I Did Not Scroll LinkedIn to Find My Summer Internship. I Wrote a Prompt.

One Prompt, No Scroll

I did not scroll LinkedIn once to find my summer internship. I wrote one prompt in Claude, connected Apify MCP, and let it do the looking. The output landed me a Mechanical Engineering Intern role at Antora Energy.

I am a solo grad-student founder. My job-hunting hours have to earn their keep. Scrolling does not. A prompt does. Once the ranked spreadsheet was open, I tailored my resume with a resume-tailor skill I had written months ago, sent one InMail, and was on a call within hours.

Why Feed Posts Beat the /jobs Page

A popular /jobs internship typically attracts hundreds of applicants in the first 48 hours. The first review batch usually closes within a few days. By the time most people scroll past it, the role is functionally filled.

A feed post from a hiring manager pulls a much smaller pool in the same window. Often single digits, occasionally a few dozen. Same role, same company, very different math at the top of the inbox.

The structural reason: feed posts are almost always written by the decision maker, not a talent-acquisition inbox. The poster is the person who runs the team, picks the candidate, and reads the InMail themselves. The hidden job market is not hidden. It is on the feed, just not in the place most candidates are searching.

The Prompt

Three rules, one ranking order, one output schema. The original prompt I used was conversational and ran across three turns; this is the cleaned one-shot version that returns the same kind of ranked spreadsheet.

Use Apify to scrape LinkedIn FEED posts (not the /jobs page) from the last 7 days
where a hiring manager or engineering lead is posting that they are hiring an intern.

Field: mechanical engineering, manufacturing, mechanical design, robotics, or any
mechanical-concepts adjacent role.

Region: US only. Exclude Canada and everywhere else.

Signal I want: an actual person on the feed saying "we are hiring", "our team is
looking for", "DM me", or similar. Not an HR-pipeline /jobs listing. Not a recruiter
account.

Rank by, in order:
  1. Recency (newer first)
  2. Low like count (fewer applicants competing)
  3. Whether the poster is the decision maker (hiring manager / engineering lead /
     founder) vs a recruiter

Output: a ranked spreadsheet (xlsx) with columns:
  Role | Location | Field | Posted | Likes | Apply-Method | Poster-Role | Post-URL

Cap at 20 results. Skip any post older than 7 days.

Paste it into Claude, attach the Apify MCP server, and let it run. Swap the field words to match your own.

What the Output Looked Like

The actual run returned 14 ranked posts. Five representative rows below, with poster names and company names redacted so the article reads as a sharable template, not a leaked list.

Role	Location	Field	Posted	Likes	Apply	Poster	Post URL
Mech Eng Intern, Summer 2026	San Francisco, CA	Climate / Energy	1 day ago	7	DM + apply link	Founder	linkedin.com/posts/[redacted-1]
ME Design Intern	Austin, TX	Energy / Manufacturing	2 days ago	12	InMail + apply link	Engineering Lead	linkedin.com/posts/[redacted-2]
Robotics Eng Intern	Boston, MA	Hardware	3 days ago	22	InMail	VP Engineering	linkedin.com/posts/[redacted-3]
Manufacturing Intern	Pittsburgh, PA	Industrial Robotics	4 days ago	31	Comment + DM	Senior PM	linkedin.com/posts/[redacted-4]
Mechanical Eng, Summer	Seattle, WA	Aerospace	5 days ago	48	Apply link only	Director of Engineering	linkedin.com/posts/[redacted-5]

Rows 1 to 3 (under 25 likes, manager-originated, under 4 days old) are the actionable ones. Row 5 is already crowded. Row 4 is borderline.

The InMail That Worked

I sent a tight InMail with two projects and the metrics that mattered. The structure is the point: greeting, shared-context line, two projects with concrete before/after numbers, full-loop framing, secondary credibility, and a closer that says I also applied through the public link.

The structural rule that actually matters is the constraint: everything above the resume attachment fits on one phone screen without scrolling. Two metrics per project. Anything more is a resume; anything less is a vibe.

Hi [First Name],

Saw your post about the [Role] role. I'm a [degree program] student at [school],
and a fellow [shared context — alma mater, hometown, community, prior employer].

I've built two [domain]-adjacent tools at [where].

[Project 1 name] — [one-line of what it does]. [Concrete before/after metric, e.g.
"cut run time from 6 hours to 3, lifted first-pass convergence from 60% to 80%"].

[Project 2 name] — [one-line of what it does], with [optional differentiator, e.g.
"a verification layer that checks every recommendation against the physics solver
before surfacing it"].

Both include the full loop: identifying the bottleneck, prototyping the tool,
tracking before/after metrics, and writing handoff docs so the team keeps using
it after I'm gone.

On the [secondary domain] side, I worked at [prior company] doing [specific
deliverables — tools, scope, system]. [Relevant cert].

Attaching my resume and have also applied through the [company] link in your
post. Happy to walk you through my projects.

[your email]

InMail outperforms cold email on response rate and deliverability, which is part of why feed-post hiring loops close so fast. The other part is structural: a short, specific note from a stranger who already understands the role lands differently when the person reading it is the one who wrote the post.

How to Run This on Your Own Field

Swap “mechanical engineering, manufacturing, robotics” for any field. The US-only, 7-days, feed-posts-only, low-likes-first ranking carries across every field. Test with data science, CS, finance, product, or design. The math of feed-vs-/jobs is field-agnostic.

Aphorism

“The students who treat AI as a second set of hands are the ones still getting offers.”

Reply to this article on LinkedIn with the first result your version of the prompt pulled up. I will DM back with tweaks for your field.

If your first run returns thin results, the fix is almost always the field words. Add three more, drop one that is too generic, and re-run.