Tailor applications, invent nothing
Job hunting is a volume game with a tailoring tax. Every posting wants a summary in its language and a letter that sounds like you read the advert, and by the tenth one you are pasting the same paragraph and hoping. Below is the version I build, including the constraint that matters most: the prompt is not allowed to add a single fact that is not already in your CV.
The questionHow do I tailor my CV and cover letter for every job application without lying or spending all day on it?
How the working version is put together
Read the tracker before anything else
The first thing the run does is read your own tracker sheet, so a posting that has already been seen is dropped before it costs you anything. The sheet is the memory. Without it you re-apply to the same role under three slightly different titles, which is the fastest way to look careless to the one employer you wanted.
Treat the master CV as the only source of fact
Your CV lives in a Google Doc as plain text and is read fresh on every run. Everything the model is allowed to say has to come from that document. This is the difference between tailoring and fabricating, and it is a prompt constraint rather than a hope: ask a model to improve your CV against a posting and it will cheerfully hand you a certification you never sat.
Throw postings out before spending a model call
A Code node drops anything already tracked, rejects titles carrying seniority words you do not want, checks your must-have and exclude keywords, and caps the batch at a number you set. Deterministic rules are cheaper and more predictable than asking a model whether a job is worth reading, and the cap is what stops a noisy feed turning into forty drafts nobody reviews.
Score the fit, and name the gap
Each surviving posting gets a fit score out of 100, three matched strengths, one honest gap, a four-line tailored summary and a letter under 180 words. The gap is the part I would not remove. Knowing you are short on one requirement before you apply is what lets you address it in the letter instead of hoping nobody reads that line of the advert.
Draft into Gmail, never send
Postings at or above your threshold get a summary file in Drive and a Gmail draft holding the letter. Nothing is ever sent by the automation. I build it this way on purpose: the tailoring is the boring half worth automating, and pressing send is the half that should stay yours.
Write every posting down, drafted or skipped
Both outcomes land in the tracker with the score and the gap. That gives you a record of what you passed on and why, which matters more than it sounds three weeks into a search when everything blurs, and it doubles as tomorrow's dedupe list.
Tailor a CV summary and cover letter for every new job posting, without inventing a thing
I packaged this one as an n8n workflow you import and own: 17 working nodes, the setup written on the canvas, and one settings node holding your keywords, your threshold and your daily cap. Instant download, and there is no account with me to maintain once you have it.
Open the kit on Etsy ↗The same workflow bundled with a Notion job tracker, in English and in French
If you want somewhere to keep the applications as well as something to draft them, the bundle pairs this exact workflow with a Notion job tracker in both languages. It holds every posting, the interviews and the contacts, and it carries the same facts bank the workflow writes from. The two Notion templates are free on Notion's own marketplace and the download links you to them, so what you pay for is the workflow and the setup written in English and French.
Open the bundle on Etsy ↗What the manual version really costs
I measured this on my own applications rather than lifting it from a study. On a board with native apply, a role takes about a minute end to end. On an employer's own applicant system it is ten minutes or more, most of it retyping a CV the form has just parsed badly.
Tailoring sits on top of that, and it is the part that quietly stops happening. The first five applications get a bespoke paragraph, the next twenty get the same one, and the difference is invisible to you and obvious to the person reading. Automating the tailoring is worth more than automating the submitting, which is the opposite of where most tools aim.
The parts that break
Feeds are the first. Boards change or retire their RSS quietly, and the failure looks like a quiet job market rather than an error. If you run this, check that the tracker is still gaining rows every week.
Duplicates are the second. The same role is syndicated across several boards with slightly different titles, so the dedupe key matters, and matching on the link alone will not catch it.
The third is the one worth being honest about: a fit score is a model's opinion of your CV against an advert, and it is not a prediction of anything. I use it to sort a morning's reading, not to decide what I am capable of.
What it costs to run
One model call per posting that survives the rules filter, and the rules throw most of them out, so the daily cost is small and predictable. Everything else runs on a Google account you already have.
The cap in the settings node is the real cost control. Set it high and you get a drafts folder nobody reads, which is exactly the failure you were trying to automate away in the first place.
Why take this from me
Because grounding a model in a document and forbidding it to go beyond it is the same problem I solve for clients, and the case studies below are the version of that with money attached. One of them is an assistant that answers only from real data, which is this constraint wearing different clothes.
I also apply for work myself, so this is not a hypothetical annoyance I read about. The honest limit is that I cannot tell you it gets you interviews. It gets you tailored applications at a volume you could not sustain by hand, and what happens next is not a thing any automation can promise.
When I would tell you not to bother
If you apply to a role a month, do it by hand and do it properly. This earns its keep somewhere north of a few applications a week, and below that the setup costs more attention than the typing ever did.
If what you actually want is applications sent automatically, this is not that, and I would push back on wanting it. Nothing here sends. A tool that fires off applications without you reading them is how people end up applying for jobs they cannot do, in words they did not write.
What lands on your desk
- I wire it into your own Google account, reading your own CV and the board you actually watch
- A rules filter you control, so a noisy feed cannot turn into forty drafts nobody reviews
- A prompt I constrain to your CV, so the letters tailor rather than invent
- One honest gap reported on every application, so you know where you are stretching before you send it
- A tracker that is your record of what you passed on and the dedupe memory for tomorrow
Open the proof
Published case studies of systems that are running. Each links to the thing itself.
My Etsy shop, all the kits ↗Booboo on GitHub (MIT, read the code) ↗
Questions this page answers
- Can the model invent something not in my CV?
- I constrain it to your master CV document, so it can't invent anything. That document lives in a Google Doc and is read fresh on every run.
- Does it send applications for me?
- Nothing gets sent by the automation. Postings at or above your threshold get a Gmail draft holding the letter, and pressing send stays yours.
- What if I'm missing a requirement?
- You get one honest gap per posting, alongside three matched strengths and a fit score. That's what lets you address the gap in the letter instead of hoping nobody reads that line of the advert.
- Is this worth it if I only apply occasionally?
- It probably isn't worth it if you only apply to a role a month. Do it by hand instead; this earns its keep somewhere north of a few applications a week.
Want this built and handed over working?
One line is enough to start. You get an honest answer on fit, and a number rather than a discovery call.
Start a brief