How AI recruiting works when one agent runs the whole pipeline
AI recruiting works by turning a role brief into a scoring rubric, then sourcing, scoring, contacting and screening candidates against it. RecruitingBot runs those 11 steps for each open role and hands your hiring manager a ranked shortlist to approve.
- GDPR built in
- Human approves every shortlist
- No hiring fees
Eleven steps from role to shortlist and who does what at each one
-
01 A structured role brief
Role intake
You describe the opening in a short form: job title, location or remote setup, salary range, seniority, must haves, nice to haves and deal breakers. Every later step reads from this brief, so a sharp intake saves rework. Deal breakers such as work authorization or a minimum years of experience become hard filters, while nice to haves only lift a score.
Your partFill in the brief and confirm the salary range and deal breakers.
-
02 A job post plus an editable weighted rubric
Job description and rubric
From the brief, RecruitingBot drafts a job post and a scoring rubric. The rubric lists each criterion, whether it is a must have, a nice to have or a deal breaker, and how much weight it carries. You can rewrite any line, add a criterion or change a weight before anything is sourced, and the post can go live on a hosted apply page.
Your partEdit and approve the rubric weights and the wording of the job post.
-
03 A deduplicated candidate pool with source and lawful basis per person
Sourcing
The agent pulls candidates from three places at once: public professional profiles and open web sources under a documented legitimate interest basis, your own talent database imported by CSV or from Greenhouse, Lever or Workable, and inbound applicants arriving through the hosted apply page. Each record keeps its source and lawful basis so you can always say where a name came from.
Your partChoose which sources are switched on for the role and exclude companies or people.
-
04 Per criterion scores with written evidence
CV and profile analysis
Every candidate is scored against the rubric one criterion at a time. Instead of a single opaque number, you see a score per criterion with the evidence behind it, such as the project, title or skill that earned the points. Candidates who hit a deal breaker drop out of the outreach list but stay visible with the reason, so nothing disappears silently.
Your partOpen any profile, read the reasoning and move a candidate up or out by hand.
-
05 A running email sequence per selected candidate
Outreach
The top matches receive a personalized multi step email sequence sent from your own connected Gmail or Outlook mailbox. Daily sending caps protect your domain, messages only go out between 8:00 and 20:00 in the recipient's local time, and every email carries the Article 14 GDPR notice plus a one click opt out link.
Your partReview the sequence copy and the sending cap before the first message leaves your mailbox.
-
06 Classified replies and answered questions
Replies
Incoming replies are classified as interested, not now, not interested or question. Interested candidates move to screening, not now replies get a polite note and a later reminder, and questions about the role are answered using only facts from the brief. Anything outside those facts, like a request to negotiate salary, is flagged for a person.
Your partTake over any thread at any moment and the agent stops writing to that person.
-
07 Screening answers with pass or fail on knockout questions
Screening
Interested candidates receive knockout and pre-screening questions you set for the role, such as notice period, salary expectation or right to work. Knockout answers that fail a hard requirement end the process for that role with a respectful message, while softer answers are recorded next to the candidate score for the hiring manager to weigh.
Your partWrite or approve the knockout questions and the answers that count as a fail.
-
08 An auto scored skills test result
Skills test
Candidates who pass screening get a short test built for the role, mixing multiple choice and short answer questions. Results are scored automatically and attached to the profile. The test is meant to add signal for the shortlist, not to replace an interview, so it stays brief enough that strong candidates are willing to finish it.
Your partReview the questions and replace any of them before the test is sent.
-
09 Booked interviews in the hiring manager calendar
Scheduling
When a candidate clears the test, they pick an interview slot directly from the hiring manager's connected Google Calendar or Microsoft 365 calendar. Only open slots are offered, the event lands in both calendars, and reminders go out before the call. No one trades five emails to find a Tuesday afternoon that works.
Your partSet the hours the hiring manager interviews and decline any booking.
-
10 A ranked, shareable shortlist with approve and reject buttons
Shortlist
The agent assembles a ranked shortlist with scores, evidence, screening answers and test results side by side. The hiring manager opens it through a shareable link and clicks approve or reject on each candidate. Nobody is hired, advanced to an offer or rejected for good by the software alone, because a person signs off on every shortlist.
Your partApprove or reject each shortlisted candidate.
-
11 Closed loops with candidates and a running retention clock
Follow ups
After decisions are made, candidates and the hiring manager get reminders, rejected candidates receive a courteous closing email, and the data retention clock starts. Sourced people who never replied are deleted automatically after 30 days by default, a window you can set anywhere from 30 to 180 days per workspace.
Your partSet the retention window and confirm the wording of rejection emails.
What one role run looks like from the inside
A role run is a single opening taken through every stage, from the brief you type to the shortlist your hiring manager signs off. The eleven steps above are not separate tools glued together. They share one rubric, one candidate record and one timeline, which is why a score you change in step two still explains a ranking in step ten. Each candidate carries their source, lawful basis, score breakdown, every email in both directions, screening answers and test result in one drawer.
That shared record is the practical difference between AI recruiting and a stack of point solutions. When a hiring manager asks why someone ranked third, the recruiter does not need to open a sourcing tool, an inbox and a spreadsheet. The evidence, the conversation and the test sit together, and the decision trail stays attached to the role until the retention clock removes it.
The first hour after you launch a role
Most of the first hour is setup, and it belongs to you. You fill in the intake, read the drafted job post, and spend real time on the rubric because it steers everything downstream. A good habit is to keep must haves to four or five items that a strong candidate would genuinely need on day one, and to push wish list items into nice to haves. Weights matter more than wording here.
Once you launch, sourcing starts immediately and the first scored candidates appear in the pipeline view within minutes, sorted by total score with the evidence visible on each card. This is the moment to sanity check the rubric against reality. If the top ten look wrong, adjust a weight or a deal breaker now, before any outreach goes out. Changes rescore the pool, so you are tuning the search, not starting over.
The first days of outreach and replies
Outreach runs on a schedule rather than in a burst. Your connected mailbox sends up to the daily cap you set, only inside the 8:00 to 20:00 window of each recipient, and follow up steps wait a set number of days between messages. That pacing looks like a careful recruiter at work, which is good for reply rates and good for the reputation of your sending domain.
Replies start to arrive over the following days. The agent sorts them, answers routine questions from the brief, and moves interested people into screening and then the skills test without waiting on you. Your daily job in this phase is short: skim the inbox view, take over any thread that needs a human touch, and watch interviews appear on the hiring manager calendar as candidates book themselves in.
Reading the example funnel correctly
Across the site you will see one example run drawn as a funnel: 500 people found, 80 selected for outreach, 25 replies, a round of screening, and 5 candidates on the shortlist. Treat those numbers as an illustration of how a funnel narrows, not as a measured average or a promise. Your own numbers depend on the role, the salary range, the location and how tight the rubric is.
What the example does show well is where the work concentrates. Reading 500 profiles is the part nobody enjoys, and it is exactly where scoring with evidence saves the most attention. Converting 80 contacts into 25 conversations is where personalization and timing matter. Moving from 25 conversations to 5 serious candidates is where screening questions and a short test earn their place.
| Stage in the example run | Example count | What drives the narrowing |
|---|---|---|
| Found through sourcing | 500 | Breadth of sources and how wide the location and seniority filters are |
| Selected for outreach | 80 | Rubric scores and deal breakers applied to every profile |
| Replied to outreach | 25 | Relevance of the role, personalization and sending times |
| Passed screening and test | Varies | Knockout questions and the skills test result |
| On the shortlist | 5 | Ranking by score and evidence, then human approval |
Where humans stay in control
The agent does the volume work, and people make the calls that carry consequences. You approve the rubric before sourcing, the outreach copy before sending, the knockout questions before screening and the test before it goes out. Any recruiter can take over a candidate conversation at any time, and the hiring manager approves or rejects every name on the shortlist. No final hiring decision is ever made by the software on its own.
This design is deliberate, and it is also how the product stays on the right side of Article 22 of the GDPR, which restricts decisions based solely on automated processing. If you want the full picture of notices, lawful basis and retention, read our page on GDPR compliant candidate sourcing. For how access, roles and audit logs work across a team, see recruiting software security.
What you need to connect before launch
Three connections cover almost every setup. A sending mailbox is required, because outreach goes from your own address rather than a shared server, so candidates reply to a real person on your team. A calendar for the hiring manager lets candidates book themselves into open slots. An existing talent pool is optional but useful, since past applicants are often the fastest route to a strong shortlist.
- Mailbox. Connect Gmail or Outlook and Microsoft 365 with OAuth, set a daily sending cap, and revoke access whenever you like.
- Calendar. Connect Google Calendar or a Microsoft 365 calendar for each hiring manager who interviews.
- Talent database. Import a CSV on any plan, or pull from Greenhouse, Lever or Workable on Growth and Agency.
- Apply page. Each role can publish a hosted apply page so inbound applicants join the same scored pool.
- Team. Invite recruiters and hiring managers as seats, each with a role that matches what they should see.
Limits, usage meters and what counts
Every plan has monthly allowances for candidates analyzed, contact reveals, outreach emails and skills tests, plus a cap on active roles at once. The dashboard shows a usage meter for each, and you see an upgrade prompt when any meter passes 80 percent. Allowances count consumption in the billing period, so deleting a role or a candidate does not hand quota back. Active roles is the one limit that measures what is open right now.
Starter fits a small team running up to three roles at once, Growth adds client workspaces and ATS import, and Agency adds single sign on, an audit log and a DPA download. Upgrading charges the difference immediately and raises your limits right away. The full breakdown of allowances by plan is on the AI recruiter pricing page, and there are no hiring fees on any plan.
| Monthly allowance | Starter | Growth | Agency |
|---|---|---|---|
| Active roles at once | 3 | 8 | 20 |
| Candidates analyzed | 1,500 | 5,000 | 12,000 |
| Contact reveals | 300 | 800 | 2,000 |
| Outreach emails | 600 | 2,000 | 5,000 |
| Skills tests | 50 | 200 | 600 |
Questions buyers ask
Anything else goes to [email protected] or the help panel inside your dashboard.
How long does it take before I see candidates?
Scored candidates appear in the pipeline within minutes of launching a role, because sourcing and analysis start right away. Replies take longer since outreach is paced by your daily sending cap and the recipient's local sending window. Interviews depend on how quickly candidates book, which is usually the slowest part and the one you control least.
Does the AI decide who gets hired?
No. It ranks candidates and explains each score with evidence, but a person approves or rejects every shortlisted candidate. Recruiters also approve the rubric, outreach copy, screening questions and test before they are used. The software never makes a final hiring decision without a human reviewing it first.
Can I change the rubric after the role is running?
Yes. Edit a criterion, a weight or a deal breaker at any point and the pool is rescored against the new version. Candidates already contacted keep their message history. It is worth getting the rubric right before outreach starts, since it decides who is contacted in the first place.
Do I still need my ATS?
Usually yes. RecruitingBot fills your pipeline with screened, interested candidates, while your applicant tracking system remains the place where offers and onboarding are managed. You can import past candidates from Greenhouse, Lever or Workable on Growth and Agency, or bring a CSV on any plan.
Which roles does this approach work best for?
It works best for knowledge work where skills show up in public profiles and CVs: engineering, data, product, design, sales, marketing, customer success, operations, finance, HR and support. Roles that depend on licenses checked in person, or on physical trials, still benefit from the outreach and scheduling steps.
Will candidates know an agent contacted them?
Messages go from your own mailbox with your team's signature and are written for the specific person and role. Every email includes the notice explaining who is contacting them, why, where their details came from and how to opt out in one click. A recruiter can step in on any thread whenever a candidate prefers to talk to a person.
Keep reading
Give it a role and see the shortlist it builds
Type a job title on the homepage. The demo runs instantly without an account and shows an example pipeline, job post and shortlist for that role.