AI Icebreaker Generator

Turn a batch of Google Maps place records into one hyper-personalized cold-outreach opening line per business. Bring your own LLM key (Gemini has a generous free tier). Two modes: base uses only fields already present on the place, enriched also fetches the homepage and folds a summary into the prompt.

Feed it the output of the Google Maps Scraper and get one ready-to-send opener per business.

What It Does

Given a Google Maps place record, the actor asks an LLM to write ONE sentence that:

• Is at most 200 characters
• References something specific about the business (rating, category + city, tenure, description, homepage detail)
• Matches the requested tone (friendly, professional, casual, direct)
• Is grounded in the input data. Never a generic template, never invented facts.
• Falls back to an empty string when the record has too little signal (no name, or missing category AND reviews AND description)

Modes

Base (default, zero-fetch)

Uses only fields already on the place record:

• title / name
• category, categories
• city, country (from complete_address)
• rating / review_rating, review_count
• description
• phone, website (reference only, not fetched)

Enriched (opt-in)

Set enableEnriched: true. In addition to base mode, the actor fetches the business's homepage, condenses it into a ~1000 character markdown summary (headings + tagline + prose paragraphs), and appends it to the LLM prompt. Falls back silently to base mode when the site returns non-200, redirects strangely, or trips an anti-bot challenge.

Input

Two shapes are accepted, and you can mix them in one run.

Flat places list (natural shape, just pipe the google-maps-scraper output):

{
  "places": [
    {
      "place_id": "ChIJN1t_tDeuEmsRUsoyG83frY4",
      "title": "Example Coffee House",
      "category": "Coffee shop",
      "website": "https://example-coffee.example",
      "rating": 4.8,
      "review_count": 1240,
      "complete_address": { "city": "Berlin", "country": "Germany" }
    }
  ],
  "tone": "friendly",
  "outputLanguage": "en",
  "geminiApiKey": "AIza..."
}

Companies shape (with a company_id passthrough for join-back to your CRM):

{
  "companies": [
    { "company_id": "acme-42", "place": { "place_id": "ChIJ...", "title": "Example Company Ltd." } }
  ],
  "tone": "professional",
  "geminiApiKey": "AIza..."
}

Input parameters

Parameter	Type	Default	Description
places	array	(none)	Flat places array — pipe directly from Google Maps Scraper output.
companies	array	(none)	Wrapped shape with company_id per row for CRM join-back. Mix with places in the same run if useful.
tone	string	"friendly"	Voice of the generated line: friendly (warm and professional), professional (buttoned-up), casual (chatty), direct (punchy, to the point).
focus	string	(none)	Optional free-text guidance passed to the LLM alongside the place record. Steers every line toward your specific outreach angle. Example: "Focus on their photo count and their newest reviews", or "Mention that we specialize in local SEO for care homes".
enableEnriched	boolean	false	When true, fetches the business's homepage, condenses it into a short summary, and folds it into the icebreaker prompt. Sharper lines, slightly higher cost. Falls back silently to base mode on 404 / anti-bot.
outputLanguage	string	"en"	ISO 639-1 language code (en, de, fr, etc.) or auto to detect from the website.
llmProvider	string	"gemini"	Primary LLM: gemini, groq, or openrouter.
fallbackProvider	string	(none)	Fallback on auth/quota errors from the primary.
fallback2Provider	string	(none)	Second-line fallback.
geminiApiKey / groqApiKey / openrouterApiKey	string	(none)	At least one is required — the actor never makes LLM calls on your behalf, keys go straight to your provider account.

The focus field — usage tips

Two shapes work well:

1. Content-specific: point the model at a signal you care about — "Focus on their photos and their newest reviews", "Mention their recent hiring signal", "Reference their tech stack".
2. Angle-specific: tell the model what YOU sell — "Mention that we specialize in local SEO for care homes", "We are a payments processor pitching lower fees to independent restaurants", "We build websites for solo dentists".

Leave it empty when you want the model to pick the sharpest angle itself. focus is a steering signal, not a directive — if the place record has no photos and you asked to reference photo count, the line will pivot to another salient signal rather than fabricating one.

LLM Configuration

You must supply at least one LLM key. All three providers are supported:

Provider	Free tier	Get a key
Gemini	Yes (generous)	https://aistudio.google.com/app/apikey
Groq	Yes	https://console.groq.com/keys
OpenRouter	Pay-as-you-go, cheap	https://openrouter.ai/keys

Set llmProvider to pick the primary, fallbackProvider and fallback2Provider for the chain. Same-provider transient errors (5xx, timeouts) get one retry; auth/quota errors fall through to the next provider.

Output

One row per input place:

{
  "company_id": "acme-42",
  "place_id": "ChIJN1t_tDeuEmsRUsoyG83frY4",
  "name": "Example Coffee House",
  "icebreaker_line": "Noticed you are running one of the highest-rated (4.8 stars, 1,240 reviews) coffee shops in Berlin. Curious how you keep that streak going.",
  "icebreaker_enriched": false,
  "llm_provider": "gemini",
  "llm_model": "gemini-2.0-flash",
  "llm_tokens_input": 340,
  "llm_tokens_output": 45,
  "generated_at": "2026-07-17T10:30:00Z"
}

icebreaker_line is an empty string when the record had too little signal to produce a specific line. The actor never invents facts to fill the gap.

How To Use It

1. Run Google Maps Scraper to collect places for a niche and city.
2. Feed the dataset into this actor as places (or wrap each in companies for join-back).
3. Pick a tone, set your Gemini key, and (optionally) turn on enriched mode.
4. Drop the output into your outbound sequencer.

Pricing

Pay-per-result via Apify Pay-Per-Event billing:

Event	Price	Description
Actor start	$0.001	Charged once per run
Icebreaker generated (base)	$0.003	Charged per row in base mode
Icebreaker generated (enriched)	$0.005	Charged per row when enriched mode succeeded

Generating 1,000 lines in base mode costs about $3. Empty input rows and rows the model refused to write for (insufficient signal) are still counted as a result. LLM API usage is billed by your provider directly, not by this actor.

Related Actors

Discovery:

• Google Maps Scraper - collect places in bulk

Enrichment:

• GBP Completeness Audit - score Google Business Profiles for lead prioritization
• Website Email Scraper - pull emails and phones from company sites
• Website Contact Extractor - LLM-powered people extraction

Notes

• Empty input is free. No enrichment-start is billed unless there is at least one valid place.
• The proxy configuration is only used by enriched mode's homepage fetch.
• Language: set outputLanguage: "auto" to have the model match the business's country language.

Issues & Feature Requests

If the model keeps refusing lines for a class of business, or if you would like a new tone, open an issue on the actor's Issues tab and we will take a look.