Tripadvisor Review Extraction

Purpose

Given a Tripadvisor entity (restaurant, hotel, or attraction) — accepted as a full tripadvisor.com URL, a d{locationId} token, or a free-form "name, city" reference — return the entity's overall rating, total review count, 1-5 rating distribution, ranking position, address, contact details, and a filterable, paginated slice of structured reviews. Honors Tripadvisor's full read-side filter surface (rating bubbles, sort order, traveler type, language, season, search-within-reviews, subcategory ratings for hotels, meal type for restaurants). Read-only — never clicks Write-a-Review, Save, Share, Sign-In, or any mutation control.

When to Use

• Pulling structured review data for a competitor or partner restaurant / hotel / attraction.
• Sentiment / reputation monitoring on a small set of POIs.
• Building a snapshot of how an entity is rated across trip types or seasons.
• Any agent flow that needs Tripadvisor review data without booking, posting, or otherwise mutating state.

Do not use this skill for bulk crawling across thousands of entities — Tripadvisor's anti-bot wall makes that economically nonsensical from a residential-proxy budget standpoint. For bulk needs, route the user to the official Tripadvisor Content API (see "Alternative methods" below).

Workflow

Tripadvisor is one of the most aggressively-defended consumer travel sites on the public web. Browserbase verified-Verified + residential proxies are mandatory — every direct-HTTP path (raw curl, Browserbase Fetch API, headless Chromium without fingerprint hardening) returns a Datadome captcha challenge (HTTP 403, ~778 bytes, captcha-delivery.com redirect script). Confirmed across /Restaurant_Review-*, /Hotel_Review-*, /Attraction_Review-*, /Search, and /TypeAheadJson paths via Browserbase Fetch with residential proxies (2026-05-18). The internal GraphQL endpoint at /data/graphql/ids is alive (returns 405 Method Not Allowed on GET, confirming POST-only) but its requests are signed by session cookies set by the same Datadome flow that gates the HTML pages — out-of-band replay is a confirmed dead-end (see Site-Specific Gotchas).

The reliable path is therefore: load the entity's /Restaurant_Review / /Hotel_Review / /Attraction_Review page in a Verified+proxied session, apply filters by rewriting the URL with Tripadvisor's documented filter tokens, and paginate using the or=N page-offset token. Extract review cards from the rendered DOM.

1. Verified + residential-proxy session

SID=$(browse cloud sessions create --keep-alive --verified --proxies | jq -r '.id')
export BROWSE_SESSION="$SID"

Both --verified and --proxies are mandatory. A bare session (or even --proxies without --verified) hits Datadome on the first request and stays walled for the session lifetime. Use a US residential proxy for .com entities; use the locale-matched Tripadvisor TLD (.co.uk, .fr, .de, .it, …) for entities in those countries and pair with a region-matched proxy.

2. Resolve the entity to a canonical Review URL

Tripadvisor entities are addressable by g{geoId}-d{locationId} pairs. The canonical Review URL shape is:

Resolution from input shape:

• Full URL supplied — use as-is. Strip any existing filter / pagination tokens (between Reviews- and the entity slug) before rebuilding with the filters in step 3.
• d{locationId} token only — open https://www.tripadvisor.com/{locationId} (Tripadvisor's bare-id short URL redirects to the canonical entity page; the resolved URL surfaces g{geoId} and the entity type).
• Name + city — drive the in-page search box (/Search?q=<urlenc-name+city>) and pick the top match whose type matches the user's hint. Do not call /TypeAheadJson?action=API directly — it is Datadome-walled even though it returns JSON (confirmed 403 with proxies).
• Ambiguous name (e.g. "Ritz" in London) — return success: false, reason: "ambiguous_name", candidates: [...] with the top 3-5 search hits and let the caller disambiguate.

3. Apply filters by rewriting the URL

Tripadvisor encodes most filters in the URL path between Reviews- and -{Slug}-, appended in a fixed order. Multiple filter tokens are concatenated with -. Applying them by URL rewrite (rather than clicking the filter UI and waiting for a re-render) is faster and stable across renders.

Example with multiple filters — Le Bernardin, 5-bubble English-language couple reviews, page 2:

https://www.tripadvisor.com/Restaurant_Review-g60763-d426543-Reviews-or10-filterRating-5-filterLang-en-filterSegment-COUPLES-Le_Bernardin-New_York_City_New_York.html

Token order is significant in some renders. Tripadvisor's URL builder emits filters in this canonical order: or{N} → filterRating-{n} → filterLang-{code} → filterSegment-{TYPE} → filterMonth-{range} → filterSearch-{keyword} → filterMeal-{n} → filterRoomType-{n}. Off-order tokens sometimes redirect to a canonicalized URL (extra round-trip) or 404. Always emit in the order above.

3a. Sort order (DOM-only filter)

Sort order is not encoded in the URL. It is set via a dropdown that updates Apollo state in-place. After page load, snapshot, click the Sort dropdown (selector [data-test-target="reviews-sort-dropdown"]), pick the matching item:

• mostRecent — "Most Recent" (default)
• topRated — "Highest"
• detailed — "Detailed"
• mostHelpful — "Most Helpful"

Wait ~1.5s for the review list to re-render before extraction.

4. Load the page and harvest entity-level fields

browse open "$URL" --remote --session "$SID"
browse get markdown body --remote --session "$SID" > page.md

Entity-level fields surface in the page header. Extract from the DOM using these selectors / patterns:

Always read these from the first page load (the or0 page), even if your caller only wants a later pagination slice — entity-level data is identical across pages but the histogram block sometimes lazy-loads on or>0 URLs.

5. Extract review cards

Review cards are <div data-automation="reviewCard"> blocks. Each card contains:

If the page returns a Datadome captcha mid-session (rare with --verified, but possible after rapid filter changes), reload once after a 5-10 s pause; if the wall persists, document the wall in the response rather than retrying indefinitely. The session is poisoned at that point.

6. Paginate

Tripadvisor's review page is server-rendered with 10 reviews per page. Paginate by re-loading the URL with the next or{N} token (or10, or20, …). Cap by the user's requested limit or by the entity's totalReviewCount. For limit values above ~50, batch the requests and respect a ≥ 2 s spacing — rapid back-to-back loads from the same session sometimes trigger an intermediate Datadome challenge that times out a long extraction.

The on-page "Show more reviews" button on the bottom of long pages triggers Apollo's LocationReviewListQuery GraphQL and lazy-appends in-place — don't use that path. It is fragile, fires duplicate-review responses on filter changes, and locks the session to one filter dimension at a time. URL-rewrite pagination is faster, parallelizable, and lets the caller jump directly to deep offsets.

7. Output

Assemble entity + reviews into the structured output below (see "Expected Output"). On any wall hit or filter combination that produces zero results, return success: true, reviews: [] with a wall_hit field if the empty result was caused by anti-bot interference rather than a legitimate zero-match filter — wall_hit: true is set when the page rendered the Datadome challenge body, wall_hit: false when the page rendered a proper "no reviews match your filters" message.

Site-Specific Gotchas

• Datadome is the wall, not Akamai. Confirmed via 2026-05-18 probe — the 403 body redirects to captcha-delivery.com and carries a dd config object with b: 1738966, s: 46694. This means Verified strategies that work against Akamai (e.g. residential-only without verified fingerprint) are insufficient — Datadome additionally fingerprints JS execution timing, canvas, and TLS JA3. Use Browserbase --verified which handles these.
• Internal GraphQL is reachable but cookie-signed. POST https://www.tripadvisor.com/data/graphql/ids returns 405 on GET (endpoint alive) — but every reviewed PoC of out-of-band replay shows that the GraphQL handler requires session cookies (TADCID, TASameSite, datadome) that are minted by Datadome's interstitial JS and are bound to the browser fingerprint that minted them. Replaying captured request bodies from curl or browse cloud fetch returns 403. Treat the GraphQL endpoint as confirmed-unreachable for out-of-band use. Don't waste turns trying to bypass.
• The official Tripadvisor Content API exists at https://api.content.tripadvisor.com/api/v1/location/... (returns 401 Unauthorized with {"message":"Unauthorized"} to unauthenticated callers, confirmed 2026-05-18). Requires a partner API key from tripadvisor.com/developers (free tier: 5,000 calls/month, paid tiers higher). For high-volume use, the partner API is dramatically more reliable than scraping. Flag this to the caller — if they have a partner key, route them to the partner-API skill (when it exists) instead of running this one.
• /TypeAheadJson is Datadome-walled even though it returns clean JSON for real browsers. Don't use it for name resolution — use the /Search page in-session (browse open /Search?q=...) or the partner-API search endpoint when keyed.
• /AllReviews, /ExpandedUserReviews, /SortReviews, /SetReviewFilter, /RatingPagingAjax, /UserReviewController, /ShowUserReviewsRestaurants, /ShowUserReviewsHotels, /ShowUserReviewsAttractions are all in robots.txt Disallow — they're legacy/internal endpoints. Don't probe them; they 403 or 404 from outside the session and aren't reachable in modern review widgets anyway.
• or{N} offsets are zero-indexed, step of 10. Page 1 = or0, page 2 = or10, page 50 = or490. Tripadvisor recently (Q4 2025) caps URL-driven pagination at or9990 (1,000 pages); deeper offsets return a "We can't show you reviews past page 1000" page. For high-review-count entities (Eiffel Tower, Statue of Liberty), this is a hard ceiling — communicate it to the caller.
• Multi-rating filter requires multiple round-trips. Tripadvisor's UI only supports one filterRating token at a time (e.g. you can request "5-bubble only" or "4-bubble only" but not "4-or-5"). To return a multi-rating slice, request each rating separately and merge client-side. Same for trip type, language, and meal — single-value tokens only.
• Locale-matched TLD beats .com for non-US entities. UK restaurants on tripadvisor.com redirect to .co.uk with a different g{geoId} (Tripadvisor maintains separate geo trees per TLD). Open .co.uk directly when the entity is UK-based to avoid the redirect and the per-locale geo-tree mismatch.
• bubble_* class encoding: bubble_10 = 1.0, bubble_15 = 1.5, …, bubble_50 = 5.0. Half-bubble ratings exist for the aggregate but not for individual reviews (individual reviews are integer 1-5 only). Some Tripadvisor markets use different class prefixes (uctUV ... bubble_* vs _F bubble_*) — match the numeric suffix, ignore the prefix.
• Translated reviews vs. native-language reviews: Tripadvisor auto-translates non-English reviews on .com pages. The translated text appears in [data-automation="reviewText"]; the original language source is in [data-test-target="reviewOriginal"] if the reviewer wrote in a different language. Capture both when available — caller may want either. The translation is machine-generated (Google Translate API under the hood) and is occasionally garbled.
• Owner responses are not displayed on or0 for some hotels when there are >50 responses — they get paginated separately. The full response thread isn't reachable through the standard review URL; document this as a known truncation if owner responses matter.
• The "Read more" button is required for reviews >200 characters. Without clicking it, [data-automation="reviewText"] returns the truncated preview ending in …. Detect the aria-expanded="false" state and click it before extraction. The expansion is in-place — no DOM re-render, no scroll.
• status: candidate — this skill was authored from out-of-band probe evidence (Datadome wall confirmed, GraphQL endpoint mapped, robots.txt scoped, URL-filter taxonomy enumerated). The skill has not been end-to-end validated through a Browserbase verified+proxied browser session in the authoring sandbox because the sandbox's network policy blocks the connect.browserbase.com WebSocket. Re-validate against a live session before promoting to launched. Specific items to verify: (1) --verified --proxies clears the Datadome challenge on first load; (2) URL-token order matches the canonical order documented in step 3; (3) data-automation selectors are still current (Tripadvisor renames these every ~2 quarters); (4) JSON-LD blocks carry the documented fields.

Expected Output

Success — entity + reviews returned

{
  "success": true,
  "wall_hit": false,
  "entity": {
    "name": "Le Bernardin",
    "type": "restaurant",
    "location_id": "426543",
    "geo_id": "60763",
    "canonical_url": "https://www.tripadvisor.com/Restaurant_Review-g60763-d426543-Reviews-Le_Bernardin-New_York_City_New_York.html",
    "address": {
      "street": "155 W 51st St",
      "city": "New York City",
      "state": "NY",
      "postal_code": "10019-7402",
      "country": "US"
    },
    "lat": 40.761742,
    "lon": -73.981903,
    "phone": "+1 212-554-1515",
    "website": "https://www.le-bernardin.com",
    "overall_rating": 4.5,
    "review_count": 4912,
    "ranking": {
      "rank": 12,
      "total": 8423,
      "category": "Restaurants in New York City"
    },
    "rating_distribution": {
      "5": 3120,
      "4": 1244,
      "3": 312,
      "2": 142,
      "1": 94
    },
    "cuisine": ["French", "Seafood", "Contemporary"],
    "price_range": "$$$$",
    "hours": [
      "Mo 17:30-22:00",
      "Tu-Fr 12:00-14:30,17:30-22:00",
      "Sa 17:30-22:00"
    ]
  },
  "filters_applied": {
    "rating": [5],
    "language": "en",
    "trip_type": "COUPLES",
    "sort": "mostRecent",
    "limit": 10,
    "offset": 0
  },
  "reviews": [
    {
      "review_id": "rr988123456",
      "permalink": "https://www.tripadvisor.com/ShowUserReviews-g60763-d426543-r988123456-Le_Bernardin-New_York_City_New_York.html",
      "rating": 5,
      "date": "2026-04-12",
      "title": "Best meal of our anniversary trip",
      "body": "Eric Ripert's kitchen continues to set the bar... [full text, Read-more expanded]",
      "trip_type": "COUPLES",
      "language": "en",
      "stay_date": null,
      "helpful_votes": 7,
      "subratings": null,
      "photos": [
        "https://media-cdn.tripadvisor.com/media/photo-l/abc.jpg"
      ],
      "owner_response": null,
      "reviewer": {
        "name": "JaneD",
        "profile_url": "https://www.tripadvisor.com/Profile/JaneD",
        "hometown": "Boston, Massachusetts",
        "contributions": 47,
        "helpful_votes_total": 92,
        "badge": "Top Contributor"
      }
    }
  ]
}

Hotel — with subratings + owner response

{
  "review_id": "rr991234567",
  "rating": 4,
  "date": "2026-03-28",
  "trip_type": "FAMILIES",
  "stay_date": "2026-03",
  "title": "Lovely stay, minor service hiccup",
  "body": "...",
  "subratings": {
    "location": 5,
    "cleanliness": 5,
    "service": 3,
    "value": 4,
    "rooms": 4,
    "sleepQuality": 5
  },
  "owner_response": {
    "text": "Thank you for the detailed feedback — we've addressed the front-desk timing issue with our team. We hope to welcome you back.",
    "date": "2026-04-02",
    "responder_name": "Hotel Manager"
  },
  "reviewer": { "name": "...", "...": "..." }
}

Wall hit — anti-bot interception

{
  "success": false,
  "wall_hit": true,
  "reason": "datadome_challenge",
  "entity": { "name": "Le Bernardin", "location_id": "426543", "canonical_url": "..." },
  "filters_applied": { "...": "..." },
  "reviews": [],
  "diagnostic": "Page returned Datadome captcha (403, captcha-delivery.com redirect) after applying filters. Session fingerprint may be flagged. Retry with a fresh --verified --proxies session."
}

Ambiguous entity — caller needs to disambiguate

{
  "success": false,
  "reason": "ambiguous_name",
  "candidates": [
    {
      "name": "The Ritz London",
      "type": "hotel",
      "city": "London",
      "country": "UK",
      "canonical_url": "https://www.tripadvisor.co.uk/Hotel_Review-g186338-d193091-Reviews-The_Ritz_London-London_England.html",
      "location_id": "193091"
    },
    {
      "name": "The Ritz-Carlton, Naples",
      "type": "hotel",
      "city": "Naples",
      "country": "US",
      "canonical_url": "https://www.tripadvisor.com/Hotel_Review-g34467-d108845-Reviews-The_Ritz_Carlton_Naples-Naples_Florida.html",
      "location_id": "108845"
    }
  ]
}

Entity not found

{
  "success": false,
  "reason": "entity_not_found",
  "query": "Foo Bar Restaurant, Nowheresville",
  "candidates": []
}

Pagination ceiling hit

{
  "success": true,
  "wall_hit": false,
  "entity": { "...": "..." },
  "filters_applied": { "offset": 9990, "limit": 10 },
  "reviews": [],
  "diagnostic": "Tripadvisor caps URL-driven pagination at or9990 (1,000 pages). Use filter dimensions (rating / trip_type / language) to narrow the result set instead of paginating deeper."
}