GMapReviewAI

GMapReviewAI is an AI Reputation Copilot for local businesses built with Lamatic. It collects live Google Maps reviews, analyzes customer sentiment, highlights recurring issues and strengths, benchmarks competitors, and drafts professional responses for unanswered reviews—all in a concise, evidence-based report that helps business owners make faster, smarter decisions.

Built for the Lamatic AgentKit Challenge.

---

The problem

A local business — a cafe, a clinic, a salon, a gym — accumulates Google Maps reviews faster than any single owner can read them. Nobody has time to read 40 new reviews a week and turn them into a to-do list, and even fewer businesses systematically compare themselves to the competitor two doors down. The reviews that most need a reply — the negative, unanswered ones — are exactly the ones an overwhelmed owner avoids opening.

This is a narrow, concrete, recurring pain, not a "build me a research agent" problem — which is what made it a good fit for a focused submission.

Why this doesn't already exist in AgentKit

AgentKit already has several review- and research-adjacent kits, so this template was designed to sit in a real gap rather than overlap them:

• review-analyzer is a browser extension that scores the trust level of e-commerce product reviews on the page you're currently viewing — shopper-facing, single-page, no external API.
• review-responder classifies and drafts replies to reviews that are already supplied as input — it doesn't fetch anything itself or compare against a competitor.
• reddit-scout searches Reddit threads for opinions about a product — a different platform, a different data shape, no ratings or competitive benchmarking.
• localboost-ai analyzes a local business's own website for lead-generation and outreach — not Google Maps reviews.
• founder-lens is a general startup/market research agent — not scoped to reviews or to a specific business's reputation.

None of them fetch live Google Maps review data, benchmark it against a named competitor, or draft replies to specific unanswered reviews. That combination is what GMapReviewAI does.

The approach

A single Lamatic flow, three steps after the trigger:

API Request  →  Fetch Reviews (Apify)  →  Generate GMapReviewAI (LLM)  →  API Response

1. Fetch Reviews (Apify) — a code node calls Apify's compass/Google-Maps-Reviews-Scraper Actor (46k+ users, 4.83★, 99.8% run success rate on the Apify Store) with the business's Google Maps URL plus any competitor URLs, all in one request. It groups the results by place, matches "the business" by name (Apify does not guarantee the dataset preserves input order — confirmed by testing, see Assumptions below), and reshapes everything into compact per-place summaries: aggregate rating, a capped sample of text-bearing reviews with any per-aspect Food/Service/Atmosphere ratings Google provides, and up to five recent negative reviews that have no owner response yet.
2. Generate GMapReviewAI — an LLM node reasons over that structured data (never the raw scrape) to produce a Markdown report: headline, what's working, what needs fixing, a competitive comparison (only if competitor data was supplied), and response drafts for the business's own unanswered negative reviews.
3. API Response — returns the full report plus two convenience numbers (business_average_rating, business_total_reviews_fetched).

Apify was the right tool here specifically because Lamatic's existing scraping integration (Firecrawl, used by article-summariser and the firecrawl-* kits) does generic page-to-markdown extraction — it has no concept of a Google Maps place, a star rating, an owner response, or a per-aspect rating. Apify's Actor catalog has a purpose-built, actively maintained scraper for exactly this structured data, which is what makes the theme-clustering and competitor-comparison reasoning possible at all.

https://www.google.com/maps/place/Aman+New+York/@40.7440484,-74.0277935,13z/data=!4m13!1m2!2m1!1snewyork+best+hotel!3m9!1s0x89c259d50b96eb33:0x1814c7086ffb6f32!5m2!4m1!1i2!8m2!3d40.7629802!4d-73.9746883!15sChNuZXcgeW9yayBiZXN0IGhvdGVsIgOoAQGSAQVob3RlbKoBRhABKgkiBWhvdGVsKAAyHhABIhq9YfxLtTD4wUCZ-vk86m7j13cfkxKTN1nItTIXEAIiE25ldyB5b3JrIGJlc3QgaG90ZWzgAQA!16s%2Fg%2F11f62dpxyl?entry=ttu&g_ep=EgoyMDI2MDcxNS4wIKXMDSoASAFQAw%3D%3D

https://www.google.com/maps/place/The+Otter/@40.7240253,-74.0793356,13z/data=!4m12!1m2!2m1!1snewyork+best+hotel!3m8!1s0x89c2590008ce9f97:0x9ab932497cc4f892!5m2!4m1!1i2!8m2!3d40.7240253!4d-74.0031179!16s%2Fg%2F11lnl53_b6?entry=ttu&g_ep=EgoyMDI2MDcxNS4wIKXMDSoASAFQAw%3D%3D

The result

See assets/sample-report.md for a complete, real worked example. It is not a mockup — the Apify Actor call in scripts/GMapReviewAI_fetch-reviews.ts was run live during development against two real Google Maps listings (Leopold Cafe and Cafe Mondegar, both in Colaba, Mumbai), and the report was produced by hand-running that exact data through the exact prompts in prompts/. A short excerpt:

Headline: Leopold Cafe is holding a strong 4.2★ all-time aggregate across 32,641 Google reviews, and the 14 most recent reviews in this sample average 4.21★...

What Needs Fixing: 1. Service lags food and atmosphere, even in happy visits — one 5-star review rated service 3/5 while rating food and atmosphere 5/5...

Setup

You need a Lamatic account and an Apify account (Apify has a free tier that comfortably covers testing this kit — a few hundred reviews costs well under $1).

1. Get an Apify API token. Sign up at apify.com if you don't have an account, then copy your token from console.apify.com/settings/integrations.

2. Create a project in Lamatic Studio and add the token as a project secret named APIFY_API_TOKEN (Settings → Secrets) — not a repo .env value, since this is a template with no standalone app.

3. Build the flow. Fastest path: create a new flow, open the Config tab, and paste in assets/lamatic-flow-config.yaml, then fill in the two placeholders (paste the script/prompt file contents where noted, and pick a model). If your Studio version doesn't support the Config-tab paste, build the same four nodes by hand:

   Node	Type	Key settings
   API Request	graphqlNode (trigger)	Advance schema — see flows/GMapReviewAI.ts header or the YAML file above
   Fetch Reviews (Apify)	codeNode	Code = contents of scripts/GMapReviewAI_fetch-reviews.ts
   Generate GMapReviewAI	LLMNode	System/user prompts = contents of the two files in prompts/; pick any chat-capable model
   API Response	graphqlResponseNode	Output mapping — see the YAML file above

   Connect them in that order, plus the standard response edge from the trigger to API Response.

4. Deploy the flow, then test it with:

   {
     "business_name": "Your Business Name",
     "business_maps_url": "https://www.google.com/maps/place/...",
     "competitor_maps_urls": [],
     "max_reviews_per_place": 30,
     "reviews_since": "3 months"
   }

5. Add one or more entries to competitor_maps_urls once the single-business report works, to see the comparison section appear.

Assumptions & tradeoffs

• Apify's dataset order is not guaranteed to match input order. A live test run with the business URL listed first still returned the competitor's rows first in the dataset. The code node matches "the business" by name against the resolved place title instead of by position — worth knowing if you extend this kit.
• personalData: false by default. Google returns reviewer names as null under this Apify setting. Report quality is unaffected (theme-clustering never needed names); only response drafts degrade gracefully to a generic greeting instead of a personalized one. This was a deliberate privacy-by-default choice, with a documented opt-in escape hatch in the script.
• Many Google Maps reviews have a star rating and no text. In real samples pulled during development, roughly half of all reviews were rating-only. The code node counts every rating toward the aggregate numbers but only clusters themes from the subset that has text — a report built on a small business's early reviews will honestly say the sample is thin rather than overstate confidence (the system prompt enforces this).
• This is a template, not a kit — a single flow, no standalone app, no scheduling. Response drafts are just that: drafts. The flow does not post them back to Google Business Profile, which would need a separate OAuth-based integration and was left out to keep this contribution focused, per the challenge's own guidance that a smaller, well-executed idea beats a padded one.
• Cost scales with max_reviews_per_place × number of places. The Apify Actor is pay-per-event (roughly $0.0006 per review scraped); the defaults (30 reviews per place) keep a typical business + one competitor run well under a cent.

Possible extensions

• Trend over time: store each run's headline stats via a Lamatic memory node and chart drift week over week, instead of only ever comparing within a single run's sample.
• Other review platforms: the same reshape-then-reason pattern works for App Store / Play Store reviews via a sibling Apify Actor — useful for software products rather than physical locations.
• Scheduled delivery: a second flow on a weekly trigger that calls this one and posts report to Slack or email, mirroring the pattern in slack-ask-bot.
• Write-back: posting the drafted responses to Google Business Profile via its own API, once a business has reviewed and approved a draft.

Files

GMapReviewAI/
├── lamatic.config.ts                              # Kit metadata
├── agent.md                                        # Agent identity/capability doc
├── README.md                                        # This file
├── flows/gmap-review-ai                      # The flow graph
├── prompts/gmap-review-ai_llmnode-540_system_0.md
├── prompts/gmap-review-ai_llmnode-540_user_1.md
├── scripts/gmap-review-ai_code-node-310_code.ts        # Apify call + reshaping logic
├── model-configs/gmap-review-ai_llmnode-540_generative-model-name.ts
├── constitutions/default.md
└── assets/
    ├── sample-report.md                              # Real worked example
    └── lamatic-flow-config.yaml                      #