The new search isn't a list. It's a recommendation — with footnotes.
That citation mechanic is the whole game now. For twenty years, search meant ranking: be position one through ten on a results page and hope for the click.
The new search isn't a list. It's a recommendation — with footnotes.
Try this today: ask an AI search engine a real trip question — "best restaurants in Kampot" or "is Kampot worth visiting in the rainy season." You won't get ten blue links. You'll get a short, confident, conversational answer, and under it a row of small numbered citations, like footnotes in an academic paper. The engine read a spread of websites, picked the ones it trusted, and wrote you an answer.
That citation mechanic is the whole game now. For twenty years, search meant ranking: be position one through ten on a results page and hope for the click. Generative search means being quoted. The sites in the footnotes are the ones whose content got folded into the answer, as if a person had read them and aggregated the best bits. If you're in the footnotes, you're part of the recommendation. If you're not, you don't exist at the moment of decision — because a growing share of travelers never scroll past the answer at all.
This is a completely different contest from the boolean search we've run since the 2000s. Ranking was about technical signals you could buy your way into. Being cited is about being the best answer to a specific question — which changes who wins, in ways that mostly favor small operators.
What the footnotes actually reward
We ran a structured audit of Kampot, one of Cambodia's rising secondary destinations, and published it this month: ten real traveler questions, asked the way AI search asks them. The results tell you exactly what the new game pays.
Across those ten questions — worth roughly 43,600 AI searches a month combined — the AI answers drew 71 citations from 37 domains. Three things stood out.
First, YouTube was the most-cited source, 13 citations, ahead of Wikipedia's 6. AI answers about Kampot lean heavily on video. If your experience exists only as a brochure paragraph, the engine has less to work with.
Second, independent travel blogs beat the industry. A single personal blog was the top-cited source for two of the highest-volume questions in the set ("Kampot to Kep day trip," "how to get from Phnom Penh to Kampot," each around 4,400 AI searches a month). The big OTAs were present but thin — one or two citations each, mostly on hotels.
Third, the biggest single prize in the set, "best restaurants in Kampot," carries about 22,200 AI searches a month — five times any other question — and the top citation went to a regional cruise-operator content page. Not a Kampot restaurant. Not a local food guide. The most valuable answer in the destination is being sourced from outside it, because nobody local had written something the engine considered a better answer.
Most Kampot businesses were absent from the footnotes entirely.
The freshness trap (and a warning from this week)
Generative engines favor current, specific content — and they punish staleness quietly. This week gave us a clean example from Ethiopia: flights to Lalibela, Axum, Mekelle and Shire were suspended on 23–24 September after fighting resumed in the north. Yet our monitoring found tour operators still ranking and still selling Lalibela itineraries with no notice that the airport is closed. Their content is good. It's just a month old, and in AI search a month can make the difference between an accurate recommendation and a misleading one.
The lesson isn't only "keep content fresh." It's that the answer engine's snapshot of your destination is now a live asset — someone's trip decision is being formed from it today, whether you maintain it or not.
The part I like: smaller destinations have their control back
The Google era rewarded scale. Domain authority, backlink portfolios, agency retainers — a national tourism board with a big contract could bury a provincial operator with better knowledge. Ten blue links is a rich man's game.
Citations don't work that way. The engine doesn't care that a one-person blog has no domain authority to speak of; it cares that the answer to "how do you get from Phnom Penh to Kampot" is genuinely the best one available. That single blog out-cited the OTAs on exactly those questions. A Kampot guesthouse that publishes an honest, detailed, current answer to "best restaurants in Kampot" — the 22,200-searches-a-month question — is competing head-to-head with whoever holds that citation now, and the incumbent is a cruise operator two provinces away.
Generative search re-prices the one asset small destinations always had in abundance — deep, specific, current knowledge of their own place — and de-prices the thing only money could buy. For a small DMO or an independent operator, that's not a threat. That's the best structural news in twenty years.
What marketing should actually do
1. Answer questions, don't describe attractions. Build pages around the exact phrasing travelers use — how to get there, is it worth it in low season, what does it cost, is it good with kids. Specific beats poetic.
2. Go on record with facts and dates. Prices, schedules, seasons, opening hours. Vague content can't win a citation, and stale content can lose one (see Ethiopia).
3. Video matters more than you think. YouTube was the top source for Kampot. Ten honest clips of your experience will do more for AI visibility than another polished landing page.
4. Fix the fundamentals once, then keep them current. Google Business Profile, TripAdvisor, OTA listings, consistent naming. The engines cross-check.
5. Audit monthly and track the movement. We run the Kampot audit monthly now; the citation set shifts as content changes, and tracking that shift is what turns AI visibility into a managed channel instead of a lucky break.
6. Claim the question nobody's answering. Find the high-volume question where the current citation is weak or outsourced — that's the cheapest win in destination marketing right now.
FAQ
What is generative search, in one line?
Search that returns a written answer assembled from multiple sources, with numbered citations, instead of a list of links.
Do I still need traditional SEO?
Yes — the fundamentals (structured data, fast pages, consistent listings) still feed what the engines read. But the objective changes: not position 3, but quoted in the answer.
We're a tiny operator. Can we realistically get cited?
That's the point of the Kampot data: a personal blog out-cited the OTAs on high-volume questions. Specific, current, genuinely useful content is what earns the footnote.
How do we find out if the AI answers about our destination mention us?
Run a structured AI-visibility audit — the traveler's real questions, asked the way AI search does, then score who gets cited. We do this for destinations and individual operators; the Kampot audit shows the method and the output.