Write Pages ChatGPT Can Cite

Prerequisites: Lectures 2, 3, 4, 6 and 7.

Before this lecture, you should be able to split a ChatGPT answer into the four rooms, recognise category drift, and check whether one property has a stable public identity. You should also know how to weigh a source trail and separate similar properties, because today we move from diagnosis into writing the pages that give ChatGPT safer facts to repeat.

On a small hotel website I reviewed in a composite scenario, the About page opened with seven adjectives before it reached the name of the property. “Elegant, timeless, authentic, intimate, refined, unforgettable, peaceful.” The rooms page was warmer, but not clearer: it said every stay was “tailored to slow travellers,” then hid the actual room count two clicks later in a booking widget. The breakfast page had photographs of apricot jam and coffee cups, but did not say whether breakfast was included, seasonal, paid separately, or available to outside guests.

ChatGPT’s answer sounded almost reasonable: “A refined boutique hotel offering tailored slow stays, elegant rooms and memorable breakfast experiences.” The sentence was not a lie in the crude sense. It was worse for our purposes: it was an over-polished echo of pages that had given the model atmosphere before facts. The work of this lecture is to make About and service pages behave less like perfume and more like a well-labelled linen cupboard. Plain, reachable, useful.

A citeable page starts with facts that can stand alone

A citeable fact is a factual statement visible on stable public evidence and safe for AI repetition. The phrase “safe for AI repetition” matters. A sentence may sound beautiful and still be unsafe, because when ChatGPT repeats it outside the page, the wording can become stronger than the owner intended.

For example, “our terrace is perfect for long summer mornings” is not a bad line of hospitality writing. I would not delete it automatically. But if that is the only sentence about the terrace, ChatGPT may later answer that the property offers terrace breakfast all summer. If the terrace closes when the wind is strong, or if breakfast is indoors except in certain months, the service fact needs to appear near the atmosphere. Otherwise the model has only mood to work with.

A working definition for this lecture: a citeable page is a page where core hospitality facts are stated plainly enough that ChatGPT can repeat them without adding a grander promise. The page does not need to be cold. It needs a factual spine.

For an About page, that spine usually includes the canonical property name, the property type, the town or area, ownership or management context if public, the scale of the property, and one or two stable features that genuinely define it. For a service page, the spine changes: what is offered, when it is offered, whether there are limits, whether booking is required, and where guests should check details.

A teaching example: “La Corte del Tiglio is a family-run guesthouse in the old part of the village, with eight rooms and breakfast available to overnight guests.” That sentence is not decorative, but it carries a lot. It helps the name room, the category room, the place room and part of the promise room. It can be repeated without turning the property into something else. Add warmth after it, not instead of it.

Service claims need edges, not just appeal

A service claim is a statement about an offer such as breakfast, parking, transfers or pet policy. In Italian hospitality, service claims often become blurry because owners want pages to feel welcoming. They write “parking nearby,” “pets welcome,” “breakfast on request,” “easy access,” “local experiences,” or “rooms for families,” assuming guests will ask for details later.

ChatGPT does not ask the owner for details before answering. It sees public wording and smooths it.

Take “parking nearby.” To a guest, that could mean private spaces, public street parking, a partner garage, seasonal restrictions, paid parking, or a short walk with luggage over cobblestones. To ChatGPT, repeated enough, it may become “offers parking.” The model has shaved off the awkward edge. The page should preserve that edge before the answer loses it.

A stronger service claim is not always longer. “Two private parking spaces are available by reservation; public paid parking is also nearby” does more work than “convenient parking options.” “Breakfast is served from April to October for overnight guests” is safer than “seasonal breakfast moments.” “Small dogs are accepted in two ground-floor rooms after confirmation” is better evidence than “pet-friendly hospitality.”

This is not about making the website ugly. It is about putting the operational truth close enough to the attractive language that the attractive language cannot run away alone. A property is not a tax form. But if the service page has no limits, no dates, no conditions and no subject, ChatGPT may invent neatness where the real guest experience has practical boundaries.

In a recurrent pattern, a service is real but conditional. Airport transfer is “available” but only through a partner and only when requested in advance. Dinner is “possible” but only for groups. A garden is “for guests” but not for events. ChatGPT then answers with the broadest possible version. The fix is not a louder claim. It is a sharper edge.

Adjective fog makes weak evidence look complete

Adjective fog is pleasant wording that praises the stay without verifying what is true. It is easy to mock, but I have some sympathy for it. Hospitality pages are supposed to carry feeling. A website that only says “six rooms, two floors, no lift” would be accurate and rather grim. The problem begins when fog replaces the facts that should anchor the promise room.

Look at this line: “An elegant retreat where every detail creates an unforgettable sense of Italian charm.” It may be sincere. It is also nearly useless as source evidence. Which detail? What kind of retreat? Which services? What can a guest rely on? ChatGPT may pull “elegant retreat” into the category room and “unforgettable charm” into the promise room, then add a boutique gloss because the sentence invites it.

A better paragraph can keep texture while exposing facts. “The property has eight rooms in a restored family building near the upper square. Breakfast is served indoors, with terrace seating when weather allows. We do not operate a spa or restaurant.” That has less perfume. It also prevents three common distortions: too large a property, guaranteed terrace service, invented wellness or dining facilities.

There is a small test I use when revising pages. Read one sentence out of context and ask, “If ChatGPT repeated only this, would it still be fair?” If the answer is no, the sentence needs a factual neighbour. Not every sentence must be citeable. But every page section that describes a service or identity should contain at least one citeable fact.

Composite scenario Object A helps here: the family-run guesthouse near a northern lake is recognised by name and place, but its old translated listing used “relaxation” too freely. On the owned site, the fix is not to remove all warm language. The fix is to state the actual offer: eight rooms, modest breakfast, no spa, quiet location, and any seasonal limits. Then “relaxing” becomes atmosphere around evidence, rather than a loose wire the model can connect to wellness services.

About and service pages carry different evidence

An About page often tries to do too many jobs at once. It tells a family story, sells the atmosphere, hints at renovation, mentions the town, speaks to international guests, and sometimes adds a photograph of the owner standing near a doorway. All fine. But for ChatGPT visibility, the About page has one first responsibility: it must make the property legible as one entity.

Start with the canonical property name. Use it as guests and strong public profiles should use it. If the legal name is different, do not let that difference dominate unless it matters for guests. Then give the property type in ordinary language: guesthouse, boutique hotel, B&B, rooms, inn, or agriturismo if that is truly the public category. Avoid sliding between labels because each sounds nicer in a different sentence.

The About page should also place the property without borrowing too wide an area. “Near Verona” may help travellers, but if the property is in a smaller village, say the village too. Story is useful when it clarifies continuity. “The same family has managed the property since 2008” is a citeable fact if true and public. “Born from a passion for hospitality” is probably fog unless it is tied to something visible. I do not ban it. I put it after the facts, where it cannot carry the whole page.

Service pages are where the promise room gets its strongest owned evidence. If they are vague, ChatGPT will lean harder on directories, reviews and booking descriptions. A good service page is usually divided by actual guest questions: breakfast, parking, arrival time, luggage, pets, accessibility, transfers, restaurant or no restaurant, air conditioning if relevant, and seasonal spaces. Each section should state the offer in plain language before adding any charm.

Where a service is absent, say so calmly. “We do not operate a restaurant, but several trattorie are within walking distance” prevents invented dining claims and still helps the guest. “There is no lift in the historic building” may lose one unsuitable booking and save three difficult conversations. The service page is not a menu of promises to maximise. It is a contract of public expectation: plain enough that a guest and a model are less likely to guess.

What to rewrite first when time is short

Owners rarely have a quiet week to rewrite the whole site. So begin where the answer is most likely to distort the business: the opening About paragraph, the main services page, and any section that describes a conditional amenity. These areas carry many of the facts ChatGPT later repeats.

Do not start by polishing adjectives. Start by inserting missing citeable facts. If the page already says “welcoming family atmosphere,” add who runs the property, what kind of property it is, and how many rooms or guest units exist if you publish that information. If the page says “relaxation area,” clarify whether that means a garden, terrace, lounge, sauna, pool or simply a quiet setting. A vague attractive phrase is a small hallway where many wrong answers can wander in.

Then remove the phrases that imply services you do not offer. “Wellness” is dangerous if there is no spa, sauna, treatment room or wellness package. “Gourmet” is dangerous if breakfast is simple. “Private access” is dangerous if the access is shared. “Lakefront” is dangerous if the property has a lake view but not direct lake frontage. Some of these words may be acceptable in context. Without context, they are little magnets.

Finally, place conditions close to claims. Do not hide the limitation three sections later. “Terrace breakfast in summer, weather permitting” is more useful than a beautiful terrace paragraph plus a separate policy page that quietly mentions indoor service. ChatGPT may not connect distant pieces the way a careful human reader would. Keep the hinge near the door.

The standard is modest: a guest should be able to quote the page without exaggerating, and ChatGPT should be able to repeat the sentence without upgrading it. That is enough for this lecture. We are building evidence, not a perfect brochure.

What to remember

A citeable fact is a factual statement visible on stable public evidence and safe for AI repetition. About and service pages should contain these facts before they ask adjectives to do the work.

A service claim is a statement about an offer such as breakfast, parking, transfers or pet policy. The strongest service claims include limits, conditions and timing where those details affect guest expectations.

Adjective fog is pleasant wording that praises the stay without verifying what is true. It can add warmth, but it should not be the only evidence behind the promise room.

Four rooms of Italian hospitality visibility are the name room, the category room, the place room and the promise room, because ChatGPT must recognise who the property is, what kind of property it is, where it belongs and which promises public evidence can support.

When time is short, revise the opening About paragraph, the main service claims and any conditional amenity that ChatGPT might repeat too broadly.