Home / CITED
The framework
CITED
Five components. Almost every delivery hour sits under one of them; measurement and reporting sit alongside the five, because you cannot improve what you are not tracking. That is what makes the work assignable, estimable and auditable rather than improvised.
The five components
Crawlability
Whether AI systems can read your site at all. This determines whether anything else matters — a property with perfect listings, excellent reviews and strong press coverage is still invisible if its own site cannot be read.
The most common failure is not a bad website. It is a content delivery network with an AI-scraper block enabled by default, which nobody chose deliberately and which overrides the site's own settings. The property ranks normally on Google and is invisible to every AI engine.
What we do: audit crawler permissions at both the robots and CDN layer, test every page template with JavaScript disabled, identify amenity lists and FAQs trapped inside tabs or accordions where they never reach the HTML, and issue the exact configuration your web agency needs to apply.
A hotel whose amenities live inside a collapsed accordion appears, to a machine, to have no amenities.
Identity
Whether machines can resolve what your property actually is. Before a model can recommend a hotel it must recognise it as a distinct entity with attributes — rooms, amenities, exact location, what it is near, which group it belongs to.
This is the least contested territory in the whole framework, because at almost every hotel group it is owned by nobody. Structured data is deployed once at site launch and never revisited. Entity records do not exist. Name and address vary subtly across forty platforms.
What we do: author complete structured data per property, create and correct machine-readable entity records, claim the Knowledge Panel, and reconcile name, address, phone, coordinates and category to be byte-identical across every surface that carries them.
"The Grand Hotel Edinburgh" and "Grand Hotel, Edinburgh" are two entities to a machine. Everything you have built splits between them.
Trust
Whether independent sources confirm the claim. Roughly 84% of AI citations come from earned media rather than a brand's own website. Muck Rack Generative Pulse, May 2026 — 25m+ citations analysed across ChatGPT, Claude and Gemini. A model treats your marketing copy the way an experienced traveller does — as a source with an obvious interest in the outcome.
What it cites instead are places that appear neutral: review platforms, tourism boards, city guides, curated collections.
What we do: claim and complete the review platforms nobody manages, audit listing accuracy across collections you already belong to, and supply a response framework that turns your existing reviews into citable content.
The technique worth understanding
"Thank you for your kind words, we hope to welcome you back" contains no extractable information. "Delighted the rooftop pool and the 4am airport transfer worked for your early flight" restates two specific, citable attributes in the property's own voice on a heavily-cited surface — at no media cost. Same effort, entirely different output.
Expression
Whether your content can actually be extracted. This is where the interests of machines and guests align completely, which makes it the easiest component to get agreement on internally.
"Superbly located in the heart of the city" is unusable to a model. There is no fact in it — and it does not help a traveller either, because it does not answer whether the hotel is walkable to where they are going. "Twelve minutes' walk from Waverley Station" can be quoted, and is more useful to a human.
What we do: specify question-shaped headings that match how guests ask, answer-first structure because models extract opening sentences preferentially, and replacement of unquantified superlatives with checkable facts. Your writers execute it in your brand voice — we never write in it for you.
The test: could a stranger extract three concrete facts from this page in ten seconds? If not, it is failing humans and machines equally.
Distribution
Whether you exist in the places models look. Map platforms, national tourism boards, city destination organisations, directories, forums and data aggregators.
Much of this is free, permanent and treated by models as authoritative — a national tourism board is a credible source on where to stay in that country. And at almost every group we assess, most of it is unclaimed.
What we do: claim, complete and maintain each surface per property, reconcile the data aggregators that silently feed dozens of smaller directories, and monitor the forums where travel recommendations are actually discussed — always with disclosed identity, never anonymously.
For a fifteen-property group this is roughly forty-five high-authority records that can exist within a fortnight, at no media cost.
Sequence
Why the sequence matters
The five components are not equal in urgency, and they are not worked at the same time. Worked in the wrong order, most of the effort is wasted — and the waste stays invisible until the next measurement round, by which point a quarter has gone.
First — Crawlability
A blocked crawler makes everything after it pointless. If GPTBot, ClaudeBot, PerplexityBot and CCBot cannot reach your property pages, no amount of work on the other four components changes a single answer.
It is also the fastest to fix: a robots.txt directive and a render check, usually inside a fortnight, usually with no budget attached.
Second — Identity and Distribution
These move next because they need nobody’s permission. No vendor coordination, no media spend, no sign-off from a brand team. We can complete a Wikidata record or claim a tourism-board listing without waiting on anyone.
That is why the first measurable movement usually appears here, before the first quarterly review rather than after it.
Third — Trust and Expression
These come last because they involve other people. Your operations team writing review responses. Your agency implementing content specifications. Your PR firm pitching the publications models actually cite.
The work is not harder. It is slower, because it moves at the speed of other people’s calendars — so starting it first simply stalls the whole programme.
The order is not a preference. It is a dependency chain: each stage is worth doing only once the stage before it holds. And measurement sits outside the chain because it has to run from day one — without a baseline taken before anything is changed, no later reading means anything.
Effort
Where the effort actually goes
Hours and impact are not the same thing. The chart below is where the time goes on a typical multi-property engagement — it is deliberately unflattering, because most of it is unglamorous record-keeping rather than strategy.
| Component | Share of hours | Who has to be involved |
|---|---|---|
| Distribution — listings, maps, tourism boards | ~40% | Us alone |
| Measurement and reporting | ~20% | Us alone |
| Trust — reviews and reputation | ~15% | Your operations team |
| Crawlability — technical specification | ~15% | Whoever holds your web build |
| Identity — entity records | ~5% | Us alone |
| Expression — content specification | ~5% | Your writers, in your voice |
Distribution at 40%
Forty per cent of the hours go into claiming, completing and maintaining records across map platforms, tourism boards, destination organisations and the data aggregators that quietly feed dozens of smaller directories. It is per-property, per-surface work that does not compress — which is the reason a multi-property group needs a dedicated team on it rather than a generalist retainer with hospitality on the side.
Identity at 5%
Five per cent of the hours, and among the highest impact per hour in the framework. A group operating under several brand names often reads to a model as several unrelated businesses; reconciling that into one entity with consistent records is a small, finite piece of work with a disproportionate effect on whether the group is named at all.
Measurement at 20%
A fifth of the hours are spent on the thing we sell rather than the thing we change: running the same query set every month, recording what was said and which sources were cited, and reporting the movement honestly — including the months where nothing moved. It is the largest single line that produces no visible output for the client except the truth.
These proportions are indicative of our own delivery model on a multi-property engagement. They are not an industry benchmark, and we have no basis for claiming they hold for anyone else.