WhatsApp recruiting chatbot: what it should do, what it must never say, and what it costs

You post a warehouse associate job on Monday at nine. By eleven there are sixty messages in the agency’s WhatsApp, and nearly all of them say the same thing: “info?”, “how much does it pay?”, “where is it?”, “is there a night shift?”. The recruiter answers the ones she can. The rest hear back at four in the afternoon, or the next day. Someone asking about an hourly job is asking two other agencies at the same time, and goes with whoever answers first.

That is the problem a WhatsApp recruiting chatbot solves, and it is why there are more and more ads offering one. This guide is for telling apart one that works from one that only replies “Hi! 👋 Thanks for your interest”, and for understanding what it really costs.

What it has to do

It is not a button menu or a form. For an agency hiring hourly workers, the assistant has six jobs, and if it is missing one, the recruiter ends up doing it by hand.

1. Answer with the job’s real information. Pay, location, shift, benefits, whether there is transportation. Exactly as the agency entered it, without rounding or dressing it up. If the job says “$17.50 an hour, weekly pay”, the assistant says that, not “competitive pay”.
2. Screen against the requirements, in the conversation. Availability for the shift, experience, whether they can lift what the job needs, whether they can get to the site. It asks naturally, in whatever order fits, and decides whether the person qualifies. Those who don’t are told politely and logged; those who do move to the next step.
3. Book the interview. In the slots the agency opened, confirming the place and what to bring. And send a reminder the day before, because in this line of work half of missed interviews are forgotten, not abandoned.
4. Hand the chat to a person when it should. If the candidate asks for someone, asks something sensitive or the conversation stalls, the assistant steps aside and the recruiter takes the chat with the full history, not from scratch.
5. Log what it couldn’t answer. “Can I start with my ID still being renewed?” isn’t in the job post. The assistant doesn’t make it up: it notes the question, tells the candidate the agency will confirm, and once the recruiter answers it, the answer is learned for the next person who asks.
6. Leave the agency a report it can use. How many wrote in, how many qualify, how many were booked and how many showed up. By job and by client company.

What it must never say

This is what separates an assistant the agency can leave alone from one that needs watching. The first three are mistakes we saw in real operation and had to fix.

• It doesn’t reveal the screening criteria. If the job requires a forklift certification and the candidate says they don’t have one, the assistant doesn’t explain “the client requires forklift certification”. A minute later the same person writes again from another number, certified. The assistant thanks them, says this job isn’t a fit, and saves the reason for the recruiter.
• It doesn’t contradict the agency. If the candidate asked where the interview is, the recruiter answered “Building B entrance, 5:40 a.m.” and the candidate asks again, the assistant repeats the official answer. It doesn’t say “that question is still pending” when it has already been answered.
• It doesn’t invent. Not the pay, not the benefits, not whether there is a break room. Whatever isn’t in the job post goes to the agency. An assistant that “fills in” with what is typical for the industry produces candidates who show up with the wrong expectations, and that costs the agency its client’s trust.
• It doesn’t ask what the law won’t let you ask. Age beyond confirming the legal minimum, national origin, religion, disability, pregnancy, marital status. Screening questions have to be about the job. A good assistant flags those criteria when the agency loads a job and never turns them into a question for the candidate, because a recruiter in a hurry will eventually write “young and energetic” in a job post.
• It doesn’t make the hiring decision. It screens against requirements the agency defined. People decide who gets the job, and it should say so if asked.
• It doesn’t mass-message. Writing again to someone who wasn’t hired is useful when a similar job opens, with their permission and one at a time. An assistant that blasts promotions to the whole database is the fastest way to get the agency’s number restricted by Meta.

What it really costs

There are two costs behind every candidate you serve, and it pays to understand them before comparing provider prices.

What Meta charges for WhatsApp

Meta charges per delivered message, by type of message and by the country code of the candidate’s phone. Since October 1, 2026 it charges for replies again, with the first 1,000 service messages per month free for each business number. These are Meta’s rates for US and Canadian numbers (its “North America” market):

A typical candidate gets four or five messages from the assistant, so Meta’s part comes to less than two cents per candidate. If you hire in other countries, the rate is the one for the candidate’s country code, and Meta publishes them all in its rate card (see the sources below). And one detail almost nobody uses: if the candidate came from a click-to-WhatsApp ad, Meta opens a window of up to 7 days in which every message is free, templates included.

What the AI costs

This is the cost that matters. Every candidate message goes through a language model that reads the job, the history and decides what to answer; in our operation, a complete candidate, from “info?” to the booked interview, costs less than a dollar in model usage. That is why serious pricing is per person served, not per job or per user: it is what actually costs money.

How to read “plans from $49 a month”

You will see ads like that. Before booking the demo, four questions clear up the price:

• What does it include? How many conversations or candidates a month, how many users, how many numbers. “From” means the plan in the ad is the smallest one.
• Are Meta’s messages included, or billed on top? And if they are billed on top, with or without a markup.
• What does setup cost? It is almost always separate and comes up on the call. Ask before.
• Do I keep my number? If the answer is “we give you a new one”, the candidates and clients who already have your WhatsApp will keep writing to the old one.

For reference, FILTRA’s prices are published, per person served, with unlimited jobs and a one-time setup that we do for you.

How to test one in ten minutes

Any serious provider gives you a demo number. Write to it as if you were a candidate, ideally at ten at night, and do these five things:

1. Ask something that isn’t in the job post: “is there parking?”, “can I start next week?”. A good assistant says it will confirm; a bad one makes something up.
2. Say you don’t meet a requirement. Watch whether it explains the filter or simply tells you, politely, that the job isn’t a fit.
3. Ask to talk to a person. It should hand you over without a fight, and the person should see what you already said.
4. Book the interview and then move it. If it reschedules without trouble, the calendar is real.
5. The next day, write again from the same number. If it greets you like a stranger, it keeps no history.

What you can’t test in ten minutes is what happens when it gets something wrong. Ask how the agency finds out about a mistake and who fixes it. If the answer is “it doesn’t make mistakes”, you have your answer.