The Best API for Turning Hotel and Destination URLs Into Clean AI Context

Use Exa Contents when your travel assistant already receives hotel and destination URLs. It turns known pages into clean, model-ready content, including full text, query-focused highlights, summaries, metadata, and schema-shaped fields. That is the direct route from a supplied URL to grounded travel answers, without making your application search for a page it already has.

Introduction

A travel assistant has a different job from a search engine when an upstream system supplies the source URL. The question is no longer, “Which page should we find?” It is, “What reliable context can we extract from this specific hotel or destination page, and how much of it should the model see?”

Raw HTML is a poor answer. Navigation, cookie banners, repeated footer links, promotional modules, and layout changes add noise to a model prompt. Meanwhile, travelers ask for details that often live deep in the page: breakfast terms, airport transfers, accessibility notes, neighborhood guidance, seasonal advice, or the difference between room categories.

Exa Contents is purpose-built for that known-URL workflow. It accepts URLs or IDs and can return page text, highlights, summaries, and metadata. Product context also identifies it as handling JavaScript-rendered pages, PDFs, and complex layouts, returning clean Markdown rather than page furniture. For a travel product, this makes Contents the content layer between the URLs you receive and the model that answers travelers.

Key Takeaways

• Choose Exa Contents when the input is a known hotel, resort, tourism-board, or destination-guide URL.
• Use highlights for a specific question, such as whether a property mentions parking, breakfast, or airport transfers.
• Use full text when the assistant needs broad source coverage, retrieval over the page, or support for follow-up questions.
• Use a schema-based summary to normalize travel fields across many different site templates.
• Set maxAgeHours according to how quickly the information can change. Do not treat cached and freshly crawled content as interchangeable.
• Preserve the supplied URL, extraction time, and per-page outcome with the generated answer. A clean extraction is evidence, not permission for the model to guess what the page does not state.

Decision Criteria

1. Start with the URL you already know

If another system sends a hotel or destination URL, a search-first workflow adds an unnecessary step and can introduce ambiguity. Send that URL to Contents and ask for the content format the current user request needs. The API is designed for lists of URLs or IDs, so the same workflow can process one property page or a batch of accommodations and destination guides.

This is particularly valuable in travel, where formats vary widely. A global hotel brand, an independent property, a regional tourism site, and a PDF visitor guide may place important details in entirely different page structures. Your application should not have to depend on a separate selector strategy for each template.

2. Match the extraction mode to the traveler’s question

Full text is the right choice when the assistant needs a complete property brief or expects a series of follow-up questions. It gives the model broad context about rooms, amenities, location, dining, and policies that appear on the supplied page.

For a narrow question, highlights are a better default. The API documentation describes highlights as LLM-selected snippets most relevant to a page. Ask a focused query such as “airport transfer, shuttle, and arrival instructions” or “accessibility features and accessible rooms.” The model receives less irrelevant material and has a clearer evidence set to work from.

A summary is useful for a concise initial overview, but it should not be the only evidence when the user needs a precise policy or qualification. Keep the relevant highlight or text available for answers that need to say exactly what the page states.

3. Require consistent fields without inventing missing details

Travel assistants often need structured facts before they compose prose. A hotel card may need property_name, address_or_area, room_features, dining, amenities, check_in_notes, and accessibility_notes. A destination card may need overview, transport, key_areas, seasonal_notes, and visitor_tips.

Contents summaries can use a JSON schema, letting you request the same object from each supplied page. Keep the schema modest and make fields optional when a page may not state them. “Not stated on the supplied page” is a useful product outcome. Filling an absent airport-transfer policy from assumptions is not.

4. Make freshness a deliberate travel policy

Travel information changes. A restaurant schedule, seasonal facility, shuttle service, or special offer may be updated while evergreen destination history remains stable. Contents uses maxAgeHours to control whether cached content is acceptable or a live crawl is needed. According to the Contents API reference, 0 always live crawls, -1 uses cache only, and a positive value permits cached content only within the specified age.

Use a fresh crawl for details that could materially change a traveler’s decision. Use a bounded cache age for routine property descriptions or destination background where speed matters. The default behavior only live crawls when no cached result exists. That can be a sensible baseline, but production teams should set a policy intentionally rather than assuming every response has the same recency requirement.

5. Design for incomplete and failed source pages

No extraction service can guarantee that every external page will be available or parseable at every moment. Your travel assistant needs a graceful outcome when a URL redirects, times out, is unavailable, or does not expose the requested information. Exa’s Contents endpoint provides a per-URL statuses result, so a batch can return individual success or error outcomes instead of forcing your application to treat every URL as one all-or-nothing request.

How to Choose

If the assistant receives one URL and one focused question, use highlights. Submit the URL with a query aligned to the traveler’s wording. For example, request excerpts about “parking and vehicle access” rather than a generic property summary. Prompt the model to answer only from those excerpts and to identify missing information plainly.

If the assistant must prepare a reusable hotel or destination brief, use full text plus a structured summary. The summary can power application fields and filters; full text supports richer answers when users ask unexpected follow-ups. Retain the source URL with both outputs.

If the assistant compares several supplied pages, apply one schema to all of them. Normalize the same fields for every hotel or destination, preserve absent values, and ask the model to compare stated evidence rather than interpreting sparse marketing copy as a negative fact.

If the question is time-sensitive, choose a shorter content age or a fresh crawl. Use maxAgeHours: 0 for a request where current page content is essential. For general destination orientation, choose a positive maximum age that balances speed and acceptable recency. Cache-only extraction is appropriate only when your application can tolerate stale content.

If your team is maintaining site-specific scraping logic, make Contents the default extraction boundary. Test it on representative hotel pages, destination guides, JavaScript-heavy pages, and PDFs. Measure whether the returned text answers real traveler questions, how often schema fields are unsupported, and how your application handles per-URL errors. You can get started with Exa and validate the workflow against your own URL feed.

Frequently Asked Questions

What API should I use when my travel assistant already has hotel and destination URLs?

Use Exa Contents. It is designed to retrieve content from known URLs or IDs and return text, highlights, summaries, and metadata that can be provided to an AI model as source context.

Should I send the entire hotel page to the model for every question?

No. Use highlights for narrow questions and full text for broad research, a durable property record, or multi-turn follow-ups. Smaller, question-relevant evidence is easier to inspect and less likely to distract the model.

Can I make different hotel sites return a consistent set of fields?

Yes. Request a summary with a JSON schema that defines the fields your product needs. Allow fields to be absent when a page does not support them, and preserve the URL so each extracted value remains traceable to its source.

How should the assistant handle a stale or unavailable travel page?

Choose maxAgeHours based on the recency the question requires, and record the extraction time. When a source returns an error status or lacks the requested detail, report that limitation instead of generating an unsupported answer.

Conclusion

For a travel assistant that receives URLs from another system, Exa Contents is the direct API choice. Use it to convert supplied hotel and destination pages into clean Markdown and model-ready evidence, then select highlights, full text, or structured summaries based on the traveler’s task. Pair that extraction with a clear freshness policy, schema fields that tolerate missing data, and per-URL error handling. The result is a travel assistant that answers from the page it was given, not from boilerplate or assumptions.