Phone-verify business locations

Bring a list of businesses. This Actor calls each one, asks the questions you set, and returns a structured row per call.

Every other Actor here reads what a business published. This one phones and asks.

That matters because published data goes stale in ways scraping cannot see. A listing says open; the number is disconnected. Hours say 9pm; the kitchen stops at 8. The website says walk-ins welcome; they stopped in March. None of that is on a page anywhere - it only exists on the phone.

Try it first, free: leave Check my list, don't call ticked and press Start. The Actor reads your list and shows the columns it found, the exact questions each business would get, the rows it would skip and the most the calls could cost. Nobody is called and nothing is charged.

What you put in

A list of businesses, from wherever you have one:

• A CSV file or link. Upload a CSV (or tab-separated) file with a header row, or paste a link to one. A Google Sheet link works if the sheet is shared as "Anyone with the link can view" or published as CSV. Excel files: save as CSV first.
• Cells pasted from a spreadsheet, header row included.
• The dataset of a previous Actor run, for example Google Maps results.
• JSON, one object per business.

If you give more than one, the first in this order is used: dataset, CSV, pasted cells, JSON. Columns are detected automatically, so an export from a scraper, a CRM or a spreadsheet works without mapping; fax and owner-mobile columns rank below the business phone.

• US numbers in any common format work: (415) 555-0142, 415.555.0142, +1 415 555 0142. They are converted to international format before dialling. Numbers outside the US are skipped with a reason.
• The same number twice is called once. Every row sharing it gets the result, the extras marked duplicateOf.
• What the list needs: a column of US phone numbers (any header; it is found from the values). A business-name column helps, since the agent says the name on the call. Every other column is passed straight through to the output, so results join back to your own records, and can be used in questions.

A sample to copy into a sheet or the paste box:

Business Name	Phone	City	My ID
Corner Cafe	(415) 555-0142	San Francisco	1001
Night Owl Diner	(415) 555-0199	Oakland	1002

Questions

The form shows the questions themselves: what you see is what the agent asks, of every business, one at a time. It starts with the opening-hours pair; edit or replace them freely.

Read a business its own data back with {column}, any column from your list: Is your address still {address}? Matching ignores case, spaces and underscores, so {Business Name} works too. If your list has no such column the run stops before calling anyone and lists the columns it does have; if only some rows have a value, those rows get one question fewer rather than a question with a hole in it. The run log shows the questions, filled in for a sample business, before the first call.

Ready-made sets, tested wording:

Checking	Questions
Opening hours	Are you open right now? / What are your hours today?
Reservations	Do you take reservations, or is it walk-in only? / For a party of 6, how far ahead should someone book?
Availability	Do you currently have oat milk? / Roughly what does it cost?
New customers	Are you accepting new patients right now? / What is the wait for a first appointment?
Details on file	Is your address still {address}? / Is this the best number for customers to reach you on?

API callers can also pass template (verify_hours, reservations, availability, new_customers, confirm_details) with an empty questions list, and templateVariables for a set's placeholders, e.g. {"item": "oat milk"}.

"Still in business?" is not a question, on purpose. If they are gone, nobody answers, so the answer is already in reached: false plus the outcome. Ask the opening-hours pair and read that column.

What you get back

One row per business:

{
  "name": "Corner Cafe",
  "phone": "+14155550142",
  "phoneRaw": "(415) 555-0142",
  "reached": true,
  "answers": [
    { "item": "Are you open right now?", "status": "answered", "answer": "Yes, open now." },
    { "item": "What are your hours today?", "status": "answered", "answer": "7 AM to 3 PM." }
  ],
  "questionsAsked": ["Are you open right now?", "What are your hours today?"],
  "keyInfo": "Open now; closes at 3 PM today.",
  "outcomeType": "success_no_booking",
  "billedCents": 10,
  "callId": "00000000-0000-4000-8000-000000000000",
  "attempts": [ { "attempt": 1, "outcomeType": "success_no_booking" } ],
  "reachability": { "signal": "answered", "evidence": "Answered on attempt 1 (1 of 1 attempts answered)." },
  "transcript": [ "..." ],
  "source": { "...your original row...": "" }
}

answers is the call analyzer's own reading, as it wrote it. It usually matches your questions one for one, but it can split a question in two or merge two into one, so match on item rather than on position. The questions actually asked for that business are in questionsAsked.

reached: false is a result, not an error. For anyone verifying a list, "nobody picks up" or "the number is disconnected" is exactly the signal worth paying for.

Nothing is dropped quietly. Rows that cannot be called (no usable phone, outside the US) come back with the reason. Businesses past your call budget or a daily limit come back as notAttempted with the reason. If you abort a run gracefully, every business it had not finished gets a row saying so; a hard abort stops the run at once and cannot write them. retryable on each row says whether a later retry pass will call it again. A check-only run writes one row per business with dryRun: true and the exact questionsToAsk.

What the call history says about a business

Every attempt is kept, the failed ones included, because for anyone verifying a list the failures are findings. Each row carries its full attempts history (when each call was placed, how long it lasted, how it ended, and on calls that did not reach the business, what the line itself said), plus a one-line reachability read built from all of it. The most recent evidence wins: a business that answered today is answering, whatever a voicemail said last week.

Signal	What it means
answered	Someone at the business picked up, even if they refused or did not understand.
closure_language	The business's own line said it stopped operating: "permanently closed", "out of business", "no longer in business".
number_not_in_service	The carrier said the number is dead: "not in service", "disconnected". The business may exist; this number does not reach it.
moved	The line said the business moved or changed number. Update the record, do not close it.
wrong_number	Someone answered, but it was not this business. The listed number may have been reassigned.
never_answered	No attempt reached anyone. The evidence line gives the count, how many days it spans and how each attempt ended.
line_problem	Only busy lines or technical failures so far. Says nothing about the business yet.
callback_requested	Someone asked to be called back later; callAfter says when. The line is live.
not_called	No call was placed (skipped, over budget, or a check-only run).

This is evidence, not a verdict. A plain "we're closed right now" is deliberately not read as closure, because that is a live business outside its hours. never_answered across several days is a strong hint a listing is stale, but we do not know the business's hours or time zone, so the dial times are included for you to judge. Retry passes add to the same history, which is what turns one unanswered call into a pattern.

Retries

Two ways, for two kinds of failure.

In the same run, for things that clear in minutes. Set Attempts per business above 1 and pick which outcomes to retry. Busy lines, technical failures and our own agent going silent are selected by default; no answer, voicemail, dropped calls, long holds, a pickup that never engaged, phone menus that never reach a person and recorded announcements can be added. A dropped call is retried only if nothing was answered before it dropped. Attempts to the same business are spaced by Wait before retrying.

When a business asks you to call back later, it gets one callback in the same run, at the time it named (or after 30 minutes), even with retries off. If that time falls after the run's timeout, the row carries callAfter, and a retry pass will not call it before then. "Don't call" is never called back.

In a later run, for things that need hours. Voicemail at 5pm usually means "try tomorrow morning", not "try in two minutes". Pick the previous run's dataset as the input, tick Only call businesses not reached yet, and only the businesses worth another try are called, each keeping its full attempt history. Every pass writes the complete picture: businesses it did not call again are carried forward unchanged (marked carriedForward), so the newest dataset always has every business and is the one the next pass should pick. An Apify Schedule reuses the same saved input, so it would keep reading the first run's dataset; pick the newest dataset for each pass instead.

Fit the retries inside the run's timeout. A run stops at its Timeout (Run options, 1 hour by default) and keeps using Apify compute while it waits between attempts. No attempt starts unless it can finish before the timeout: a business whose next attempt would run past it is not called again in this run, keeps retryable: true, and its row says run timeout in laterAttemptNotPlaced (or reason, if it was never called). The run log warns at the start when your list or your retry settings will not fit. Raise the Timeout, or finish with a retry pass. Waits longer than about 30 minutes are usually cheaper as a later run than as an idle one.

What never gets retried, in the same run or a later one, whatever you select: a business that answered (even if the call later dropped), a business that hung up during or right after our introduction, a number answered by someone who is not the business, and a call whose result we lost track of (it may have succeeded). Calling back someone who just hung up is how numbers get blocked.

Every attempt counts against Maximum calls this run, because PlaceCall's daily limit counts attempts. First attempts always get the budget before any retry, so a retry never costs a business its only call.

Pricing

• $0.10 for each call that reaches a real conversation with someone at the business, whatever they answer, including a "no" or a call that drops mid-conversation. Charged through your Apify account; no separate sign-up and no API key.
• Your first 10 such calls are free, once per Apify account.
• Free: voicemail, no answer, busy lines, phone menus and recordings that never reach a person, hang-ups in the first seconds, wrong numbers, and our own errors. Retries of unanswered calls cost nothing.
• "Check my list, don't call" is always free.
• On Apify's free plan, you get the 10 free calls; after that, calling needs a paid Apify plan. This limit is set by PlaceCall, not by Apify.
• Your maximum cost per run (Run options) is respected: when it is reached, new calls stop, calls already in progress finish, and the rest come back as notAttempted with the reason.

Each row reports costUsd and freeCallsUsed; the run's status message sums up free calls used and left, and what was charged. A row with chargePending: true had a call that was not yet settled when the row was written (a brief outage on our side), or whose charge was interrupted by a restart and could not be confirmed, so its cost is not in the row. The run's status message has the outcome: it counts the call if it was charged later, and says so if it was never charged.

Apify compute is small: the Actor runs at 256 MB by default (512 MB when Maximum calls this run is above 1,000), and a business with two attempts used about 0.03 compute units in our tests. A run waiting between retries still uses about 0.25 compute units an hour at 256 MB.

Big lists and memory. The Actor holds your whole list in memory, including rows beyond Maximum calls this run. 256 MB covers about 2,000 Google Maps rows even with reviews included; 512 MB covers 5,000. If a dataset is too big for the run's memory, the Actor stops within a few hundred rows, before calling anyone, and tells you which Memory to set under Run options.

What you need to place calls

Accept the PlaceCall terms in the field at the top of the form (API: acceptTerms, with the exact text of the option). By placing calls you agree to the PlaceCall API Terms and Privacy Policy. When the terms change, the option changes, and runs (including saved Tasks and Schedules) ask you to accept again before any call.

Limits worth knowing before a big run

• Per Apify account per day: up to 500 dial attempts on a paid Apify plan, 30 on the free plan. Retries and unanswered calls count as attempts. The count resets at midnight UTC (5pm Pacific in summer, 4pm in winter). When it is reached, new calls stop and the rest come back as notAttempted; run again after the reset with Only call businesses not reached yet.
• US destination numbers only. English is the most reliable language.
• A call takes 60-90 seconds; the run log estimates how long your list will take and warns when it will not fit the run's Timeout.

Writing good questions

Two or three short, factual questions per run. The agent asks one at a time and a long opening is the main reason people hang up.

Ask for facts a person at the counter would know. "Are you open right now" works; "what is your refund policy for orders placed through third-party delivery apps" does not.

Nobody is watching an Actor run, so if a business asks something your questions did not cover, there is no one to answer. Put any facts it might need into Answer if a call asks a question. If that field is empty, the agent is told at once to carry on without the answer rather than leaving the business on hold. Every row reports what it was asked in questionsAskedMidCall, so you can fold that into the next run.

Good manners, enforced

Every call opens by identifying PlaceCall and stating that the line is recorded. Asked directly whether they are speaking to a person, the agent answers honestly and never claims to be human. Calls are recorded; recordings and transcripts are kept for 90 days.

Only call businesses you have a legitimate reason to call, and only on behalf of yourself or an organization you represent. These are real phones answered by real people who did not ask to hear from you.

Support

https://github.com/voygr-tech/placecall