Buyer questions for Airbnb, drafted by four Claude models

CiteLadder's Prompt Builder drafts the buyer questions a brand tracks in AI answers. For this study we ran the same step for Airbnb on Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5 and Claude Fable 5.1. Each model ran at low, medium and high effort, twice each. The instructions, the business profile and the 12 planned question slots stayed the same. Only the model and the effort changed. The tables below this text list every question each model wrote. This page describes what can be observed in them. Judging their quality is left to the reader.

What Airbnb sells, according to the business profile

The business profile describes Airbnb as a vacation rental marketplace with consumer buyers. Its category terms include vacation rentals, short-term rentals, holiday homes, monthly rentals, travel experiences, cabins and beach houses, and in-home services booking.

The profile lists four offerings. Each one became a topic:

• Vacation home rentals: entire homes, cabins, beach houses, apartments and unique stays for short trips.
• Monthly furnished rentals: furnished homes and apartments for stays of a month or longer, with long-stay pricing.
• Tours and local experiences: activities, tours and classes led by local hosts.
• Bookable in-stay services: private chefs, massages, photography and personal training at or near the rental.

Each offering also carries attributes, situations and audiences. The slot plan draws on these.

How the 12 slots were planned

Each topic received three slots, one per buyer stage. Each stage was tied to one facet of the offering:

• Awareness slots use a situation. Allowed intents: learn, solve, recommend.
• Consideration slots use an audience. Allowed intents: solve, recommend, compare, validate.
• Decision slots use an attribute. Allowed intents: compare, validate, buy, recommend.

The facets chosen per slot were:

• Vacation home rentals: family vacation (awareness), families (consideration), entire place to yourself (decision).
• Monthly furnished rentals: relocating to a new place (awareness), digital nomads (consideration), dedicated workspace (decision).
• Tours and local experiences: special occasion celebration (awareness), couples (consideration), activity duration (decision).
• Bookable in-stay services: capturing trip photos (awareness), celebration planners (consideration), available time slots (decision).

All 12 slots belong to the core cohort. Every model received this same plan.

What differs between the models' questions

Length. The code measured a median of 10 words for Claude Haiku 5.5, 11 for Claude Sonnet 5.5, 11 for Claude Opus 5.5 and 12 for Claude Fable 5.1. No model wrote a question over 16 words.

Question marks. Claude Fable 5.1 ended 81.9% of its questions with a question mark. The other three models sit between 58.3% and 62.5%. Claude Haiku 5.5 often wrote requests in statement form, and one run ended them with a full stop, as in "We're a family of four needing a holiday home to book."

Openings. Claude Haiku 5.5 rotates between openings such as Looking for, Where can I, Recommend, Compare, Any and Who offers. Claude Opus 5.5 repeats openings across slots and runs. For the digital nomad slot it opens with the same self-description in all six runs, as in "I'm a digital nomad, what platforms have good monthly furnished apartments?"

First-person framing. All four models write some questions as a buyer describing themselves. Claude Sonnet 5.5 often starts with a situation in the first person, such as travelling with young kids or planning a birthday weekend. Claude Opus 5.5 uses a family of five for the families slot in most runs. Claude Fable 5.1 adds people and occasions from the buyer's life, as in "Any local experiences I can book to celebrate my mum's 60th birthday?"

Detail. Some questions add details that are not in the slot. Claude Fable 5.1 adds items such as a pool, a kitchen or toddlers to the families slot. Claude Opus 5.5 turns activity duration into two-hour tours in every run. Claude Haiku 5.5 and Claude Sonnet 5.5 mostly use half-day tours for the same slot. No model named a city or region.

Low, medium and high effort, run 1 and run 2

Effort did not produce one pattern shared by all four models. Within each model, the framing of a slot tends to hold across effort levels while the wording shifts.

Claude Opus 5.5 keeps several questions almost fixed. For the celebration planners slot, every run at every effort opens with the same anniversary framing, as in "Planning an anniversary getaway, where can I hire a private chef for the rental?" The exact wording "Apps to book an in-home massage with same-day availability" appears at low, medium and high effort. "Best sites to find a vacation home for a family trip" appears in both low effort runs and in medium effort run 1.

Claude Sonnet 5.5 also repeats itself between runs. "Any good local experiences to book for an anniversary celebration?" appears word for word in both runs at low effort and both runs at medium effort. At high effort the same slot shifts to a memorable anniversary and a milestone birthday.

Claude Haiku 5.5 shows the most change in wording between run 1 and run 2 of the same setting. Slots keep their facet, but the openings and sentence forms change.

Claude Fable 5.1 repeats a few questions between runs at low effort, for example on the family vacation and entire-place slots. At medium and high effort its two runs share fewer exact questions.

What the code measured

For each model the code counted 6 outputs and 72 prompts. It measured:

• Median words: 10 for Claude Haiku 5.5, 11 for Claude Sonnet 5.5, 11 for Claude Opus 5.5, 12 for Claude Fable 5.1.
• Over 16 words: 0% for every model.
• Names a studied brand: 0% for every model.
• Names a place: 0% for every model.
• Joins two asks with or: 0% for three models and 2.8% for Claude Opus 5.5, for example "Recommend romantic tours or classes for couples on vacation".
• Ends with a question mark: 58.3% for Claude Haiku 5.5, 59.7% for Claude Sonnet 5.5, 62.5% for Claude Opus 5.5, 81.9% for Claude Fable 5.1.
• Near-duplicate pairs: 0% for every model. The repeats described above occur between separate runs of the same slot.
• Distinct openers: 52.8% for Claude Haiku 5.5, 38.9% for Claude Sonnet 5.5, 31.9% for Claude Opus 5.5, 43.1% for Claude Fable 5.1.
• Mean output tokens: 3278 for Claude Haiku 5.5, 800 for Claude Sonnet 5.5, 799 for Claude Opus 5.5, 1291 for Claude Fable 5.1.

These figures describe form. They do not score quality.

Every question, side by side

Run 1 of each setting. Each row is one planned slot; the model and the effort are the only things that change across a row and down a stage.

Vacation home rentals

Monthly furnished rentals

Tours and local experiences

Bookable in-stay services

What the code measured for Airbnb

All six outputs per model (three efforts, two runs), 72 questions each.

Frequently asked questions

See how AI answers your buyers

CiteLadder runs buyer questions like these for your own brand across ChatGPT, Gemini, Claude and Google AI Overviews. It shows which pages those answers cite and gives you a ranked list of what to change on your site.

Explore Platform overview