A growing number of large employers run cover letters through some form of automated scoring before a human reads them. Most of the time it is keyword and semantic-similarity matching against the job ad, not a sophisticated language model. The good news is that the same letter that performs well in that step is also the kind a recruiter wants to read.
You are not trying to game an algorithm. You are trying to write a letter that says, in plain words, the same things the job ad asked for.
What the scoring step looks at
When a cover letter is auto-scored, two signals tend to dominate.
The first is overlap with the job ad: how many of the nouns and noun phrases in the ad appear in your letter. Tools, methodologies, domains, named systems. If the ad mentions Postgres twice and you do not mention it at all, your similarity score drops.
The second is structure: does the letter open with something specific, contain a body that looks like work, and close with a clear next step. Automated readers cannot tell good writing from bad. They can tell whether your letter is mostly filler ("I am thrilled to apply") or mostly substance.
A simple four-paragraph shape
A letter that does well in both human and automated reading tends to follow the same shape. You do not need to label the paragraphs.
Paragraph one. One or two lines naming the role, the company, and a specific reason you are writing to them. Not a generic mission line. Something you read, watched, used, or noticed.
Paragraph two. A short story about a piece of work you have done that maps to the most important requirement in the ad. Use the same vocabulary the ad uses. Numbers where you have them.
Paragraph three. A short paragraph about a second relevant strength, or about how you would approach the role. This is where you can use a couple of words from the "nice to have" list in the ad.
Paragraph four. A short closing with a concrete next step. "I am available for a short call any afternoon this week" or "I have attached my resume and would be glad to walk through any of these projects."
That is the entire structure. About 200 to 300 words. Anything longer starts to drag for the human reader.
Vocabulary to keep, vocabulary to drop
Match the ad's vocabulary as closely as you honestly can. If the ad says "observability stack", use that, not "monitoring and alerting infrastructure". If it says "stakeholder communication", use that, not "talking to people across the business".
Drop the phrases that show up in every AI-generated letter and add no signal:
- "I am thrilled to apply for this opportunity."
- "I am a results-oriented professional with a proven track record."
- "My background aligns perfectly with the role."
- "I would love the chance to bring my passion and expertise to your team."
A recruiter reading dozens of letters in a row tunes those out by lunch. A scorer that uses semantic similarity does not penalize them, but they take up space that could carry your real keywords.
One last pass
Read the letter aloud. If a sentence sounds like something you would never say in an interview, rewrite it in plain words. That single test catches almost every habit you want to break.