Refolk
TeardownApplying at volume

The Auto-Apply Sent-Log Audit, 60 Bot Rows to Fixes and a Dated Stop Rule

Sort every bot-sent application into five classes, send each fix through the channel its portal supports, and date a stop, restrict, or keep rule for the tool.

18 min readLast reviewed November 14, 2026Read as Markdown

Key takeaways

  • An auto-apply dashboard count is a ceiling, not a tally: one vendor reports roughly one in three applications its tool starts does not go through, and the biggest failure bucket is being unable to confirm the submission.
  • Sponsorship is the most damaging field for a bot to get wrong, because it is a yes/no question Greenhouse can auto-reject, and Greenhouse and SmartRecruiters make the rejection email optional, so the error can fail silently.
  • Never withdraw and reapply to fix a wrong answer: at least one Workday employer blocks reapplying to the same requisition after a withdrawal, and Huntr data shows the per-application interview rate falling from 5.9% at one application per company to 2.4% at eight or more.
  • In Refolk's index, US recruiting profiles listing Workday outnumber those listing Greenhouse 998 to 276 (3.6x), so many corrections will run through Workday's per-stage Withdraw button or a recruiter, and Greenhouse corrections always mean emailing the company.
  • A 60-row log with 2 interviews (3.3%) cannot be told apart from a 3.5% baseline, so decide whether to keep the tool on counted errors, not on its interview rate.
  • Setting a Greenhouse application to Inactive only tidies your own view and cannot be undone; a withdrawal counts only once the employer replies to confirm it.

This is the audit for anyone who let an auto-apply tool send applications in their name and now needs to know what actually went out. I follow one 60-row sent log from two weeks of bot sending through receipt checks, answer checks, duplicate counts and the corrections each portal really allows, including the wrong turns. By the end, every row on your own log will be sorted into one of five classes, every material error will have a note sent, and you'll have a dated decision to stop, restrict, or keep the tool.

What does an auto-apply sent-log audit produce?

A sent-log audit turns the tool's "60 applied" into five counts you can act on: Confirmed, Ghost, Wrong-Fit, Wrong-Answer and Duplicate. It also produces a correction or withdrawal note for each material error and a dated rule for the tool. A sent log is the tool's own record of what it says it submitted. None of it counts as proof until the employer's side agrees.

Each row gets a receipt status first (Confirmed or Ghost), then any error flags. One row can carry more than one flag, for example Wrong-Answer and Duplicate. Record every flag, but act on the most severe.

ClassWhat it provesHow it lies
ConfirmedThe employer side shows it as receivedIt can still hide a silent knockout
GhostThe tool claimed it, but the employer side has no recordAn email alias or third-party route can hide a real one
Wrong-AnswerA screening answer contradicts your recordsGreenhouse free-text slips can't auto-reject
Wrong-FitThe role misses your target criteriaVague criteria make every role look like a fit
DuplicateMore than one application to one employer or requisitionSeparate requisitions are not always a problem

Two terms come up throughout. An ATS (applicant tracking system) is the software an employer uses to receive and route applications. A knockout field is a screening question the ATS can use to reject you automatically. Severity runs in this order: a Wrong-Answer on a knockout field, then Duplicate, then Wrong-Fit. Ghost rows sit outside that order, because the employer has nothing on file for you to correct.

Why is the tool's "applied" count not a receipt?

Many tools infer that an application was submitted instead of confirming it, so the dashboard count is the most it could be, not what actually happened. One vendor reports that roughly one in three applications its tool starts does not go through. The largest single failure bucket is the tool being unable to confirm the submission at all.

The other published figures point the same way. All of them come from vendors or reviewers, not independent studies. A review of user reports put failing or erroring submissions at 25 to 40 percent. One published test ran eight tools, and only three completed the submission.

SourceApplicationsOutcome
remotejobassistant, 14-day test473 callbacks (6.4%)
scale.jobs, LazyApply test470 interviews
hrlens tool test8 tools3 completed submission

For the 60-row log, these figures set an expectation before you open a single inbox. At one in three, about 20 rows would be Ghosts. At 25 to 40 percent, between 15 and 24 would be. Neither is a prediction for your tool. They are a reason not to trust the number 60.

A dashboard that says applied is a claim. A portal that says Submitted is a receipt.

The audit procedure, step by step

Seven steps, in this order, because each one changes what the next one is allowed to do. The most important constraint is that duplicates get counted before any fix is attempted, so a fix can't create a new duplicate.

Sent-log audit, from export to dated rule

  1. Export the tool's sent log
    One row per claimed submission, with employer, requisition, timestamp, portal, and the screening answers the tool entered.
  2. Check receipt for every row
    Look for a confirmation email, confirmation ID, final confirmation page, or Submitted status in the employer portal, across every email alias. Mark each row Confirmed or Ghost.
  3. Check knockout answers field by field
    Compare authorization, sponsorship, salary, relocation and years of experience against your own records. Mark each Confirmed row Wrong-Answer or clean.
  4. Check fit against your target criteria
    Hold each role against the criteria you wrote down before the audit. Mark Wrong-Fit rows and the criterion each one missed.
  5. Count duplicates per employer and requisition
    Count before touching anything, so no fix pushes an employer's count higher.
  6. Correct or withdraw through the supported channel
    Use the Workday Withdraw button where it shows, and email the employer for Greenhouse, LinkedIn Easy Apply and undocumented portals. Never reapply to fix an answer.
  7. Compare results and set a dated rule
    Set your counts against published baselines, then write down stop, restrict, or keep with today's date and a review date.

Here are my own working estimates, not sourced figures. The export takes about 30 minutes. The receipt check takes 2 to 3 hours for 60 rows, because each row means a search in your inbox and a login to the employer's portal. The answer and fit checks go faster once your records are open next to the sheet. One vendor advises checking receipt within 48 hours of each attempt. If the tool has run for two weeks without checks, the oldest rows are already past that window, so start with them.

Sent-log audit sheet columns
Row | Employer | Requisition ID | Job title | Portal (Workday / Greenhouse / LinkedIn Easy Apply / Other) | Tool timestamp | Email or alias used | Receipt evidence (email / ID / page / portal status) | Receipt class (Confirmed / Ghost) | Authorization answer | Sponsorship answer | Salary answer | Relocation answer | Years answer | Wrong-Answer field(s) | Wrong-Fit criterion | Applications to this employer | Applications to this requisition | Action (none / correct / withdraw) | Channel used | Date sent | Employer reply | Final class

Paste as the header row of a spreadsheet; one row per claimed submission, never merge rows.

Receipt and answer checks on the 60-row log

The worked log belongs to a US-based seeker who is authorized to work, does not need visa sponsorship, has a salary floor, and will only relocate to a short list of cities. The tool ran for 14 days and logged 60 rows across Workday, Greenhouse, LinkedIn Easy Apply and a few portals whose candidate help I can't document. Below is each fork the audit reached, including the ones it got wrong the first time.

Fork 1: Confirmed or Ghost

A row is Confirmed only when the employer's side shows it. Workday sends a confirmation email to the address on the Workday account. Under Candidate Home, My Applications lists the job title, requisition number, date submitted and status. That status has to read Submitted. The tool can mark a row "applied" when Workday actually holds an application in progress, and that draft also blocks you from reapplying to the requisition.

Greenhouse was the first wrong turn. On the first pass, several Greenhouse rows went into Ghost because nothing appeared in MyGreenhouse or the main inbox. Greenhouse lists three reasons an application may not show: the employer hasn't shared it, you applied with a different email, or you applied through a third-party site. The tool had applied from a relay address. Employer mail sent to relay inboxes is not trivial. One vendor reports that 221 users sent 2,813 applications through its relay inbox over about three weeks, and those inboxes received 4,226 employer emails. Searching the relay address moved those rows back to Confirmed.

Rule for this fork: no row goes into Ghost until you've searched every address the tool could have used and logged in to the employer portal under each one.

Fork 2: Wrong-Answer

A Wrong-Answer is any screening answer that contradicts your own records. It matters most on knockout fields. The usual ones are work authorization, visa sponsorship, location, a required license, and minimum years of experience.

Where a screening answer goes wrong

  1. Tool profile
    The answers you stored once, which the tool reuses on every form
  2. Form mapping
    The tool picks an option on each employer's own wording of the question
  3. ATS knockout rule
    Yes/no, single-select and multi-select answers can trigger auto-reject in Greenhouse
  4. Rejection email
    Optional in Greenhouse and SmartRecruiters, so a knockout can stay silent
A wrong answer can enter at the tool, get locked in by the form, be enforced by the ATS, and never be reported back to you.

Sponsorship is where an error does the most damage. In Greenhouse it is a yes/no question, one of the three answer types that can auto-reject; free-text answers cannot. iCIMS documents an Initial DNQ (did not qualify) status, and Oracle Taleo uses disqualification questions. One recruiter estimates the sponsorship question knocks out "probably 30%" of applicants for technical roles. This isn't hypothetical with bots. A LazyApply reviewer wrote that the tool "inaccurately stated that I needed an H-1B visa when I actually did not." The knockout behaviors above come from secondary summaries of vendor documentation. If a row depends on one, check the vendor's own page.

For each Confirmed row, read the answers the tool entered, field by field, against a single written source of truth. This check is only as good as that source. If your work history, dates and authorization status exist in one place, the comparison is mechanical. Refolk writes your resume from your own history, so the facts the tool should have used are already written down to compare against.

Fork 3: Wrong-Fit

A Wrong-Fit row is a role that misses the target criteria you set before the tool started: title, level, location, pay floor and must-have skills. Write those criteria down before you read any rows. Otherwise you'll grade each posting on how it looks rather than against your criteria, and almost everything will pass. In the worked log, Wrong-Fit rows clustered where the tool had matched on a keyword in the title and ignored the location or pay line. Refolk scores how well you actually fit a posting, and that score is a fast way to grade the Wrong-Fit column without rereading every job description.

How do duplicates happen, and why is reapplying the trap?

Duplicates come from two sources: the tool hitting the same employer through several postings, and you reapplying to fix a wrong answer. The second is worse because you chose it. Huntr's data shows the interview rate per application dropping from 5.9% with one application to a company to 2.4% with eight or more.

Count both columns before any fix: applications per employer and applications per requisition. Two applications to the same requisition is a true Duplicate. Several requisitions at one large employer is not automatically wrong, but each one lowers the per-application rate above, and the next fix you consider would add to it.

The worst wrong turn in this audit happened here. A Workday row had the wrong sponsorship answer, and the obvious fix was to withdraw and reapply with the right one. Published guides disagree on this. One says withdraw and reapply. Another says not to submit a second application to the same posting, because that can create a duplicate. The employers settle it. One Workday employer's FAQ says that once you withdraw, you cannot reapply to the same requisition. BCG's candidate FAQ says: "Please do not withdraw your application with the intention of reapplying, as you can only apply to each role once."

I couldn't find ATS vendor documentation of per-candidate caps or detection of identical text, so I can't tell you whether an employer's system flags your duplicates automatically. Assume a recruiter will see them.

Which correction channel does each portal support?

Each employer's portal setup decides the correction channel. Workday may give you a Withdraw button, depending on the stage. Greenhouse and LinkedIn Easy Apply give you no in-portal edit or withdrawal, so the channel is a direct message to the employer. Whether you can withdraw depends on how each employer configured its ATS, not on the ATS brand.

PortalEdit after submitWithdrawChannel
WorkdayNot documentedButton only at stages the employer enabledButton, else recruiter
GreenhouseNoNot in MyGreenhouseEmail the company
LinkedIn Easy ApplyNoNoMessage the employer
Lever, iCIMS, SmartRecruiters, AshbyNot establishedNot establishedAsk via the confirmation email

In Workday, the employer adds a candidate withdrawal reason to each stage where withdrawal is allowed. If a stage has none, the Withdraw button doesn't appear for external candidates. So the same application can be withdrawable at intake and not later. Greenhouse says MyGreenhouse can't change an application or contact recruiters for you, and it can't change the email on an application after submission. I didn't retrieve Indeed's rules, so check its help pages yourself if your log includes Indeed rows.

Which channel dominates depends on your employer mix. Refolk's index gives a rough picture of where each ATS shows up.

3.6x
US recruiting profiles listing Workday versus Greenhouse as a skill
In Refolk's index, 998 US recruiters, technical recruiters, talent acquisition partners and recruiting coordinators list Workday, against 276 listing Greenhouse.
SegmentProfilesTechnical Recruiter share of top-titles sample
US, Workday9981/25 (4%)
US, Greenhouse2767/25 (28%)
UK, Greenhouse92/9

In Refolk's index, the Workday-skill US sample includes recruiters at enterprises like Boeing, Siemens, Lenovo and Takeda, where a portal button may exist. The Greenhouse-skill sample includes Snowflake, Warby Parker, Datavant and Rent The Runway, where a correction always means writing to a person. US Greenhouse profiles outnumber UK ones 30.7x. Skill tags undercount the people who actually use each tool, so read these as proportions, not a census.

Correction note and withdrawal note
Subject: Correction to my application, [Job title], Req [Requisition ID]

Hello [Name or Recruiting team],

I applied for [Job title] (Req [Requisition ID]) on [date]. One screening answer on that application is wrong.
Question: [question as written on the form]
Submitted answer: [what was entered]
Correct answer: [your answer]
I'd be grateful if you could note the correction on my application. I have not resubmitted, to avoid a duplicate.

Thank you,
[Full name] | [email used on the application] | [phone]

---

Subject: Withdrawal request, [Job title], Req [Requisition ID]

Hello [Name or Recruiting team],

Please withdraw my application for [Job title] (Req [Requisition ID]), submitted on [date] from [email used on the application]. It was sent in error and doesn't match the roles I am pursuing. Could you confirm once it has been withdrawn?

Thank you,
[Full name]

Send to the recruiter or recruiting inbox for that requisition; replace the bracketed fields, keep it to the facts, and send one note per requisition.

The correction note says "I have not resubmitted" on purpose. It tells the recruiter you know about the one-application rule and won't send a second copy.

How does this audit go wrong?

Most audit errors are classification errors: a row ends up in the wrong class, and then the wrong action follows. Each failure mode below comes with what it looks like and the check that catches it.

  1. False Ghost from an email mismatch. A Greenhouse application can look missing because the tool used a different address or routed the application through a third-party site. Check: search every alias and relay inbox before you classify a row as Ghost.
  2. Workday draft counted as applied. The tool logs "applied" while Workday holds an application in progress, which also blocks you from reapplying. Check: the status in My Applications must read Submitted, not a draft.
  3. Silent knockout read as "pending". Greenhouse and SmartRecruiters make the rejection email optional, so a row with no response may already have been auto-rejected on a knockout answer. Check: verify the knockout answers before you count a row as Confirmed and still in play.
  4. Duplicate created by the fix. Reapplying to correct an answer can lock you out of the requisition after a Workday withdrawal and raises your count at that employer. Check: count requisitions per employer before taking any fix action.
  5. Missing Withdraw button read as failure. No button doesn't mean the application failed to submit. It means the employer didn't enable withdrawal at that stage. Check: confirm receipt separately, then email the recruiter.
  6. Greenhouse Inactive read as withdrawal. Inactive only changes your own view. Check: a reply from the employer confirming the withdrawal.
  7. Baseline comparison overread. Huntr's rates are self-tracked by its own users, and a 60-row log is small. Check: work out whether your rate could plausibly be the baseline before calling the tool better or worse. The next section shows the arithmetic.

The pattern across all seven: every check uses the employer's record, never the tool's. When the tool and the portal disagree, trust the portal.

Stop, restrict, or keep: how do you set the dated rule?

Decide on counted errors, not on the interview rate, because 60 rows can't separate a bot's rate from the baseline. The worked log produced 2 interviews from 60 rows, or 3.3%, against Huntr's 3.5% average. On my own arithmetic, that difference is too small to be distinguishable from the baseline at this sample size.

Condition (Huntr, self-tracked)Interview rate per application
Annual average3.5%
Untailored resume, one quarter2.07%
Tailored resume, one quarter4.23%
8 or more applications to one company2.4%
LinkedIn, trailing 12 months3.6%

These figures come from Huntr users tracking their own searches. One quarter covered 375k job entries from 25,887 job seekers. They don't isolate manually sent applications, and no independent study of interview rates for bot-sent applications exists. Use them as a rough reference, not as a measurement standard.

What the table does show is where volume is weakest. Untailored applications convert at about half the tailored rate. Piling applications onto one company drags the per-application rate down. Huntr also reports Google job search at 9.9% against LinkedIn at 3.6% over 12 months, and employer career sites at 6.2% against job boards at 4.9%. If you keep the tool, narrowing it toward career-site postings and fewer applications per employer is the change the evidence supports.

Stop, restrict, or keep

High receipt rateLow receipt rate
Keep, narrowed
Tighten filters to career sites and target criteria, cap applications per employer, re-audit on the review date
Restrict to review mode
The tool drafts, you approve every screening answer before it submits
Restrict and fix detection
Rows are clean but unconfirmed; require proof of receipt before the tool logs a row
Stop
Unreliable and wrong; withdraw or correct the damage and turn it off
No knockout errorsAny knockout error
Receipt reliability and knockout errors decide the rule, and either one being bad is enough to take the tool off autopilot.

Three more factors can override the matrix:

  • LinkedIn automation. LinkedIn's prohibited software page bans third-party bots and extensions that automate activity on its site. Its restrictions page says automated inauthentic activity can lead to temporary or permanent restriction. Semafor reported that LinkedIn restricted the account of AIHawk's creator. LinkedIn confirms an Easy Apply daily limit, a temporary speed limit for rapid submissions, and possible lower limits for activity that looks inauthentic, but it publishes no numbers. Third-party figures for the limit are unverified. If your tool automates LinkedIn, stop it on that channel whatever the matrix says.
  • Recruiter reaction. In TopResume's survey of 600 US hiring managers, 19.6% would reject a candidate with an AI-generated resume or cover letter, and another 20% saw heavy AI reliance as a red flag.
  • Wrong sponsorship or authorization answers. One of these is enough to move the tool out of unattended mode. Those errors fail silently, and you only find them by auditing.

If you choose review mode, the cost moves to your time, and that cost is mostly tailoring. Refolk tailors your resume to each posting and drafts the cover letter, which puts review-mode applications in the tailored condition (4.23% in the table) instead of the untailored one (2.07%).

Write the rule as one dated line in the sheet, for example: "On [date], restricted to review mode after [n] Wrong-Answer rows; re-audit on [date two weeks later] using the same columns." Then run this audit again on that date, on the new rows only.

Before you close the audit

  • Every row in the sent log has a receipt class, and no row is Ghost until every alias has been searched.
  • Every Workday row marked Confirmed shows Submitted, not a draft, in My Applications.
  • Every Confirmed row has its authorization, sponsorship, salary, relocation and years answers checked against your records.
  • Applications per employer and per requisition are counted, and no fix added a new application.
  • Every material Wrong-Answer row has a correction note sent through the channel its portal supports.
  • Every withdrawal is logged as done only after an employer reply, not after a Greenhouse Inactive change.
  • The stop, restrict, or keep rule is written down with today's date and a review date.

Questions job seekers ask

How do I check what an auto-apply tool actually submitted?

Export the tool's sent log, then check each row against the employer's side: a confirmation email, a confirmation ID, a final confirmation page, or a Submitted status in the employer's candidate portal. In Workday, Candidate Home under My Applications shows job title, requisition number, date submitted and status. Search every email alias or relay inbox the tool used before you call a row a Ghost, and treat a Workday draft as not submitted.

Can I correct a wrong sponsorship answer after the application is submitted?

Usually only by contacting the employer, because no portal I can document lets you edit a submitted answer. Greenhouse says plainly that corrections go to the company you applied to, and LinkedIn Easy Apply applications can't be edited after submission. Send a short note to the recruiter on that requisition giving the correct answer. Assume the wrong answer may already have caused a silent auto-rejection.

How do I withdraw a job application that was submitted by mistake?

Use the Withdraw button in Workday if it shows. If it doesn't, the employer hasn't enabled withdrawal at that stage, which is not the same as the application having failed. For Greenhouse, LinkedIn Easy Apply and other portals, email the employer and ask them to withdraw it. A withdrawal counts only when the employer confirms it; setting a Greenhouse application to Inactive does not withdraw anything.

Should I withdraw and reapply to fix a wrong answer?

No. At least one Workday employer's FAQ says that once you withdraw, you cannot reapply to the same requisition, and BCG's candidate FAQ asks people not to withdraw in order to reapply because each role takes one application. Reapplying also raises your count at that employer, and Huntr data shows the per-application interview rate falling from 5.9% at one application per company to 2.4% at eight or more.

Is auto apply worth it, based on actual results?

The evidence can't tell you yet. Published results come from vendors and reviewers: one 14-day test got 3 callbacks from 47 applications (6.4%), and another got 0 interviews from 47. No independent study exists. On your own log, 60 rows with 2 interviews (3.3%) can't be told apart from a 3.5% baseline, so base the decision on receipt failures and wrong answers, which you can count.

Will LinkedIn restrict my account for using an auto-apply tool?

It can. LinkedIn's prohibited software page bans third-party bots and extensions that automate activity on its site, and its restrictions page says automated inauthentic activity can lead to temporary or permanent restriction. Semafor reported that LinkedIn restricted the account of the creator of AIHawk, an open-source bot. LinkedIn doesn't publish a fixed restriction length or its Easy Apply daily limit.

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.

  1. 01Drop your resume

    A PDF or a LinkedIn URL. About a minute, once.

  2. 02I rank the openings

    Every weekday morning, the live catalog scored against your history. Up to 20 worth your time, not two hundred links.

  3. 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.

Read next