Stop spam calls using up your AI answering minutes by blocking numbers before the AI answers, not after. Use your receptionist's spam list, mark and block repeat offenders from the call log, check how your plan counts calls (per call, per minute or per unique caller), and claim any spam credit your vendor offers. Smith.ai, for one, excludes its known-spam numbers from the quota.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe catch is that no filter is complete. A robocall that slips past gets answered, and on most plans an answered call is a billable call until you do something about it. So the work has two halves: a front filter that stops most junk, and a short routine that catches the rest.
Which calls use up your AI receptionist's allowance?
Which calls use up the allowance depends on the vendor: Smith.ai counts real calls its AI answers, while Goodcall counts unique phone numbers and says spam does not count. Before you fight spam, find out what your own plan meters, because a robocall costs you nothing on one model and a full call on another. Here is what two vendors say on their own pricing pages at the time of writing (November 2026).
| Vendor and plan | What is counted | What happens to spam |
|---|---|---|
| Smith.ai AI Receptionist (Free, Pro, Enterprise) | Real calls answered by the AI. Free: 25 a month then $3 each. Pro: tiers of 75, 150 or 300 calls with extra calls at $2.50, $2.30 or $2.17 | Calls from its list of more than 20 million known spam numbers are filtered and not counted. Calls the AI does answer count by default. Customers can remove up to 10 percent of a cycle's calls for selected spam |
| Goodcall (Starter $79, Growth $129, Scale $299) | Unique customers: one phone number counted once a month. Starter includes 100, Growth 500, Scale 2,000. Minutes and calls are not metered | Spam and wrong numbers are stated not to count |
| A plan that bills per minute (several do) | Every minute of talk time, including a robocall that plays a recording to your AI | Nothing is excluded unless the vendor says so in writing |
Two things in that table matter more than the headline prices. First, Smith.ai separates calls "filtered before answering" from calls "answered", and only the first group is free. Second, Goodcall's unique-customer count means a robocaller who dials you forty times in a month is still one number, and if the vendor treats it as spam it is zero.
The third row is a reminder rather than a vendor claim: I have not named any particular per-minute product there, because pricing models change. If your plan is metered in minutes, read the terms for the words "spam", "unwanted" and "test calls" before you sign, and keep the reply you get. If you are still choosing, the comparison of an AI receptionist and an answering service shows how per-call and per-minute models change the monthly bill.
A worked sum for a coworking space
Here is an illustration with made-up numbers. A coworking space gives its front-desk number to an AI receptionist on a plan with 150 calls included and extra calls at $2.30 each (the middle Pro tier on Smith.ai's page). In one month the phone gets 130 genuine calls: tour requests, day-pass questions, a few members locked out. It also gets 40 junk calls that the vendor's filter let through, so the AI answered them.
That is 170 answered calls against an allowance of 150. With nothing done, the bill is 20 extra calls at $2.30, which is $46 on top of the monthly fee. Now apply the credit the help centre describes: up to 10 percent of the 150-call quota, so 15 calls, can be removed for spam and test calls combined, provided you mark them and block the number.
After the credit, the count is 155 and the overage is 5 calls, $11.50. The remaining 25 junk calls cost nothing extra in credit terms because the allowance is used up, but blocking those numbers means they cannot repeat in later months. If the same 25 numbers each called again next month, that would be a recurring leak of up to $57.50 (25 calls at $2.30) before the next credit.
| Step | Answered calls | Calls over 150 | Overage at $2.30 |
|---|---|---|---|
| Nothing done | 170 | 20 | $46.00 |
| 15 spam calls marked and removed | 155 | 5 | $11.50 |
| Next month, 25 repeat numbers already blocked | 130 plus new junk | 0 if new junk stays under 20 | $0 to a few dollars |
The pattern to take from this: the credit is a ceiling, not an entitlement to unlimited free junk. It covers a tenth of your allowance, so it rescues a bad month and does nothing for a structural problem. The blocking is what fixes the structure.
If you would rather run the allowance sums and call-log sorting in a spreadsheet yourself, try The Complete Excel Course with AI Integration on Coding Liquids.
How do you mark and block a spam call in Smith.ai?
Smith.ai's help centre says to open the call's details page in the Call Log, click More options and choose Mark as spam and block, which blocks the number straight away. The article describes the routine for paid plans. I am giving the steps as it lists them, and I would still do the first one on a test call in your own account before you trust it with a busy month.
- Open the call's details page from the Call Log (under Call History) in your dashboard, or from the link in the call summary email.
- Click More options.
- Choose Mark as spam and block and confirm.
The number is blocked straight away, so later calls from it do not reach your AI receptionist. The help article lists some limits that catch people out:
- You can block as many numbers as you like, but only spam calls inside the 10 percent allowance come off the bill. Its example: a plan with 150 calls a cycle allows up to 15 removed calls.
- Once you pass the allowance, the call that took you over it stays billable. Future calls from that number are blocked and not billed.
- After you hit the limit the Block caller button still works, but extra calls are not credited, and the override toggles stay disabled until the next cycle.
- Marking a call as spam or test cannot be undone, so check the call first.
- Filtering a caller without blocking does not qualify the call for removal.
The article does not say whether Free plan users can mark spam or earn a credit. Smith.ai's own page is also silent on it, so if you are on the free tier with 25 calls, ask support before assuming anything. I have not confirmed it, and I would not budget on it.
Which filters stop robocalls before the AI picks up?
Three layers can stop a robocall before the receptionist answers: the vendor's own spam list, your phone provider's call blocking and the way the AI greets callers. The cheapest robocall is the one that never reaches the receptionist, and you control the second and third layers.
The vendor's own spam list
Smith.ai says it filters calls from more than 20 million known spam numbers and does not count them. Goodcall says spam does not count towards unique customers. You do not set either one up; you only need to know it exists, and to remember it is a list. Numbers that are new, rotating or imitating local numbers can get through.
Your phone provider or business line
Many business phone systems and carriers offer call labelling, anonymous-caller rejection or a personal blocklist. Which ones you have depends entirely on your provider, so I will not promise a feature. The practical move is to ask your provider three questions in writing: can you block calls with no caller ID, can you add a list of blocked numbers at the line level, and does a forwarded call keep the original caller's number when it reaches the AI? That last one matters because a blocklist on the AI side only works if the receptionist sees the real number.
How the AI answers
Some robocalls hang up the moment they hear a voice that is not a human greeting. A short greeting that asks a question, for example "Which of our desk or meeting-room services are you calling about?", gives the AI something to listen for. A recording playing at the other end will not answer. Check whether the call ends quickly and whether your vendor treats very short calls as billable; some do, and the terms will tell you. If you are writing the greeting, the tutorial on AI call disclosure wording covers the first line a caller should hear.
How do you spot junk calls in the call log?
Spam in an AI receptionist's call log clusters at the extremes: calls of two to eight seconds, calls of exactly the same length, and calls whose AI summary says "no message left" or "silence". You do not need a data analyst. A month of calls fits on one screen if you sort by duration.
Here is an illustrative extract from a coworking space's call log, with numbers shortened. The reasoning in the last column is what you are doing in your head as you scan.
| Caller | Length | AI summary | Verdict |
|---|---|---|---|
| +1 555 010 0147 | 0:04 | No speech detected | Spam: block |
| +1 555 010 0148 | 0:05 | No speech detected | Spam: same prefix, block |
| +1 555 010 0212 | 2:41 | Asked about a four-desk team room next month | Real: keep |
| Withheld | 0:19 | Recorded message about an extended warranty | Spam: block if you can |
| +1 555 010 0390 | 1:12 | Member locked out of the building | Real: keep |
| +1 555 010 0149 | 0:03 | No speech detected | Spam: same prefix, block |
The first, second and last rows share a prefix and a length, which is the signature of an auto-dialler walking through a number block. You block each number you can, and you note the prefix. If your phone provider can block a range, that is the better place for it; if it can only block numbers one by one, the AI-side block is your only tool.
Using AI to read the log (a prompt you can paste)
If you get more than a hundred calls a month, an assistant such as ChatGPT or Claude can do the first pass. Export the log as a CSV and paste only the columns you need, which are number, duration and summary. Remove anything that would identify a customer first, and check your vendor's terms and your own privacy policy before putting call data into any third-party tool.
I run a coworking space and my AI receptionist answers our main number.
Below is last month's call log (number, duration in seconds, AI summary).
1. List numbers that look like robocalls or spam, with the reason for each
(very short, no speech, same prefix as another number, recorded message).
2. Group them by shared prefix.
3. List any calls you are unsure about and say what would settle it.
4. Count answered calls, spam calls and genuine calls.
Do not guess: if a summary is missing, say so. Ask me questions if anything is unclear.
[paste the log here]
A realistic reply, labelled illustrative, looks like this: "Likely spam: 31 numbers, 24 of them under 6 seconds with no speech. Shared prefix: 11 numbers start +1 555 010 01. Unsure: 3 calls of about 15 seconds whose summary is only 'wrong number'. Counts: 174 answered, 38 spam, 129 genuine, 7 unsure." What you fix: the assistant cannot know that "wrong number" calls are billed by some vendors and not others, and it cannot know which of the three unsure calls is the member who dials from a different phone. You make those calls. The assistant narrows two hundred rows to a dozen.
Why can a spam block stop a real customer?
A spam rule built on one sign, such as call length, can block a real customer. A catering company I will describe as an illustration, rather than a real client, gave its AI receptionist a blocklist and marked every call of under ten seconds as spam. A week later a regular customer complained that the phone "just hung up". The cause was a loose connection on her mobile that dropped her calls after a few seconds; she rang back five times, and the fifth number was blocked. Blocking cannot be reversed in some systems, so the damage was real.
The fix was a rule, not a tool. Only block a number when two signs agree: very short and no speech, or a recorded message, or a prefix cluster. One sign alone goes into a "watch" list for a week. A number that rings repeatedly in short bursts and then connects once is more likely a poor signal than a robot. Before you block any number you do not recognise, spend ten seconds searching it, and check your customer records.
What is a ten-minute monthly routine for catching spam calls?
Run a ten-minute spam check in the last three days of each billing cycle, because a credit allowance is still available to use then.
- Open the call log and sort by duration, shortest first.
- Mark and block the clear spam: no speech, recorded messages, shared prefixes. Stop when you reach the allowance if your vendor has one; remember it covers test calls too.
- Look at the number of answered calls against your allowance. If you are within 10 percent of the limit, check your forecast for the next cycle before the reset.
- Check your invoice or usage page the day it arrives and compare it with the log. If a spam call you removed still shows as billed, send the vendor the call ID.
- Write the spam share in a note: spam calls divided by answered calls. If it passes 15 percent two months running, talk to your phone provider about line-level blocking, or to the vendor about a plan that counts unique callers.
That last ratio is the useful one. A single bad month is noise; a persistent share tells you which layer is failing. For a wider view of whether the receptionist is paying for itself once the junk is gone, see how to measure an AI receptionist's return in 90 days.
When does counting unique callers beat counting calls?
Counting unique callers beats counting calls when the same few clients ring repeatedly. Take a translation agency whose main number is rung repeatedly by the same dozen clients chasing delivery dates, plus a steady drizzle of robocalls. On a per-call plan every one of those chasing calls counts. On a plan that counts unique phone numbers, as Goodcall's does, a client who rings six times in a month is one customer.
Illustrative sum: 120 calls from 40 different numbers in a month, plus 30 robocalls from 30 different numbers. A per-call meter sees 150 calls. A unique-number meter sees 40 clients and, if the vendor excludes spam as Goodcall states, nothing for the 30 robocalls. Goodcall's Starter plan includes 100 unique customers and the next one up includes 500, so that firm sits well under the limit on either.
The reverse also happens. A venue that gets one-off enquiries from new people every day has no repeat callers to merge, so unique counting saves almost nothing, and each new number is a new customer. That is why I would not switch plans on a hunch. Take your own last month, count calls and unique numbers, and see how far apart they are. If they are close, spam handling matters more than the counting rule.
Whichever model you are on, check whether a plan's price per extra unit changes as you grow; Goodcall's page shows overage of 79 cents, 26 cents and 15 cents per extra customer across its three published tiers, and Smith.ai's extra-call rate falls as the included tier rises. The tier where you sit matters as much as the spam share.
What should you ask an AI receptionist vendor in writing?
Email the vendor five questions about spam, billing and blocking, so you have a reply to point to later if its page leaves one open. The five below cover nearly everything about spam and allowances:
- Is a call that is hung up after a few seconds counted or billed?
- Are calls to the AI that you later mark as spam credited, and is there a cap?
- Do blocked numbers still appear in the call log, and are they billed?
- If I forward calls from my main line, does the receptionist see the original caller number so my blocklist works?
- Does the free or entry plan get the same credit as paid plans?
Keep the answers with the contract. If you want help comparing what several vendors say on these five points for your own line, that is the kind of job a 1:1 call on my AI implementation consultation covers. Most owners can do it alone with the list above.
What this looks like for an event venue and a caterer
An event venue's main number is on its website and in search listings, so it attracts both robocalls and a real mix of enquiries on weekends. Its highest-value calls come in clusters after an advertising push or a wedding fair. The risk there is not the average month but the busy one: a campaign pushes genuine calls past the allowance just as a robocall burst arrives, and the 10 percent credit is spent on spam when you wanted headroom for enquiries. For a venue I would raise the included tier before the season and clean the blocklist the week before.
A catering company has the opposite shape: a few large daytime orders, many repeat customers and a spike around holidays. Repeat callers make unique counting attractive, and a stable set of known customer numbers makes a simple allow-list sensible. Save the customers' numbers in your phone system and mark them as known if your provider offers it. Then any unknown number that rings and hangs up inside five seconds is far easier to classify.
If you want to see the finished calls in one place for review, a simple automation can drop each call summary into a sheet. The tutorial on sending AI receptionist summaries to a sheet with Zapier shows the pattern for one vendor, and the same idea works for others.
What stays unsolved
You will not get to zero. Spoofed numbers that mimic local ones, number rotation and rare real calls from odd numbers mean some junk always gets through, and a few real callers will occasionally be treated as junk. The goal is a spam share you can see and a routine that keeps it small. A call log that you read once a month, a blocklist built from two agreeing signs, and a plan whose counting rule matches your call pattern together do most of the work.
The facts here were checked against the vendors' own pages in November 2026. Prices, allowances and credit rules change, so read the current pricing and help pages before you decide. If a vendor's terms read differently from what I have written, trust the vendor's page and your written reply from their support.
Spam calls and AI answering allowances: questions owners ask
Do spam calls count against my AI receptionist plan?
It depends on the vendor's counting rule. Smith.ai does not count calls from its known spam list, but a spam call its AI actually answers counts by default, with up to 10 percent of a billing cycle removable. Goodcall counts unique phone numbers per month and says spam and wrong numbers do not count.
Can I get credit for robocalls my AI receptionist answered?
Sometimes, depending on the vendor. Smith.ai's help centre describes a monthly allowance: on paid plans you can remove up to 10 percent of your call quota from the bill by marking spam or test calls and blocking the number. Other vendors differ, so ask support in writing what they credit and how you claim it.
Will blocking a number stop all robocalls from it?
Only that exact number. Robocallers often rotate numbers or spoof ones that look local, so a blocklist trims repeat offenders but never finishes the job. Pair it with vendor-side spam filtering and a monthly check of the call log for new patterns.
Should I switch to a plan that counts unique callers?
It can help if repeat contacts, not spam, drive your volume, because each phone number is counted once a month. Run last month's call log through both pricing models before you move. If most of your calls come from different numbers, the saving is small.
How often should I review my call log for spam?
Weekly for the first month after launch, then once a month. Sort by shortest duration first, because silent calls and hang-ups under ten seconds are where most junk sits. Block what you find the same day so it cannot repeat inside the billing cycle.
Why does my AI receptionist answer robocalls at all?
Because no spam filter is complete. Spoofed numbers that mimic local ones, rotating numbers and new numbers can get past a vendor's list, and on some plans a spam call the AI answers counts towards your quota by default. A monthly check of the call log catches what slips through.
How do I avoid blocking a real customer by mistake?
Block a number only when two signs agree: very short and no speech, a recorded message, or a prefix cluster. One sign alone goes on a watch list for a week. Before you block any number you do not recognise, spend ten seconds searching it and check your customer records.
Further reads
- What Is an AI Receptionist and How Does It Handle Bookings? — The basics of how an AI receptionist handles a call.
- My AI Front Desk Pricing: 200 Minutes and Overage Costs — Another receptionist's allowance and overage rules in detail.
- What to Do When an AI Receptionist Gets a Booking Wrong — Fix the other costly failure: a wrong booking.
- AI Receptionist vs Front-Desk Hire: The Real Cost for a Salon — See per-call costs next to a front-desk hire.
- What Does an AI Receptionist Cost a Dental Practice in 2026? — Worked monthly costs for one call-heavy practice.
- Best AI Receptionists for Salons, Compared by Price — Receptionist plans compared by price.
Show 3 more related tutorials
- How Much Does an AI Answering Service Cost a Contractor? — Smith.ai, Dialzara, Goodcall, Rosie and Jobber Receptionist priced at 100 calls a month, with the call-length sums and the plan limits that change the bill.
- AI Receptionist for Junk Removal: Costs, Photo Requests and Booking — Plan prices at three call volumes, nine intake questions, a text-for-photos flow and same-day booking rules for a junk removal company.
- How to Test an AI Voice Agent for Plumbers Before Real Calls — Fifteen scripted test calls for a plumbing AI voice agent, from a burst pipe to a wrong number, with pass and fail rules, a scoring sheet and a fix list.
Browse all AI-for-business tutorials
Sources: Smith.ai AI Receptionist pricing page and its help-centre article on managing spam callers; Goodcall pricing page (checked 17 Nov 2026).
