The Search Channel Score, Six Dimensions to Scale, Hold, Fix, or Cut
From a tracker with two weeks of logged sends, score every channel on six dimensions and write a Scale, Hold, Fix, or Cut call plus a 14-day hour split.
Key takeaways
- Score a job search channel by where the application was submitted, not where the posting was found, because Google Jobs sends applicants to the employer's own form.
- Employer career sites beat job boards 1.76x on interview rate in pooled Huntr data, but only 1.27x (6.2% against 4.9%) when the same people used both routes.
- At personal volumes, interviews alone cannot condemn a channel: 0 interviews from 20 sends still leaves a 95% Wilson upper bound of 16.1%.
- At 44 minutes per send, a job board at 3.4% costs about 1,294 minutes per first-round interview, against about 733 for an employer career site at 6.0%.
- Before cutting a channel, check your own behaviour: users sending 100 or more applications interviewed at 2.58% per application, against 9.25% for those sending 11 to 20.
- In Refolk's index, US agency recruiters outnumber in-house recruiters at US software, IT services and financial services employers 5.4x, so agency and recruiter inbound need separate tracker rows.
Every two weeks, someone running a search across many companies has to decide which channels get more hours and which ones they stop feeding. This guide is a fixed model for that decision. It covers six dimensions, a minimum-sample rule, a price for each channel in minutes per first-round interview, and a written Scale, Hold, Fix, or Cut verdict per channel with an hour split for the next 14 days. The model is fixed so that two people holding the same tracker reach the same call.
What counts as a channel in the Search Channel Score?
A channel is the route an application was submitted through, not the place where you found the posting. Score channels on submitted-via. Keep found-on as a separate column, because it answers a different question: where to look, not where to apply.
This split matters because discovery tools and submission routes are often different. Google Jobs, for example, routes applicants to the company's own application form. Huntr's data has 189 job seekers who used both Google job search and LinkedIn and got interviews at 9.0% on Google against 2.7% on LinkedIn. If you log "Google" as the channel, you are crediting a search engine with an employer form's results.
Huntr also attributes a Greenhouse job served at careers.acme.com to Acme rather than to Greenhouse. A branded applicant tracking system (ATS) page is the employer's career site for scoring purposes. An ATS is the software employers use to receive and screen applications.
Use these rows in your tracker, one per submitted-via route:
- Job board apply: the form lives on the board itself.
- Employer career page: the employer's own site, including ATS forms on the employer's domain.
- Standalone ATS page: an ATS-hosted form on the vendor's domain.
- Niche board: a specialist board, scored on its own row, not pooled with general boards.
- Referral: an employee submitted you or flagged you to the recruiter.
- Recruiter inbound (in-house): an employer's own recruiter contacted you and you replied.
- Agency: a third-party recruiter submitted you to a client.
- Direct outreach: your own message to a hiring manager or team member about a role.
- Former employer: any application back to a past employer, kept apart from everything else.
Former employer gets its own row because Huntr found that applying back to a former employer roughly doubles the interview rate, at 17.6% against 8.6%. Leave those sends inside "career page" and that channel looks better than it is.
Why agency and recruiter inbound are separate rows
Agency recruiters far outnumber in-house recruiters, so a lot of what feels like "recruiters reaching out" is agency traffic. If you pool the two rows, you are scoring two different businesses as one channel.
| Segment in Refolk's index | Recruiters | Multiple |
|---|---|---|
| US agency (staffing and recruiting) | 25,659 | 5.4x US in-house |
| UK agency (staffing and recruiting) | 19,916 | 0.78x US agency |
| US in-house, 3 industries | 4,758 | 1.0 |
Two limits apply here. Titles are matched by keyword, and the in-house row covers only software, IT services and financial services employers. In a sample of 25 top UK agency profiles, 18 carried the title "Recruitment Consultant". That title is a quick tell when a message lands in your inbox.
What do published baselines say about interview rates by channel?
Only three submission routes have published tracker-side interview rates: employer career sites, ATS pages and job boards. Career sites lead boards in every cut, but by 1.27x rather than 1.76x when the same people used both routes. Referral, recruiter and agency figures exist only as employer-side hire or offer rates.
This is the honest answer to "which job search channel gets the most interviews" and to "job board vs company career page interview rate". The table below is Huntr's, from two consecutive quarterly reports.
| Route | Q2 report | Q3 report | Same person, Q3 |
|---|---|---|---|
| Employer career site | 6.42% | 6.0% | 6.2% |
| ATS page | 5.29% | 4.8% | n/a |
| Job board | 3.82% | 3.4% | 4.9% |
| Career site / board | 1.7x | 1.76x | 1.27x |
The Q2 ATS figure comes from a secondary page. The Q3 sample was 84,753 applications. The same-person column covers 5,855 job seekers who used both routes. Over a longer window of 1,244,654 tracked applications, boards and aggregators reached an interview only 1.95% of the time.
The same-person column is the one to trust. In pooled data, stronger candidates apply direct more often, so part of the career-site lead is selection, not channel. That is why this guide uses 3.4% to 4.9% as the board baseline band: the pooled figure at the low end and the same-person figure at the high end.
What employer-side data says about referrals, sourcing and agencies
Employer-side figures measure hires and offers, not interviews. Use them to rank channels, never as targets for your own interview rate.
- Referrals (Ashby): about 1% of applications but roughly 18% of hires, across nearly 250,000 hires.
- Referral offers (Ashby): referred applicants got offers at 7.3%, against 0.2% for inbound applications.
- Referrals (Gem, secondary citation): referrals convert to hires at 11 times the rate of inbound applicants.
- Referrals at small businesses (CareerPlug): about 2% of applicants but 11% of hires.
- Direct sourcing (Gem): about 2.5% of applications but nearly 10% of hires.
- Job boards (Gem): 49% of applications but only 24.6% of hires.
- Trend (Ashby): inbound rose from 38% to 52% of all hires over about five years, while sourced and referral hires both fell as a share.
- Agency submittals: sources conflict. One cites a 46% submit-to-interview rate, and another says no public benchmark exists.
There is also a measurement disagreement. Huntr measures application to interview in one step, while some auto-apply analyses track a separate screen rate. Pick one definition of first-round interview and stick to it. I use the first live conversation scheduled about a specific role, recruiter screens included.
The six dimensions and how each is scored
The Search Channel Score has three gates and three scored dimensions. The gates decide whether a channel can be scored at all. The scores decide the verdict.
The Search Channel Score, outermost check first
- Attribution gateEvery send has a submitted-via value, and repeat and former-employer sends sit in their own rows
- Response window gateOnly sends older than 10 days are scored, and younger ones stay pending
- Sample floorFewer than 20 closed sends means Hold, whatever the rate looks like
- Scored dimensionsRate interval against the board baseline, minutes per interview, sending hygiene
| Dimension | Type | How it scores | How it lies |
|---|---|---|---|
| 1. Attribution | Gate | Pass if no blank submitted-via | Found-on logged as channel |
| 2. Response window | Gate | Pass if no pending send over 10 days | Fresh sends counted as misses |
| 3. Sample floor | Gate | Pass at 20+ closed sends | One reply flips the verdict |
| 4. Rate interval | 0 to 2 | Wilson bounds vs 3.4% to 4.9% | Hire multiples read as interview odds |
| 5. Minutes per interview | 0 to 2 | Under 733 / to 1,294 / over | Minutes never logged |
| 6. Sending hygiene | 0 to 2 | Flags on volume, repeats, tailoring, bots | Your behaviour blamed on the channel |
What each dimension proves
- Attribution proves you are scoring the route that converted. It lies when a search engine or aggregator is logged as the channel while the employer's form did the work.
- Response window proves a non-response is a real non-response. It lies when the most recent sends are scored as misses before employers could have replied. Huntr's median from application to interview is 5 days, but that median covers only applications that reached an interview, so the tail is longer. My cutoff is 10 days, twice that median. No per-channel cutoff has been published, so if your own tracker shows interviews arriving later, lengthen it.
- Sample floor proves the rate is more than noise. It lies when a single interview on 10 sends is read as a 10% channel.
- Rate interval proves whether the channel clearly beats or trails the board baseline. Score 2 if the Wilson lower bound is above 4.9%, 1 if the interval overlaps the 3.4% to 4.9% band, and 0 if the upper bound is below 3.4%.
- Minutes per interview proves what the channel costs you. Score 2 at or under 733 minutes, 1 from 734 to 1,294, and 0 over 1,294. A channel with zero interviews scores 0 once its logged minutes pass 1,294.
- Sending hygiene proves the channel was given a fair test. Score 2 with no flags, 1 with one, and 0 with two or more. The flags are covered in the Fix-versus-Cut section below.
How many sends before a channel's interview rate means anything?
Treat 20 closed sends as the floor for any verdict and about 40 sends as the point where a strong channel can clearly beat boards. No job-search-specific rule has been published, so this guide uses the Wilson score interval. It is a standard confidence interval for a proportion that behaves better than the textbook formula at small samples.
The Wilson formula, at z = 1.96 for 95% confidence:
- centre = (p̂ + z²/2n) / (1 + z²/n)
- margin = z × √(p̂(1 − p̂)/n + z²/4n²) / (1 + z²/n)
Here p̂ is your observed rate (interviews divided by closed sends) and n is closed sends. The lower bound is centre minus margin, and the upper bound is centre plus margin.
| Interviews / sends | Observed | Lower | Upper |
|---|---|---|---|
| 0 / 20 | 0% | 0% | 16.1% |
| 1 / 10 | 10% | 1.8% | 40.4% |
| 1 / 20 | 5% | 0.9% | 23.6% |
| 2 / 40 | 5% | 1.4% | 16.5% |
| 4 / 40 | 10% | 4.0% | 23.1% |
Read the table as two rules.
- Silence cannot condemn a channel at normal fortnightly volumes. Zero interviews from 20 sends still leaves an upper bound of 16.1%, which is far above any published channel rate except former employers. When an event is rare, a zero tells you little. A Cut verdict therefore has to rest on cost, not on missing replies.
- A good channel needs about 40 sends to prove itself. At 4 from 40, the lower bound of 4.0% only just clears the 3.4% board rate. It still sits inside the 3.4% to 4.9% band, so it scores 1, not 2. The interval narrows roughly with the square root of n, so doubling your sends does not halve your uncertainty.
How to price a channel in minutes per first-round interview
Minutes per first-round interview is all the minutes you spent on a channel divided by the first-round interviews it produced. That includes finding roles, tailoring, filling in forms and writing outreach. It is the one dimension that can justify cutting a channel before its interview rate is statistically settled.
The only published time figures are per application. In a 1,000-person survey, job seekers sent 62.6 applications and spent about 44 minutes on each. Laid-off job seekers reported 95.3 applications at about 32.6 minutes each. Nobody has published minutes per interview by channel, so the table below is derived. It assumes every channel takes the same 44 minutes per send, which has not been verified, and it inherits Huntr's self-flagging caveat.
| Route | Interview rate | Sends per interview | Minutes per interview |
|---|---|---|---|
| Former employer | 17.6% | 5.7 | 250 |
| Career site | 6.0% | 16.7 | 733 |
| Board | 3.4% | 29.4 | 1,294 |
Those two figures, 733 and 1,294, are the cost thresholds. A channel that costs less per interview than a typical career-site send scores 2. A channel that costs more than a typical board send scores 0.
What to log so the price is real
- Discovery minutes: time spent searching on that channel, split across the sends it produced.
- Tailoring minutes: resume and cover letter edits for that send.
- Form minutes: account creation, screening questions and uploads.
- Outreach minutes: drafting and following up on referral asks, recruiter replies and direct messages.
Your own minutes will not be 44 everywhere. A referral ask might take far longer to set up than a board form but convert far better. That is exactly why "how much time to spend on referrals vs applying online" cannot be answered from a pooled rate. Some practitioners say to spend 80% of your time on networking and referrals up front. This model measures first and lets each channel's minutes per interview set its share.
If tailoring is what pushes a career-page send to 44 minutes, that is the cost Refolk takes off the row. It tailors the resume to each posting from your own history and drafts the cover letter, so the minutes per send fall without dropping the tailoring that Huntr ties to a higher rate.
Broken or low-yield: how to tell Fix from Cut
A channel is broken when your sending behaviour is dragging its rate down, which you can fix. It is low-yield when you sent well and it still costs too much, so you cut it. Check four behaviour flags before you blame any channel.
| Flag | Benchmark | Effect on measured rate |
|---|---|---|
| High total volume | 100+ sends: 2.58% vs 9.25% at 11 to 20 | Over 3.5x lower per send |
| Repeat company | 8+ to one company: 2.4% vs 5.9% at 1 | Less than half |
| Untailored sends | 2.9% vs 5.95% tailored | About half |
| Auto-apply use | 22% admitted bots in one Greenhouse study | Volume without targeting |
Each flag lies in a specific way:
- Volume comes from Huntr's per-user counts over a quarter. It lies when one channel absorbs most of your high-volume sending, so the channel inherits the 2.58% drop. Higher volume tends to go with less targeting, which is why sending behaviour gets mistaken for channel yield.
- Repeat company lies when you apply to many roles at one employer through a single channel. Count those sends separately or exclude them.
- Tailoring lies when one channel gets your tailored resume and another gets a generic one. You are then comparing effort, not channels.
- Auto-apply figures conflict. One Greenhouse study found 22% of active job seekers admitted using bots, and a Greenhouse US survey found 3% use automated tools to mass-apply. Do not lean on either number. Just flag any channel where you used auto-apply.
Fix or Cut, by cost and sending hygiene
A missing reply is not evidence. A missing reply after 1,300 clean minutes is.
The repeat-company and volume flags usually come from sending to roles you were never a strong fit for. Scoring fit before you send is the cheapest fix. Refolk scores how well you actually fit each posting, so you can drop the weak sends before they ever reach a channel's denominator.
How to run the fortnightly score
The score takes about 80 minutes the first time and less after that, assuming you logged minutes per send. Run it on the same day every two weeks, on sends older than the 10-day cutoff.
The fortnightly Search Channel Score
- Tag every send with found-on and submitted-viaGive every application or outreach message both fields. Done means no blank submitted-via cell anywhere in the tracker.
- Close the response windowMark every send older than 10 days with no interview as a non-response, and leave younger sends pending. Done means no pending send older than the cutoff.
- Count closed sends and first-round interviewsCount both per submitted-via channel. Move repeat applications to one company and former-employer applications into their own rows.
- Apply the minimum-sample ruleCompute Wilson 95% bounds per channel. Under 20 closed sends is Hold, and no Cut or Scale while the interval contains the 3.4% to 4.9% baseline.
- Price each channel in minutes per interviewSum discovery, tailoring, form and outreach minutes and divide by interviews. Score 2 at or under 733, 1 up to 1,294, and 0 above.
- Diagnose broken versus low-yieldCheck volume, repeat-company count, tailoring share and auto-apply use per channel. Any flag sends the channel to Fix before it can be Cut.
- Write the call and the 14-day hour splitRecord Scale, Hold, Fix or Cut per channel using the verdict rules, then move Cut hours to Scale channels. Done means a dated written allocation.
The template below is the minimum tracker layout and the spreadsheet formulas for one channel row on a summary tab. It answers "how to track job applications by source" in a form you can score.
LOG TAB COLUMNS Date sent | Company | Role | Found on | Submitted via | Repeat company (Y/N) | Former employer (Y/N) | Tailored (Y/N) | Auto-applied (Y/N) | Minutes spent | Status (pending / no response / first-round interview) SUMMARY TAB, ONE ROW PER CHANNEL B2 closed sends | C2 first-round interviews | D2 total minutes Observed rate (E2): =C2/B2 Wilson lower bound (F2): =(C2/B2 + 1.96^2/(2*B2) - 1.96*SQRT((C2/B2)*(1-C2/B2)/B2 + 1.96^2/(4*B2^2))) / (1 + 1.96^2/B2) Wilson upper bound (G2): =(C2/B2 + 1.96^2/(2*B2) + 1.96*SQRT((C2/B2)*(1-C2/B2)/B2 + 1.96^2/(4*B2^2))) / (1 + 1.96^2/B2) Minutes per interview (H2): =IF(C2=0, IF(D2>1294, "over 1,294 with no interview", "none yet"), D2/C2) Rate score (I2): =IF(F2>0.049, 2, IF(G2<0.034, 0, 1)) Cost score (J2): =IF(C2=0, IF(D2>1294, 0, 1), IF(D2/C2<=733, 2, IF(D2/C2<=1294, 1, 0)))
Put one send per row on the log tab, and one channel per row on the summary tab, with closed sends in B, interviews in C and total minutes in D.
How the Search Channel Score goes wrong
Most bad reallocation calls come from one of seven errors, and each has a check you can run before you write the verdict. Run all seven every fortnight. They take less time than undoing a wrong Cut.
- Found-on logged as the channel. Google Jobs looks like a winner when the real submission was the employer's form. Check: the submitted-via column is fully populated before anything is scored.
- Pending counted as a miss. Scoring a fortnight of sends on day 14 penalises the newest sends. The 5-day median covers only the winners, so the tail runs longer. Check: score only sends older than the 10-day cutoff.
- One reply flips the verdict. One interview from 10 sends has an interval of 1.8% to 40.4%, so it is consistent with a terrible channel and a superb one. Check: no Cut or Scale while the interval still contains the 3.4% to 4.9% board baseline, unless cost decides it.
- Behaviour mistaken for channel. A board scored after you passed 100 total sends inherits the drop to 2.58%. Check: compare total volume, repeat-company count and tailoring share across channels before blaming one.
- Employer hire multiples used as interview odds. Referral 7.3% against inbound 0.2% is an offer rate across employers' pipelines, not your interview rate. Check: use employer multiples only to rank channels before you have data, never as a target your referral row must hit.
- An agency benchmark taken as fact. The published agency figures conflict, and one source says no public benchmark exists. Check: keep the agency row on Hold until you have 20 of your own closed submittals.
- Survivorship in published rates. Huntr's rates depend on users flagging interviews themselves, and people who keep tracking are not a random sample. Check: wherever a Huntr rate sets a threshold here (3.4%, 4.9%, 733, 1,294), treat it as approximate and let your own data overrule it once a channel passes 40 sends.
Recruiter timing deserves a separate note. Pin's data on more than 230,000 outreach threads puts the median candidate reply at 3.9 days. The median interview comes 5.8 days after first contact, with 58.9% interviewing within a week. That is recruiter-side timing, not yours. No medians exist for agency or referral sends, which is one more reason the 10-day cutoff is a judgement you should revisit against your own log.
Writing the call and the next 14 days
Apply the verdict rules in a fixed order, then split hours with one formula. The order matters: a channel that qualifies for Scale is never sent to Fix, and a channel with hygiene flags is never Cut.
| Verdict | Condition, checked in this order | Hours next 14 days |
|---|---|---|
| No verdict | Attribution or window gate fails | Fix the tracker first |
| Hold | Under 20 closed sends | Same as last fortnight |
| Scale | Rate 2 and cost 1+, or cost 2 and hygiene 2 | Gains Cut hours |
| Fix | Hygiene 0 or 1 | Same, with flagged behaviour changed |
| Cut | Cost 0 and hygiene 2 | Zero |
| Hold | Anything else | Same as last fortnight |
The hour split
Keep your total search hours constant, and move hours only from Cut channels.
- Add up last fortnight's hours on every Cut channel.
- Give those hours to Scale channels in proportion to interviews per hour, which is 60 divided by minutes per interview.
- If nothing qualifies for Scale, give them to the Hold channel with the lowest minutes per interview that has at least one interview.
- Fix channels keep their hours. Spend the first part of them clearing the flag, for example tailoring every send or stopping repeat applications to one company, before sending more.
This also answers "when to stop using a job board": stop when it scores Cut, which needs 20 or more closed sends, clean sending and a cost above about 1,294 minutes per interview. A quiet board alone does not meet that bar.
Review date: Sends scored: sent on or before (review date minus 10 days) Channel: Closed sends: First-round interviews: Wilson 95% interval: % to % Rate score (0-2): Cost score (0-2): Hygiene score (0-2): Minutes per interview: Hygiene flags (volume / repeat company / untailored / auto-apply): Verdict (Scale / Hold / Fix / Cut): Reason in one sentence: Hours last 14 days: Hours next 14 days: What changes in how I send on this channel:
Fill one block per channel, and keep every fortnight's record so you can see verdicts change.
Keeping the score current
The thresholds come from published baselines that move quarter to quarter. Each time Huntr publishes a new Job Search Trends report, re-check the board and career-site rates and recompute the 733 and 1,294 cost lines with the 44-minute assumption, or better, with your own average minutes per send. Once any channel passes 40 closed sends, your own rate is a better baseline than anyone's pooled figure.
Before you call the fortnight done
- Every send in the scored window has a submitted-via value
- No send older than 10 days is still marked pending
- Repeat-company and former-employer sends sit in their own rows
- Agency and in-house recruiter inbound are separate channels
- Every channel has a Wilson lower and upper bound written next to its rate
- No channel under 20 closed sends carries a Scale or Cut verdict
- Every Cut channel shows a cost score of 0 and a hygiene score of 2
- Every Fix channel names the specific behaviour that changes next fortnight
- Total hours next fortnight equal total hours last fortnight
- The verdict record is dated and saved alongside last fortnight's
Questions job seekers ask
Which job search channel gets the most interviews?
Among channels with published tracker data, employer career sites lead: Huntr measured 6.0% interviews per application against 3.4% for job boards, and applying back to a former employer reached 17.6%. Referrals likely do better still, since Ashby found referrals are about 1% of applications but roughly 18% of hires, but no tracker-side referral interview rate has been published. Your own tracker is the only reliable answer for your search.
Is applying on a company career page better than a job board?
Yes, but by less than pooled figures suggest. Huntr's pooled data shows career sites beating boards about 1.76x, but for 5,855 people who used both routes the gap was 6.2% against 4.9%, or 1.27x. Part of the pooled gap is that stronger candidates apply direct more often. Where a board links to the employer's own form, apply there and log it as a career-page send.
When should I stop using a job board?
Stop when it is expensive and your sending was clean, not when it simply went quiet. Under the Search Channel Score, a board is Cut only after at least 20 closed sends, when it costs more than about 1,294 minutes per first-round interview, and when your volume, repeat-company count, tailoring share and auto-apply use show no flags. Zero replies from 20 sends alone proves very little.
How much time should I spend on referrals versus applying online?
Let your own minutes per first-round interview decide, not a fixed split. Some practitioners advise spending 80% of your time on networking and referrals, and employer data supports ranking referrals highly, with Ashby reporting a 7.3% offer rate for referred applicants against 0.2% for inbound. Those are offer rates, though, so log your referral asks as sends and price them like any other channel.
How many applications do I need before a channel's interview rate is reliable?
Treat 20 closed sends as the floor for any verdict and about 40 as the point where a strong channel can clearly beat boards. One interview from 10 sends has a 95% Wilson interval of 1.8% to 40.4%. At 4 interviews from 40 sends the lower bound reaches 4.0%, just above the 3.4% board rate.
Put this to work
Paste your career in once. Every application after that is written for you.
Drop a resume or a LinkedIn URL. I rank the live openings against it, rewrite the resume and write a cover letter for the best of them, and fill in the employer's form when you press the button. You read, you decide what goes out.
01Drop your resume
A PDF or a LinkedIn URL. About a minute, once.
02I rank the openings
Every weekday morning, the live catalog scored against your history. Up to 20 worth your time, not two hundred links.
03Each one is written up
Resume rewritten for the posting, a cover letter, a fit score. Press send, or let me fill in the form.
- New matches ranked and written before you are up.
- Every bullet stays inside what your history supports. Nothing invented.
- Queued, submitted, interviewing, offer: one screen, not a spreadsheet.
Free, no card. Up to 25 applications sent a week, and nothing is sent until you say so.