← Help centre

Phone screens and the AI voice screen

Signed in? Ask Lovelio this question inside the app - it answers from this same page.

Phone screening is how a consultant checks an applicant by phone before putting them in front of a client. Lovelio supports two ways to run it: the consultant makes the call themselves using a question sheet inside the app, or Lovelio sends the candidate an AI voice screen, a short voice conversation with an AI agent that the candidate takes in their browser, which comes back scored. Both paths use the same question set and produce the same kind of score on the application, calculated the same way. Phone screening is included in the one Lovelio plan; without an active plan the phone screen options do not appear at all, and the actions refuse on every surface including Slack and the API.

The screen wizard

Clicking Screen on an applicant opens a short two-step wizard. Step one asks how to screen: Phone screen (the consultant makes the call, questions and scoring on screen) or AI screen (Lovelio interviews the candidate by voice and scores the answers). Step two asks where the questions come from: AI writes them fresh for this candidate (from the job, the AI assessment, and the client's DNA) or one of the forms attached to the Phone Screen row in Settings under Forms, copied verbatim - so different desks can keep different screens (tech, sales, temp) on that same row. The default from Settings is marked "(your default)" and sorts first, but the choice is made per screen, every time; there is no company-wide lock.

Every template row in the picker has a Preview button showing the actual questions with their answer type, weight (Critical, Important, or Nice to have) and required flag, before committing to it. The list also has a "Write a new one" row: it builds a brand new screening template right there, without going to Settings under Forms. The consultant speaks or types what the call must cover into one box (the microphone streams into the same text field, so typing works fine in a busy office), AI turns each point into a proper question with a proposed answer type, weight and required flag, and the consultant confirms every question one card at a time - keep it, change any of the three settings, edit the wording, or drop it. A review step shows the whole template so far and takes more questions through the same speak-or-type box, as many rounds as needed. Naming it saves it as a real screening template the whole team can pick from then on (editable later in Settings under Forms), and the screen runs from it immediately.

Backing out of a screen and opening it again asks both questions afresh, so a consultant can try the phone screen, close it, and come back for the AI screen instead. Picking the same source again reuses the questions already prepared, instantly; picking a different one rewrites them. The wizard is only skipped when there is work to protect: an answer already saved on the form, or an AI screen out with the candidate (sent, in progress, or back and scored). In those cases the screen opens on its current state and nothing is re-asked or wiped. Rewriting questions is refused outright once any answer has been saved.

The screening form (consultant makes the call)

  • With "AI writes them" chosen, the questions are specific to that candidate and job, 6 to 8 in four buckets: two alignment questions (notice period, salary, location, hours, work rights, whichever look likeliest to be deal-breakers), two to three resume probes (verifying claims, gaps, depth on skills the job needs), two job-fit questions tied to the role's actual requirements, and one motivation question. Lovelio's AI assessment of the candidate is in the mix too: a clear doubt it flagged becomes a question that tests it, but the assessment is treated as a first take to verify, never as fact.
  • Questions are written about the client the job is for, never about the agency; the candidate is being screened for the client's role. For the agency's own internal hires, the agency is treated as the employer.
  • If the candidate record is too thin to generate from (no resume summary and no cover letter), the form asks the consultant for a short free-text brief of what the call should cover and generates from that instead, swapping resume probes for experience probes.
  • During the call, the consultant scores each answer 1 to 5 stars and can flag concerns. Typed answers can get an AI-suggested score with a short reasoning line; the consultant always has the final say. Each question has a microphone button for dictating the answer using the browser's built-in speech recognition.
  • Submitting computes a weighted 0-100 score that is saved to the application. The score is compared against the company's screening thresholds (advance at 70 or above, reject at 45 or below, by default; editable in Settings under workflow rules). A strong score shows a "Looks strong - worth submitting" hint; a weak one shows "Below your bar - consider parking". These are hints only; they never move a candidate on their own. Borderline scores show no hint.
  • Concerns travel forward: answers scored 1 or 2 stars feed into interview question generation later as flag probes, so the interview probes exactly what worried the screener.
  • Regenerating questions is available if the first set misses the mark, but only before any answer has been saved.

The AI voice screen

  • Choosing AI screen in the wizard ends on a confirmation showing the candidate's email and the full list of questions the candidate will be asked - each with its answer type, weight (Critical, Important, or Nice to have) and required flag, and a help icon on AI-written ones giving the one-line reason it is being asked. Nothing is sent until the consultant has that list in front of them. AI-written sets carry a "Write them again" link beside the list, so a set that misses the mark is rewritten without leaving the step. Sending is also available via the assistant, Slack, and the API. The candidate gets an email with a link and opens it straight from the email; no account is needed, the link is the key. The link works for 7 days.
  • The candidate has a spoken conversation of roughly 10 minutes with the screening agent (hard cap 20 minutes). The agent asks the question set the wizard prepared (template or AI-written); if the consultant already curated questions, the voice screen asks those. Automatic sends, where no consultant is present to choose, always use AI-written questions.
  • When the call ends, the transcript comes back automatically. AI maps the transcript to per-question answers, scores each answered question 1 to 5, and computes the same weighted 0-100 score the manual form produces. The inviting consultant gets an in-app notification, and the answers, AI scores, and a "Read full transcript" disclosure appear in the phone screen panel, where the consultant can adjust any score and submit as usual.
  • A scored screen also reads in full from the home page's deal file: the Screen step shows the screen summary in front and the scorecard behind a tab - each question with its score out of ten and the candidate's answer word for word, plus the overall score. It is the same record the screen panel holds, read from the deal file. While the deal has not moved on, the verdict card also suggests the next moves - Submit to client, and Book interview - each button disappearing once its own step has started.
  • Sending again while a link is still open and unexpired re-sends the same link rather than creating a new one. If the invite email fails to send, the consultant can copy the link and send it themselves. Once a link has lapsed, the card offers a fresh one.
  • Auto-screening: a job can have AI screening switched on. When it is on, the invite goes out automatically as soon as the assessment completes, but only to promising inbound applicants: the candidate applied themselves (careers page, job board, social, referral) and the AI verdict is yes or maybe. Candidates a consultant added by hand are never emailed automatically; the manual send button is the only way to invite them. Clear no's are never auto-screened, and each application is only ever auto-invited once, even after a re-score.
  • Sending an AI screen to a candidate at Yes moves them to the Screen stage - the screen is in motion, and the strip shows it. Funnel and Maybe applicants keep their stage (triage is still yours): a sent but unfinished AI screen still shows them in the strip's Phone Screen position immediately, without their stage changing.

Logging a phone screen from chat, Slack, or the API

Outside the dashboard form, a phone screen is logged as a short draft: an outcome (Advance, Reject, Hold, or No show) plus required notes, with optional call time and duration. Confirming refuses until both outcome and notes are filled in. On confirm, one timeline note is written to the application (including call time and length if given). The outcome advances or parks. A Reject outcome moves any pre-client application to No; the rejection email is not sent from here, the consultant sends it separately, so rejections can be reviewed, edited, or bulk-actioned later. Advance and Hold move a Yes candidate to Screen (the screen happened). No show leaves the stage unchanged, and a candidate already with the client is never pulled back by a screen log.

Voice input elsewhere in the app

  • Hold the Space bar anywhere in the dashboard (when not typing in a field) to talk to Lovelio. Release to send; the transcript goes through the assistant exactly as if typed, including the same confirmation cards for actions. Taps under 300ms and pressing Escape discard the recording.
  • The search bar, assistant voice capture, and referee forms accept voice input transcribed server-side. The phone screen form's per-question dictation uses the browser's own speech recognition; in a browser without that support the microphone button does nothing.

Limits and edge cases

  • Phone screening does not appear without an active plan, and a cancelled workspace never auto-invites even if a job kept the AI screening toggle on.
  • The screening form shows an AI screen card at the top only when an AI screen is actually out with that candidate (sent, in progress, scored, or lapsed). Choosing Phone screen in the wizard means no card at all - the AI option is offered in the wizard, not re-offered over the questions. If the AI screen option is greyed out in the wizard on an active workspace, the AI voice screen has not been configured for that deployment; running the screen yourself still works.
  • The AI screen link expires after 7 days; sending again after expiry issues a fresh link.
  • Sending requires the candidate to have an email address on record.
  • A submit that saves but then hits an error shows a retry message rather than losing the answers; reopening the form shows what saved.

Common questions

  • "Does a good score move the candidate forward automatically?" No. Completing a screen on a Yes candidate records them at the Screen stage, but the score itself only produces a hint - submitting to the client or parking is always the consultant's action.
  • "Will attaching a candidate to a job with AI screening on email them?" No. Auto-invites only go to candidates who applied themselves.
  • "Can the candidate retake the AI screen?" While a link is open, re-sending reuses the same link. After a screen finishes or expires, the consultant can send again and a fresh link is created; nothing re-sends automatically.
  • "Where do I change the advance and reject score thresholds?" Settings, in the workflow rules section.