Written by the ChatterMate team — we build an open-source AI support agent.
Picture a staffing branch at 7:40 on a Friday morning. The phone rings. "Did my timesheet go through?" It rings again. Same question. By nine, nobody on that team has recruited anyone. They've done support, and nobody planned for it.
That's the thing to understand about an AI chatbot for staffing agencies. Your inbox isn't mostly "I'd like to apply." It's "when do I get paid," "which site do I report to Monday," "did you get my right-to-work document," and "is my invoice approved." Candidates, temps on assignment, and client managers all ask them, all day, usually at the exact moment your team is busy placing someone.
So here's our position, stated early. Point the bot at those questions. Keep it out of the hiring decision. The rest of this post is why we think that split is both the better product and the safer one.
Where the volume actually comes from
Agencies talk about candidate experience as a sourcing problem. A lot of it is a silence problem.
Pin analysed more than 200,000 active recruiter conversations between January 2024 and May 2026 for its Employer Ghosting Index and found that 72% of active candidate conversations stall for 30 days or more, with a median silence of 75 days (Pin, Employer Ghosting Index 2026). Greenhouse surveyed 2,500 workers across the US, UK and Germany in late 2024, and 61% reported being ghosted after an interview (Greenhouse State of Job Hunting report, as cited by Pin). And SHRM's 2025 talent trends report has 41% of employers saying candidates ghost them back (as cited by Pin).
Silence is expensive in both directions.
Notice what those numbers describe, though. They describe a status gap. Someone applied, interviewed or got placed, then heard nothing. A person who hears nothing does the rational thing and asks again, by email, by text, by phone, by calling the branch. Each of those is a ticket you never wrote down.
The same shape shows up in other businesses that live on a back-office pipeline. Our MSP guide covers a different industry with the identical problem: the answer exists in a system (a PSA board there, an ATS and payroll platform here), but the person asking can't see it, so they ask a human.
The three audiences, and what each one asks
Staffing is unusual because you serve three groups at once, and their questions barely overlap.
Candidates
Mostly about process and paperwork. What documents do I need for onboarding? Is the role still open? How long does a background check take? Who is my consultant? These are in your onboarding pack, your FAQ page, your compliance checklist. A bot grounded in those documents answers in seconds, at 9pm, with a citation to the paragraph it used.
Temps and contractors on assignment
This is where the Friday-morning phone storm lives. Timesheet cutoffs, pay dates, holiday pay rules, how to report an absence, what to do when a site manager has not signed off. The policy answers live in documents. The personal answers ("has my timesheet been approved") live in a system, and we'll come back to that, because it's where most agencies get hurt.
Clients
Hiring managers want to know where their open requisitions stand, how invoicing works, what your notice terms are, how to extend a contractor. Fewer messages than from temps, but higher stakes per message. And they notice when you're slow.
The case against letting the bot screen candidates
This is the opinionated part. Plenty of vendors will sell you a recruiting chatbot that asks screening questions, scores the answers and ranks candidates. We think agencies should be cautious about that, and not only on ethics.
Start with the law. New York City's Local Law 144 bars employers and employment agencies from using an automated employment decision tool unless the tool has had a bias audit within the past year, the audit results are publicly available, and candidates have been given notice, which must go out 10 business days before use (NYC Department of Consumer and Worker Protection). Employment agencies are named in the law explicitly. In the EU, the AI Act treats AI used for recruitment and selection as high-risk, which brings its own documentation and oversight duties. We're not lawyers, and you should ask one. The point is that the moment a bot influences who gets seen, you've stepped into regulated territory.
Now the practical side. A screening bot's mistakes are invisible. If it wrongly rejects a good forklift driver or a locum nurse, nobody complains. They just never show up in your pipeline. A support bot's mistakes are loud: someone says "that's wrong" or asks for a person. You can measure the second kind and fix it. The first kind quietly shrinks your talent pool.
So we're deliberately boring about this. ChatterMate answers questions from the documents you give it and cites them. It doesn't rank people, and we wouldn't build one that does for you. If you want a screening tool, buy a screening tool, run the bias audit, and keep it separate from your support channel.
What to feed it
The quality of an AI chatbot for staffing agencies is mostly a function of what's in its knowledge base. Roughly in order of payoff:
- Pay and timesheet policy. Cutoff times, pay calendar, what happens when a timesheet is late, holiday and overtime rules by contract type. Highest-volume content you own.
- Onboarding checklist. Right-to-work documents, forms, background-check timelines, uniform or PPE rules.
- Client-facing terms. Notice periods, conversion fees, how to extend or end an assignment, how invoicing works.
- Branch details. Hours, addresses, who to call for what.
Don't feed it your internal margin sheets, your client rate cards, or anything you wouldn't hand to a candidate on paper. Scope matters. A bot that can see a client's bill rate can be talked into repeating it. We wrote about that failure mode in the prompt injection post, and the fix is mostly "don't put it there."
Our RAG guide for customer support goes into how retrieval keeps answers tied to your documents rather than the model's imagination.
The personal-data trap
"Has my timesheet been approved?" is a different kind of question from "when is the pay cutoff?"
The first needs live data about one specific person. The second is policy. A document-grounded bot handles the second perfectly and the first not at all, and it shouldn't pretend otherwise. If the bot guesses at someone's pay status, you've built a liability.
The sane design: the bot answers the policy question, recognises the personal one, and hands it to a person with the context attached, so the temp doesn't repeat themselves. That handoff is the single most important feature in this whole use case. We covered what makes one work (and what makes people hate it) in our human handoff guide. Short version: pass the transcript, say what happens next, and never make someone re-explain a payroll problem to a stranger.
Also worth a decision up front: where does this data go? Staffing agencies hold passports, national insurance or social security numbers, bank details. If that makes your compliance person twitch about sending chat transcripts to a third party, self-hosting is on the table. ChatterMate is open source and self-hostable, and our self-hosted AI chatbot guide walks through what that involves.
A realistic first month
We'd rather you start small than buy a transformation project.
Week one, pull the last two months of inbound emails and texts to your branch inbox and group them by question. You'll probably find that five or six questions account for most of it. Don't take our word on that; count.
Week two, write the answers down properly if they only live in one coordinator's head. This step finds more problems than the bot does. Don't be surprised to find two or three different versions of the holiday pay rule in circulation.
Week three, load the documents, put the widget on your candidate portal and your temp login page only, and turn on handoff to a named human inbox.
Week four, read every conversation. All of them. Mark the ones that went wrong. Fix the document, not the bot.
Then decide whether to widen it. There's a longer treatment of measuring whether it's working in our piece on the confidence score, which is how you tell the bot when to say "I'm not sure, let me get someone."
What this won't fix
Be honest with yourself here. A chatbot will not repair the 75-day silence if the silence is caused by hiring managers sitting on feedback. It answers questions about status; it can't create status that doesn't exist. If a candidate asks "where do I stand" and the truthful answer is "we're waiting on the client," the best bot in the world says exactly that. That's still better than no reply. But it isn't a pipeline fix.
It also won't replace the recruiter relationship. Placing a nurse or an electrician is a trust business. We think the bot's job is to clear the 7:40am timesheet calls so your people have the hours back for the conversations that need a person.
Try it on your own documents
ChatterMate is open source and free to start: the Free plan includes 50 AI replies a month, and you can self-host it. Load your pay policy and onboarding pack, ask it the ten questions your coordinators hate answering, and see what comes back. If an answer is wrong, you'll see which document it used. Start at chattermate.chat.



