A recruiter I know filled a mid-level marketing role in April. The posting was live for nine days and drew 412 applications. She read 31 of them closely. Not out of laziness, but because the first pass was a twelve-second scan, and most letters failed it before the second paragraph.
That ratio explains why AI cover letter generators have become so popular. It also explains why most of them are useless. Speed was never the bottleneck. Getting read is. A tool that produces a polite 300-word letter in eight seconds will take you exactly as far as the generic letter you would have written yourself, only faster.
Used with a bit of care, though, these tools do something genuinely awkward to do alone: they force you to map your experience onto one specific job description before you start typing. Here is how to get that, and how to spot the output that is quietly working against you.
What an AI cover letter generator actually does
Under the interface, nearly all of them are the same product wearing different logos. You paste a job advert and a CV or LinkedIn profile into a form, and a large language model produces 250 to 400 words in a house style. Some bolt on a job-board integration, a tone slider, or a button labelled “humanise.” The button does not do much.
Where they earn their keep
- They kill the blank page. Writing the first sentence of a cover letter at 11pm is miserable. Reading a mediocre draft and fixing it takes a fraction of the willpower.
- They mirror the vocabulary of the advert. If the posting mentions “stakeholder management” three times, the draft will use it, which helps with automated screening tools that score keyword overlap.
- They compress. Boiling eleven years of work history into three tight paragraphs that stay on topic is real labour. The model is happy to do the boring half of it.
Where they fall apart
Ask for a complete draft and you usually get three failures at once. Invented detail comes first: metrics that were never in your CV, a certification you never earned, a job title quietly inflated. Flattery comes second. “I am excited by your innovative culture” could be pasted into a letter for any company on earth, which is precisely the problem. Then there is rhythm. Every paragraph lands at the same length, every sentence opens with the same subject-verb shape, and the whole thing reads like it was written by someone with no opinions.
That rhythm is the tell. Hiring managers read dozens of these a week and have gotten very good at spotting it. The letter gets skimmed, judged generic, and closed.
The 20-minute workflow that beats a one-click draft
You can get a letter worth sending in about twenty minutes, which is roughly the same time it would take to write one from scratch on a good day and far less than on a bad one. The trick is to make the model do structure and phrasing, and keep the substance for yourself.
- Paste the advert as text, not a link. Models that fetch URLs happily hallucinate the contents. Copy the actual block of requirements.
- Feed it five or six hard facts. Numbers where you have them, names of real systems or teams, a specific result with a date attached.
- Ask for an outline first. Three paragraphs, one line each, explaining what goes where. This is the step most people skip, and it is the one that matters.
- Rewrite the opening line yourself. Everything downstream inherits its voice, so a sentence the model didn’t write is the cheapest way to make the whole letter sound human.
- Ask it to cut 20%. First drafts from every tool I’ve tested run 80 to 150 words too long.
- Read it out loud. Anything you stumble over gets deleted, not fixed.
Specifics beat adjectives every time
“Cut onboarding time from nine days to four by rewriting the first-week checklist” is worth more than every adjective a model can generate about your work ethic. The problem is that most people feed these tools their job titles and nothing else, so the only raw material available is platitudes. Garbage in, “I am a results-driven professional” out.
Before you open the tool, spend five minutes writing down anything you did last year that has a number attached. Revenue, headcount, days saved, tickets closed, budget managed, percentage points of conversion. Six facts is plenty. The letter writes itself once those exist, and the model’s job shrinks to stitching them together.
A prompt that works better than most
Typing “write me a cover letter for this job” gets you exactly what that deserves. Give the model a brief instead. Mine usually looks like this: act as a hiring manager who has read 400 applications today and has no patience left. Here is the advert, here are six facts from my background, and here are two things I want this letter to accomplish. Give me a three-paragraph outline, flag any claim that doesn’t follow from the facts I supplied, and stop at 280 words.
That last instruction does more work than any tone setting. The length cap alone removes most of the padding.
Checks to run before you hit send
The most common failure I see is not bad writing. It’s a leftover company name from the last application, sitting in the first sentence like a billboard. Read the letter once purely for names and titles, then read it again for these:
- Does every claim in the letter appear in your CV or your notes? If not, delete it. Invented metrics end interviews early.
- Could this letter have been sent to a different company with the name swapped? If yes, it isn’t finished.
- Is there one sentence that could only have been written by you? Every good letter has one.
- Is it under 300 words? Length is not a virtue here.
Times to skip the tool entirely
Three situations where the draft will hurt you. Referrals, because the person forwarding your letter knows you and will notice the shift in voice. Small companies, where the founder reads every application and a generic letter is more conspicuous than no letter. And any role that lives on writing quality, since the letter is the work sample, and a hiring editor will clock a generated one in about four seconds.
What’s worth paying for
Honestly, not much. The free tiers of ChatGPT and Claude handle this job as well as anything, and a $25-a-month dedicated cover letter product mostly sells you a nicer form. If you are applying to thirty-plus roles and want saved profiles or bulk tailoring, a paid tier earns its money. Otherwise the subscription is decoration. There’s a wider version of this argument in this breakdown of which AI tools are worth keeping and which to skip, and the logic applies to job-hunting tools as neatly as to anything else.
If you want the bigger picture of how these tools fit into a working week, this practical stack of AI tools for content creators covers the same trade-off from a different angle: pay for the thing you use daily, drop everything you open twice a month.
Track outcomes, not drafts
Here is the part nobody does. Keep a plain spreadsheet with one row per application: date, company, role, which version of the letter you sent, and what came back. After twenty rows you will have data rather than a feeling. Sometimes it turns out the short, blunt version gets interviews and the polished three-paragraph one gets silence. You cannot know that from memory.
It also helps to remember that the screening software everyone fears is not the gatekeeper it’s made out to be. Automated systems get fooled by keyword stuffing and miss obvious quality, and as this experiment with AI image judges showed, these models fail in ways their designers genuinely didn’t anticipate. Which means the letters that win are still the ones a tired human reads and thinks, for a second, that a person wrote it.
So use the generator for the boring 70%. Do the last 30% yourself: the opening line, the one specific detail nobody else could claim, and the cut that takes it from 380 words to 280. That’s the difference between an application and a rejection email you’ll never see.