Will AI search reduce our dependence on marketplaces and aggregators?

Do AI answers still send buyers through marketplaces and aggregators?

Mostly, yes: for booking-style questions, intermediaries still supply most of the sources AI cites.

The clearest evidence comes from hotels, an industry built on paying intermediaries for demand. Commission rates commonly run from 15% to 25% of booking value. For independent hotels, bookings from online travel agencies can exceed 61% of room revenue.

Zhu and Chang (opens in a new tab), researchers at Blossom AI, a company, put 156 Tokyo hotel questions to Gemini 2.5 Flash with Google Search in March 2026, in English and Japanese. Online travel agencies such as Booking.com, Expedia and Jalan supplied 55.3% of all citations. For booking-style questions such as “cheap hotel in Shinjuku”, they supplied 69.2%. The authors’ own verdict: disintermediation is not imminent.

Where can a brand’s own content compete?

In experience-led questions, where the intermediaries’ share fell to 44.1%.

The study paired each booking-style question with an experience version, such as “good value hotel with local charm in Shinjuku”. That one change in question framing cut the agencies’ share of citations by 25.1 points. The gap held in every category tested:

Budget questions were the agencies’ stronghold, because price comparison is what they do best. Business travel was the most open, thanks to content about workspaces and quiet rooms that listings rarely carry. In Japanese, 62.1% of citations for experience-led questions came from sources other than agencies.

Most of that freed-up share went to blogs, editorial sites and other third parties, not to hotels. Hotels’ own sites made up a steady 19–24% of the non-agency citations.

Do shopping answers lean on retailers and marketplaces too?

It depends on the engine: Google’s AI answers draw on retailers, while chat assistants lean on review sites.

Uberti-Bona Marin and colleagues (opens in a new tab) put 117 real product questions to ChatGPT, Gemini and Google’s AI Overviews, the AI summaries at the top of Google results, from the Netherlands in September 2026. In AI Overviews, retailers and marketplaces made up 17.8% of the source websites shown, and manufacturers and brands 16.1%. ChatGPT and Gemini leaned mainly on editorial and product-review sites. The authors treat these source categories as exploratory.

Our own AI Overview study of 1,248 US Google searches found marketplaces and retailers were a small share of citations overall but 7.9% on transactional searches. On those searches, Google’s own pages took 37.9% of citations, the largest single source type. For buying questions, Google itself is becoming one of the intermediaries.

How often do AI answers cite a brand’s own website?

It varies by engine, from under a fifth of answers to nearly all of them.

In our study of 79 brands asked “Is this brand legit?” on four engines in September 2026, Perplexity cited the brand’s own site in 94.9% of answers and Google AI Mode in 17.7%. Brand-owned pages are rarely the main evidence for a recommendation, though. Chen and colleagues (opens in a new tab) found that for US consumer electronics questions, a search-enabled GPT drew 92.1% of its sources from independent media and review sites.

Assistants also go looking for directories. In our hidden-searches study, ChatGPT ran a search aimed at a platform or directory, such as Yelp, Avvo or G2, in 31.2% of its answers to buyer questions. Leaving those platforms is not an option yet.

What makes a brand’s own site get cited instead of the middleman?

Depth: in one small audit, hotels with deep, question-answering pages were cited directly, and shallow ones were not.

Zhu and Chang scored 14 hotel websites on content depth: FAQ, area guide, blog, access information and unique content. Hotels Gemini cited directly averaged 8.6 out of 15; hotels it did not cite averaged 3.4. Every hotel scoring 6 or more was cited; every hotel below 6 was not.

The contrast between two hotels makes the point. Kadoya Hotel, an independent with no technical search work, was cited through a 33-question FAQ and a 13-attraction sightseeing guide. Hotel K5, with a 9.6 rating on Booking.com and full technical setup, had only brief pages. Gemini mentioned K5 but drew its facts from Expedia, Hotels.com and an editorial site. The hotel was found, but through intermediaries.

With 14 hotels, this is an association, not proof. Better-known hotels may both write more and get cited more. Other signals, such as guest rating and price, also shape which hotel an assistant recommends.

What should you do about it?

Keep the marketplace channel healthy while building the content that competes for experience-led questions.

1. Keep marketplace and directory listings complete. They still supply most citations for price and booking questions.
2. Build deep answers on your own site. Long FAQ pages, area or use-case guides and specific details beat brief pages ticking every box.
3. Target the questions intermediaries answer badly. Experience, fit and specialist needs were the most open in the research.
4. Earn coverage in editorial and review sources. They took much of the share the intermediaries lost.
5. Measure share of citations by question type and engine. Track how often answers cite you, an intermediary or a third party.
6. Do not cut marketplace spend on citation data alone. No study yet shows AI citations turning into direct bookings.

To set up that tracking, see our generative engine optimization service.

What does the research not tell us yet?

The research shows who is cited when a buyer is deciding, not who captures the sale.

• The intermediary evidence covers one engine, one city and one month, and comes from researchers at a company.
• Citation is not booking. A hotel cited through Booking.com may still be booked directly, and the reverse.
• The content-depth finding rests on 14 hotels and cannot rule out that famous hotels simply write more.
• The retail source categories were sorted by an AI model and checked only by spot inspection.
• No study here covers Amazon or other product marketplaces from the seller’s side, or tracks changes over time.

Frequently asked questions

Will ChatGPT replace Booking.com and Expedia for hotel discovery?

Not on current evidence. In a Gemini audit of Tokyo hotel questions, online travel agencies still supplied 55.3% of cited sources, and the authors concluded disintermediation is not imminent.

Can AI search help a hotel get more direct bookings?

It may help with discovery, but bookings have not been measured. Hotels with deep FAQ and area-guide pages were cited directly by Gemini, while the study measured citations, not reservations.

Do AI shopping answers link to retailers or to brand sites?

Both, depending on the engine. Google’s AI Overviews drew 17.8% of source websites from retailers and marketplaces and 16.1% from manufacturers and brands in one audit.

What kind of content helps a brand’s own site get cited by AI?

Deep, specific answers to buyer questions. In one hotel audit, every site scoring 6 or more out of 15 for content depth was cited directly, and every site below 6 was not.

Sources

• Zhu and Chang (2026), The End of Rented Discovery: How AI Search Redistributes Power Between Hotels and Intermediaries (opens in a new tab), arXiv:2603.20062.
• Uberti-Bona Marin et al. (2026), "If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations (opens in a new tab), arXiv:2609.18729.
• Chen, Wang, Chen and Koudas (2025), Generative Engine Optimization: How to Dominate AI Search (opens in a new tab), arXiv:2509.08919.
• Underneath (2026), When does Google show an AI Overview? 1,248 US searches
• Underneath (2026), “Is this brand legit?” How AI assistants build a reputation
• Underneath (2026), The hidden searches AI assistants run before they answer