about 200 job applications went out from my tracker. 49 said no. one invited me to an interview. most never answered.
AI did the heavy lifting: it found roles, wrote a CV for each one and filled in the forms. it never pressed submit. i did that, after checking every form.
honest part first: AI got me volume, not interviews. this guide is the tracker that kept 395 rows from turning into chaos, and the review step i'd keep even if i did it all by hand.
what you need
- a spreadsheet or a plain CSV file
- an AI agent that can read and write that file. i use claude code, with a browser connected for the forms
- one CV with only true facts in it. the AI builds every tailored version from this, so it can't invent anything
01copy the header
seven columns, one row per job. that's all mine has.
company,role,status,date,url,source,note
- company and role: as written in the post
- status: one word from the list in step 2
- date: the day the status last changed, as 2026-11-08
- url: the job post, so you never apply twice
- source: where the agent found it, or which batch it came in
- note: one line on anything open, like "salary question left blank" or "posting needs fluent czech"
02copy the statuses
these are the ones my tracker actually uses:
draft_held needs_account blocked applied_unconfirmed reported_submitted applied interviewing rejected closed skipped excluded
- draft_held: form filled, parked at the submit button, nothing sent
- needs_account: the site wants an account, so i make it myself
- blocked: something only i can do, like a captcha or personal details
- applied_unconfirmed and reported_submitted: i think it went, but i never saw the confirmation page
- applied: i saw the confirmation page. only i set this one
- interviewing, rejected: the answer came back
- closed: the posting is gone
- skipped, excluded: not applying, and the note says why
the split between "i think it went" and "applied" matters. without it, your count of applications is a guess.
03how the AI fills it
for each batch, the agent works down this list and writes a row at every step:
- finds roles on job boards
- drops the obvious misses in plain code first: wrong city, wrong language, a must-have i don't have, already in the tracker
- asks AI the fuzzy questions: is this really an AI role, does it need on-site work. it scores what's left from 0 to 10 against my real CV and drops anything under 6
- shows me the shortlist. i pick
- builds a tailored CV for each pick from my true facts only, and runs it past my AI hiring panel (that's the PANEL freebie)
- fills every form up to submit, sets draft_held, and leaves the tab open
the prompt i'd start with:
here is my tracker: applications.csv, columns company,role,status,date,url,source,note. here is my CV: <file>. find roles that match <job titles> in <location>. skip any url already in the tracker. for each match, add a row with status draft_held only after the form is filled. never press submit. never set status to applied. if a form asks something my CV doesn't answer, leave it blank and say so in the note.
04review before you submit
this is the part i never hand over. for every draft_held row:
- open the tab and read every field, top to bottom
- check the CV attached is the right version for this job
- check any answer the agent wasn't sure of: salary, years of experience, right to work
- press submit, wait for the confirmation page, then set applied
check it worked
count your statuses. if applied and "i think it went" are mixed up, or rows have no url, fix that before the next batch.
three ways this breaks
- one form per tab, and close finished tabs one at a time. i once closed a whole tab group and lost every open draft at once.
- you let the agent mark things applied. an agent can think a form went through when the page never confirmed it. the confirmation page is the only proof.
- you chase volume. when i counted the contract roles that honestly fit me, the pool was 27, not hundreds a day. sending more of the wrong ones doesn't change the answer.
what i'd tell you
AI made applying cheap. it didn't make me stand out. in my panel tests the hiring manager was the real filter, and real numbers on the CV moved it. rewording never did. use the tracker to stay honest about the count, and spend the time you save on fewer, better CVs.