Which ATS has a built-in assistant?

Most ATS platforms now ship some kind of assistant, so the useful question is not which one has it but which one has an assistant that actually does the admin. The test that separates them: can it act on your live candidate and job data, with your approval, from the screen you are already on? Recruitifly’s assistant, Fly, is one published example of the built-in kind: it drafts outreach, schedules interviews, updates stages and builds shortlists on every tier and every screen, always proposing the change for you to confirm. A chat window that only answers questions is a search box with better manners.

Built-in or bolted-on: how do you tell the difference?

A bolted-on assistant lives somewhere else: a separate chatbot, a browser extension, a generic writing tool you paste candidate details into. It can produce a decent email, but it does not know your pipeline, it cannot see that the candidate already declined twice, and everything it writes you carry back by hand. Pasting candidate data into an external tool also creates a GDPR exposure that is now yours to manage.

A built-in assistant reads the records it acts on. Ask it to follow up with everyone stuck in screening for more than a week and it knows exactly who that is. The draft lands on the candidate record, the stage change happens on the real board, and every action is logged. Most major vendors, Workable, Greenhouse, Bullhorn and Teamtailor among them, now advertise assistants or copilots, and the marketing language is nearly identical, so run three checks on live data instead of reading feature pages:

1. Ask a question only your data can answer: “which candidates for this role have heard nothing in ten days?”
2. Ask it to do something, not just say something: “schedule the interview”, “move her to offer”.
3. Check where the output lands: on the record, or in a chat transcript you still have to act on.

Which jobs should the assistant handle?

Make the vendor demonstrate each of these on live data before you believe an “AI-powered” label:

• Draft outreach and follow-ups from the candidate record, in your tone, ready to approve and send.
• Schedule interviews: propose slots, invite the panel, absorb the inevitable reschedule.
• Update pipeline stages by instruction, “move these three to interview”, instead of dragging cards.
• Summarize a candidate from CV, notes and message history into two readable paragraphs before a call.
• Build shortlists: rank candidates against the role and compare the front-runners side by side.
• Answer reporting questions in plain language, “how many offers did we send this quarter?”, without a spreadsheet export.

Four or more of these and the assistant replaces real admin hours. Only the first, and you have bought a writing tool with a login. For the wider map of which recruiting admin can be handed off entirely, see what recruitment work can actually be automated.

Propose and confirm, or fire and forget?

Once an assistant can write to your data, the control model becomes the most important line on the spec sheet. Fire and forget executes immediately: you ask, it sends. Propose and confirm drafts the action and waits for a named human to approve before anything reaches a candidate or changes a record.

Dimension	Propose and confirm	Fire and forget
Speed	One extra click per action	Fastest possible
Risk of a bad send	Low: you see it before the candidate does	Real: mistakes reach candidates directly
Audit trail	Every change has a named approver	Only as good as the vendor’s logging
Accountability	Clearly yours, with evidence	Murky when something goes wrong
Best for	Anything candidate-facing or record-changing	Read-only lookups and summaries

Approval-first is safer for a structural reason: recruiting mistakes are public. A mislabeled stage is recoverable; outreach sent to the wrong shortlist is not. The sensible split is to let the assistant answer and summarize freely, and to require confirmation for anything that sends or changes. That is how Fly works in Recruitifly: every write is proposed first, you confirm or cancel, and the confirmation is logged with your name on it. No vendor should sell you fully autonomous hiring, and you should not buy it.

Where does an assistant save a small team the most time?

Three places, in this order:

1. Follow-ups. The polite chase, candidates waiting on feedback, hiring managers sitting on CVs, eats more recruiter hours than any single task. An assistant that watches the pipeline and drafts the nudges turns an hour of remembering into five minutes of approving.
2. Scheduling. Slot-finding, panel coordination and reschedules are pure coordination cost. We cover what changes when this happens inside the system in interview scheduling inside your ATS.
3. Reporting. “How is the pipeline for the finance role?” should be a question you ask, not a Friday afternoon of exports.

For a one to three person desk, these add up to several hours a week. Be honest about the boundary, though: an assistant compresses admin. It does not take the intake call, read the room in an interview, or close a wavering candidate. Buy it to free those hours, not to replace them.

What does assistant usage cost?

Three pricing patterns dominate:

• Bundled per seat: the assistant is included in the seat price. Predictable, though deeper capabilities often sit in higher tiers.
• AI add-on: a separate monthly fee per user to switch the assistant on. The sticker is not the bill; always price tool plus add-on.
• Usage-based: you pay per unit of assistant work. Efficient if usage is spiky, provided the unit and the overflow prices are published.

Recruitifly publishes the third model with the assistant included everywhere: Fly is on every tier, and the metered unit is an Action, one task that produces work, like drafting an outreach message or parsing a CV. Freelancer is EUR 70 per month with 150 Actions, Pro is EUR 250 with 600, Agency is EUR 599 with 2,500, and Enterprise starts at EUR 2,500 with custom volume. Overflow packs cost EUR 30 for 250 or EUR 100 for 1,000 and never expire. The pricing page keeps the numbers current next to a calculator.

Whichever model you pick, get two answers in writing: exactly what counts as billable assistant usage, and what happens when you hit the cap mid-search.

The short version

Demand built-in rather than bolted-on, test the six jobs above on live data, prefer propose and confirm for anything that touches a candidate, and price your busiest month rather than the sticker. Recruitifly was built around exactly this model and is currently in private beta; talk to us if you want to put Fly through the checklist on one of your real roles.