Pantai Melasti at 14.30 with the Rp 5.000 entrance fee attached, sunset at Jimbaran, through to a backpacker budget of Rp 2.5 to 3.5 million with itemized lines for the motor rental, hostel, meals and tickets. The lead citation, repeated through the answer, was tiket.com's itinerary content, a content-marketing article rather than a booking page, structured almost exactly like the answer built from it.

That mirroring is the practical lesson of the study. Pages built from labeled lists, day plans with real numbers, and claims that stand alone one sentence at a time map directly onto how these answers are assembled. Long unstructured prose can be correct and still lose the slot to a thinner page shaped like the answer, because the AI is not ranking essays, it is harvesting parts.

The AI quotes with confidence, including its mistakes

The recorded answers contain real errors. One quoted a garbled price of "$20 - 28 hari" in a hotel savings tip, one placed Belitung's airport under the wrong name, one confused the PPIU and PIHK license types that separate umroh operators from haji operators, and one repeated the incognito-mode pricing myth as a money-saving tip. An Australia visa answer cited an article about China visas, and a Japan halal-travel answer cited a Trip.com page about a mosque in India. One answer argued for travel insurance more strongly than its own cited source, which had concluded domestic coverage is often unnecessary.

None of those errors came with any drop in confidence, and a reader has no way to separate the quoted facts from the broken ones. The defensive move for brands is unglamorous. Keep prices, names and claims unambiguous and consistent on every page you own, including the Instagram captions, because the AI will quote them somewhere, and inconsistency is what turns your own content into the source of a wrong answer about you.

What to do with this, by seat

If you run a small operator, an open trip, an umroh agency or a rental fleet, the trust page is the highest-leverage move in this dataset. Publish the license number, the verification steps a customer can take, the real price bands and the named packages, structured as labeled sections, and keep the Instagram account information-dense because reels from operator accounts were cited across packages, umroh and transport. You do not need Traveloka's domain authority to be named in an answer, the operators being recommended today are your size.

If you run a platform or a hotel, the finding cuts differently. Your category pages already win citations on commercial queries, but the planning and advice layer is where tiket.com's itinerary content won the richest answer in the study, and that is a content investment, and the bookable queries where AI answers vanish still run on your Google Hotels feed, your flight data and classic rankings. The two systems need separate owners and separate scorecards, because a citation and a booking module are won with different work.

We check this for brands as part of our free AI Visibility Audit, including which queries in your category trigger AI answers, what shape those answers take, and who currently owns the citations. Talk to us if you want your category mapped, and if you want the raw dataset from this study, ask.