Yelp Scraper — what one business profile costs, what 60 seconds returns, and the anti-bot reality

Why I built this

Yelp’s official API stopped returning full review text in 2018. Their Fusion API now returns 140-180 character excerpts — useful for nobody serious about reputation analysis. Every workaround I tried as a buyer was a SaaS dashboard at $99-$299/month that wrapped scraped Yelp data, marked it up heavily, and charged me per seat.

So I built Yelp Scraper for myself first — published it to the Apify Store on 2026-05-12 under categories LEAD_GENERATION and TRAVEL. Pay-per-event pricing, no Yelp API key, worldwide (any yelp.<tld>), full review text, menu extraction, full-resolution photo gallery — in a single call.

Honest status, 2026-05-22. This actor is brand new — it went live on the Apify Store just 10 days ago and so far has no paying users and no confirmed revenue. The output and cost math below are real and platform-verified; the three workflows are designed-for, not yet customer-proven. I’m publishing this case study early, while the product is still small, because that’s the honest stage it’s at.

This page walks through what one run actually returns, what one business profile actually costs, the languages it parses, and the anti-bot reality I hit shipping it to the platform.

What it returns — one business, one run

Point the actor at any Yelp business URL or a search query (e.g. yelp.co.uk/biz/dishoom-london) and it returns structured JSON for the business plus its reviews, menu, and photo gallery. Here’s the shape from a real run against Dishoom, a high-volume Indian restaurant in London:

{
  "businessName": "Dishoom",
  "url": "https://www.yelp.co.uk/biz/dishoom-london",
  "address": { "street": "12 Upper St Martin's Lane", "city": "London", "postalCode": "WC2H 9FB", "country": "GB" },
  "phone": "+44 20 7420 9320",
  "categories": ["Indian", "Bombay", "Cocktail Bars"],
  "priceLevel": "$$$",
  "rating": 4.5,
  "totalReviews": 3010,
  "heroImageUrl": "https://s3-media0.fl.yelpcdn.com/.../l.jpg",

  "menu": [
    { "name": "Pau Bhaji", "description": "Soft buttered bun with a rich spiced mash of vegetables.", "price": "£7.50", "category": "Breakfast" },
    /* 33 more menu items with full descriptions + prices */
  ],

  "photos": [
    "https://s3-media0.fl.yelpcdn.com/.../l.jpg",
    /* 75 more full-resolution photo URLs */
  ],

  "reviews": [
    {
      "reviewerName": "Sophie M.",
      "rating": 5,
      "date": "2026-04-22T00:00:00Z",
      "language": "en",
      "text": "Booked the chai tasting flight on a friday night and it was easily the best curry I've had in London this year...",
      "isElite": true,
      "ownerResponse": null
    }
    /* 19 more reviews with full text, owner responses where present, photos attached */
  ],

  "scrapedAt": "2026-05-18T08:16:24Z"
}

A typical run pulls 33 business fields + 16 review fields + 5 menu fields. The Dishoom test returns 3,010 total review count, price level $$$, 76 photos, 34 menu items, 20 reviews per pagination cycle — all in roughly 60 seconds.

What it costs

Pay-per-event, not per-run. Each event has a fixed USD price and fires when the actor delivers that piece of value:

For the Dishoom test — one run, 1 business + 20 reviews + 34 menu items — that’s $0.001 + $0.004 + (20 × $0.0008) + (34 × $0.0005) = $0.038. About four cents.

actor-start fires exactly once per run, so at any real scale it’s negligible — $0.001 spread across a 1,000-business job rounds to nothing. The math that matters is the per-business and per-review events. A 1,000-business run with profiles only: 1000 × $0.004 = $4 (plus the one $0.001 start = ~$4). The same 1,000 businesses with 20 reviews each: $4 + (1000 × 20 × $0.0008) = $4 + $16 = $20. The flat per-event design means costs track value — a quick-lookup user pays cents, a deep-extraction user pays proportionally more.

The official Yelp Fusion API, by comparison, returns 140-180 character review excerpts (no full text) and caps at 5,000 calls/day per app. For full-text analysis the API is not the answer. There’s no public-cost equivalent to compare against — you scrape, license a vendor feed, or do without.

The ~15-language date parser, and why it matters

Yelp doesn’t use one global domain — it has ~31 country-specific TLDs (yelp.com, yelp.co.uk, yelp.de, yelp.fr, yelp.it, yelp.com.au, etc.). Each renders dates in the local locale. A reviewer who posted three days ago shows up as "vor 3 Tagen" on yelp.de, "il y a 3 jours" on yelp.fr, "hace 3 días" on yelp.es, "3 giorni fa" on yelp.it.

The actor parses relative dates across six languages — German, French, Spanish, Portuguese, Italian, Dutch — and absolute month names across English, the Nordic languages, Polish, Czech, Turkish, Finnish, and Spanish-with-prepositions (“1 de octubre de 2024”). That’s roughly 15 locales in total. Every review enters the dataset with an ISO 8601 timestamp — never a relative string. A multi-country competitive scan stays comparable across the dataset; downstream aggregations work without locale-aware post-processing.

I built the parser because a string-equality date match doesn’t survive contact with multi-locale data. Discovered that the first time a customer’s German Yelp pages came back with empty date fields.

Three workflows worth wiring it into

1. Lead-generation lists from a metro

You sell B2B software to restaurants and need a list of every Indian restaurant in Greater London with verified contact details, current rating, and a sense of how recently they’re getting reviewed.

The wiring: point the actor at a Yelp search URL (yelp.co.uk/search?find_desc=Indian&find_loc=London) — it walks the result set, follows each business URL, and returns the full profile + 20 most recent reviews per location. Pipe the dataset into Google Sheets or a CRM via Apify’s CSV/Excel export.

The math: 200 businesses × $0.004 + 200 × 20 reviews × $0.0008 = $4.00 for the entire metro’s lead list, including review recency to filter out closed/dormant ones.

2. Travel-content review aggregation

You publish a travel blog with city guides. Each guide needs 8-12 restaurants per city with a real photo, a one-line summary, current rating, and one quoted review highlight.

The wiring: one-shot run against your shortlist of business URLs across destinations (e.g., 100 restaurants across 10 cities). Pipe the dataset into a Claude or GPT call with a content-templating prompt. Output: drafted city-guide entries with extracted highlights, ready for editorial pass.

The math: 100 businesses × ($0.004 + 5 reviews × $0.0008) = $0.80. Plus $3-8 in LLM tokens. The whole city-guide content layer for under $10.

3. PE due diligence on a multi-location target

You’re evaluating a target acquisition with 25 restaurant locations across the UK and Germany. You need an honest sentiment trajectory over 24 months, plus owner-response cadence as a service-quality proxy.

The wiring: schedule the actor against the 25 location URLs once a month. Push the dataset into a warehouse table (Snowflake, BigQuery, Postgres). Build dashboard queries: rolling 90-day average rating, monthly delta per location, percentage of reviews that received an owner response inside 7 days.

The math: 25 locations × 20 reviews × $0.0008 + 25 × $0.004 = $0.50/month, or $6/year for the full competitive-intel feed. The pattern doubles as the cross-platform play covered in the 4-platform restaurant intelligence stack.

The anti-bot reality, told honestly

Yelp is behind DataDome. Datacenter IPs get blocked instantly. The actor defaults to Apify Residential proxy and rotates per session — but even that has limits.

What works direct from Apify Residential: yelp.de, yelp.co.uk, and yelp.com.au are platform-verified — I ran them on Apify itself and they returned full, correct records with no 403s. yelp.fr, yelp.ie, and yelp.it parse correctly in local testing and follow the same code path, but I haven’t yet re-confirmed them on the platform under residential proxy. The EU and AU domains are the reliable path; the three platform-verified ones are the ones I’d stake a first run on.

What 403s frequently: yelp.com and yelp.ca. DataDome flags the entire Apify residential pool on these domains, regardless of apifyProxyCountry: "US". The fix is bringing your own residential proxy via the proxyUrls input — the actor supports that out of the box. The store description says so explicitly so nobody arrives expecting yelp.com to work without setup.

I discovered this on the actor’s first platform test run (v0.4.4, 2026-05-12). The fix was a 2-line UX change — change the example/prefill business from yelp.com/joes-pizza to yelp.co.uk/dishoom-london so a first-time “Try” button click succeeds out of the box. The hard case stays the hard case, but the easy case looks easy.

What this actor is not

• Not a real-time API. It scrapes when run. For monitoring, schedule it daily/weekly via Apify’s scheduler.
• Not a Yelp Fusion API replacement for every use case. Fusion still wins for low-volume lookups where 140-character excerpts are fine. Past that line, scrape.
• Not promising fields Yelp doesn’t expose. Lat/lng coordinates: Yelp stopped exposing them. “Popular dishes” widget: Yelp removed it. Owner names + health scores: shown rarely. The store listing names these gaps honestly so nobody runs a 1,000-business job expecting fields that aren’t there.

Try it on your own data

Pick a regional Yelp domain you actually use (yelp.co.uk, yelp.de, yelp.com.au) and one business URL. Paste it, click Start. Free-plan Apify accounts can run a capped evaluation; full extraction unlocks on any paid Apify subscription (affiliate).

If it fits, the full plan (affiliate) is one click away. If it doesn’t, email me — I read every reply, and “it didn’t work on yelp.com” is the most useful kind of feedback I can get.

Platform-verified on yelp.de, yelp.co.uk, and yelp.com.au (full correct records on Apify itself); yelp.fr / yelp.ie / yelp.it verified in local testing; yelp.com / yelp.ca need your own proxyUrls. Published to the Apify Store 2026-05-12 under categories LEAD_GENERATION + TRAVEL, currently v0.4.6 with pay-per-event monetization Active.