On this page
The best person for your open role already has a job, and they are not reading job boards. This playbook runs the same two-prompt hunt on either of two platforms, Perplexity Computer or Claude with Research turned on, and shows you which one to pick first.
🎯 You get: a ranked shortlist of 25 real candidates, top 5 with approach notes
✅ Bring: a Claude Pro or Max plan OR Perplexity with Computer access (or both, to test) · your job description · Perplexity users, confirm your credit balance at perplexity.ai/computer before you start (Pro includes a one-time 4,000-credit bonus, not a monthly allowance)
👤 Best for: Anyone hiring right now with a real job description ready to paste in. Skip if you're still defining the role, because a vague JD just produces fuzzy dream-candidate profiles downstream.
🎯 What you'll be able to do
By the end of this playbook:
What you're building toward (illustrative example, your real results will differ):
| Name | Score | Platform | Best Proof | Why They Fit |
|---|---|---|---|---|
| Alex R. | 9/10 | GitHub | Shipped a 400-star open-source tool solo | Active in the exact Discord servers your scorecard flags |
| Jamie T. | 8/10 | Led a 3-person team to a public case study | Posts weekly about the problem your role solves |
📚 Background — why this works
One thing to know before you start hunting.
What the two prompts do
The Scout turns your job description into a dream-candidate profile, a platform map, and a qualification scorecard.
The Hunt takes that kit into the field and comes back with 25 real people ranked by score.
Both prompts run the same on either platform.
🧭 Why I choose what
Both tools run this play. Here is how we pick between them at TUB.
Perplexity Computer: the stronger social hunter
Computer is a cloud agent that runs the whole job on its own and spawns sub-agents when it hits a problem. You close your laptop and come back to a finished shortlist.
In our work at TUB, it is much better at hunting social platforms like LinkedIn and Instagram. It searches them the way an insider would and comes back with fuller candidate tables.
Login gates still hide most profiles from it, so even Computer only sees public data.
Claude + Research: the smarter buy for most hires
Research mode sends Claude on a multi-step web investigation and returns a cited report. It comes with every paid Claude plan, including the $20 Pro plan.
It is weaker on social platforms, so expect thinner LinkedIn and Instagram tables.
The archetype, platform map, and scorecard come out every bit as sharp, and for most roles the shortlist is still strong.
The money math
Perplexity Pro at $20 a month includes Computer access, but only a one-time 4,000-credit bonus. Recurring credits mean Max at $200 a month.
For one or two hires a year, that math rarely works. Run the play in Claude with the Pro plan you may already have and spend nothing extra.
Hiring every month, or filling social-heavy roles like marketing and content, is when Max starts to earn its keep.
The login-gate workaround: Claude for Chrome
Both tools get stopped by login walls, but the Claude for Chrome extension works inside your own browser, where you are already logged in.
Turn it on and Claude can open LinkedIn or Instagram as you, scroll the profiles the Hunt flagged, and pull what the public web hides.
Use it as a second pass on your top candidates rather than the whole 25, since you supervise it live in your browser.
Go gently on LinkedIn: its anti-bot systems have flagged accounts running browser agents too aggressively. A handful of profile visits is fine, bulk scraping is not.
🧭 Part 1: Pick your platform
One decision before anything else. Every step after this is the same on both platforms.
Step 1: Pick your platform
Pick Perplexity Computer if you are on Max or filling a social-heavy role. Pick Claude with Research if you make occasional hires and already pay for Claude Pro.
When in doubt, start in Claude. You can rerun the Hunt in Computer later if the shortlist comes back thin.
🛠️ Part 2: Build your hunting kit
The Scout turns your job description into three things: a dream-candidate profile, a platform map, and a qualification scorecard.
Step 2: Open your workspace
Claude: go to claude.ai and start a new chat, then turn on Research mode from the tools menu.
Perplexity: go to perplexity.ai/computer and log in. You'll land on the Computer screen with a prompt box for describing your task.
Claude users with a Project loaded with business context: start the chat inside it. The archetype comes out sharper.
Confirm it worked: you see the blue Research indicator at the bottom of your Claude chat, or you're on Perplexity's Computer task screen. If neither shows up, don't paste the Scout prompt yet.
Step 3: Paste the Scout prompt
Copy the prompt below, fill in the four brackets, and send it. The same prompt works on both platforms.
Role: You are an elite headhunter and talent scout.
You find A-players who are not looking for jobs, and you know that the best people in any field cluster in specific corners of the internet.
Context: I am hiring for this role: [paste full job description].
Here is the person I picture in this seat: [describe them: what they do all day, what they've already built, what they care about, what would make them say yes].
Companies or people whose talent I admire: [name 2-3 — this calibrates the quality bar].
Do NOT return people from: [competitors, sectors, or companies to exclude — or "none"].
A-players in this role are not on job boards, so I need to hunt them where they already live.
Request: Do four things, in order.
Sharpen my picture into a dream-candidate profile I can hunt with. Distill it to: must-have signals (5 max), nice-to-haves (5 max), and deal-breakers. Go past the resume: describe their day, their public work, and the proof an A-player leaves behind. Anchor "great" to the caliber of the companies I named above.
Research where these people live online. Find the 3 to 5 platforms and communities where the best ones are most active RIGHT NOW — and at least 2 must be non-obvious channels (Discord servers, Substack comment sections, niche forums, professional communities), not just LinkedIn and Instagram. Do not guess from memory. Verify with sources from the last 6 months, and tell me how an insider searches each one.
Write the exact searches to run on each platform, using only the operators that platform supports (LinkedIn allows only AND, OR, NOT, quotes, and parentheses — no wildcards). Give me 2 variations per platform.
Build a qualification scorecard: the 5 to 7 public signals that separate an A-player from an average one on those platforms. For each signal, show what "great" looks like with one real example.
Clarifying Questions: Before you start, flag any part of the job description too vague to hunt against and ask me one question per flag, up to 5 questions total. Ask anything that would change who you hunt for or where you look.Confirm it worked: within seconds you should see up to 5 clarifying questions, not a finished report. If it starts researching immediately with no questions, tell it "ask me your 5 questions before you start" and resend.
Step 4: Answer the questions, then let it run
Both tools ask up to 5 questions about the role, the budget, and the person you picture. Answer them.
The run takes 5 to 15 minutes on most roles. Leave the tab open and check back.
Step 5: Save the three outputs
Copy the dream-candidate profile, the platform map, and the qualification scorecard into one doc.
You'll paste all three into the Hunt prompt next.
Keep this doc after the hire. The next time that role opens up, you skip the Scout and go straight to the Hunt.
🕵️ Part 3: Run the hunt
The Hunt opens each platform on your map, searches like an insider, and comes back with 25 real people.
Step 6: Paste the Hunt prompt
Copy the prompt below and paste your Scout outputs into the bracket. Send it in the same chat or task.
Role: You are the same elite headhunter, and now you are in the field with your own computer.
Context: Here is the dream-candidate profile, the platform map, the searches, and the qualification scorecard from our scout run: [paste all of it].
I want real, reachable humans, not a summary of what candidates like this look like.
Request: Open each platform on the map and hunt.
Run the insider searches: the hashtags, communities, repos, leaderboards, and threads where this exact person shows up. Prioritize people active in the last 12 months, and flag anyone whose latest public activity is older.
Find 25 real people who match the profile. Respect my exclusion list — wasted slots cost me real candidates.
VERIFICATION IS THE JOB: only include a person if you opened their profile during this run and the link resolves to them. If you cannot verify someone is real, drop them. A verified list of 14 beats a padded list of 25. Never invent a name, a URL, or a piece of work — if you are not sure, write "not verified" and move on.
Score every verified person against the qualification scorecard, and show the evidence link behind each score.
For each person give me: their name, links to every profile you verified, their score, the single best piece of public work that proves they fit, and one line on why they match. Return it all in a table sorted by score.
Then for your top 5: draft a three-sentence outreach message that names one specific piece of their public work. If a message could be sent to anyone else on the list, it is a template — rewrite it.
Clarifying Questions: If a platform blocks you or a talent pool looks thin, stop and tell me. Suggest where else to look instead of padding the list with weak matches.
When I reply "go again," find 10 more verified people, excluding everyone you already showed me and anyone I rejected.Confirm it worked: the tool should visibly open and search specific platforms, not hand back a written summary instantly. An instant answer with no visible searching means it likely pulled from memory. Stop and resend with "go verify this live, do not answer from memory."
If credits are burning fast with no progress, see Trip-Up #1 below.
Step 7: Let it run, then check in once or twice
Perplexity users can close the laptop, because Computer runs in the cloud. Claude users, leave the tab open while Research works.
Perplexity users: watch the progress panel. If it repeats the same step over and over, stop the task, because it is burning credits without progress.
Thin results on LinkedIn or Instagram are expected. See Trip-Up #2 for the fix.
Step 8: Review the shortlist
You get a table of 25 people sorted by score, plus a top 5 with approach notes.
Every row includes profile links and the one piece of public work that proves the fit.
Need more names, reply "go again" and it finds 10 more verified people, excluding everyone you've already seen.
✅ Part 4: Verify before you reach out
Ten minutes of checking protects you from messaging the wrong person.
Step 9: Open every link on the top 5
Click each profile link and confirm the person, the work, and the numbers are real.
AI can attach the right work to the wrong person with the same name. Catch it here, not in a DM.
Before you message anyone, see Trip-Up #4 on name collisions.
Step 10: Send your first outreach
Start with the approach notes on your top 5, and build each message around the specific piece of public work that impressed you.
A-players answer people who did their homework.
📖 Reference: What the prompts produce
You don't need this section to run the playbook. The prompts build all of it for you. Open these toggles if you want to understand each piece, or build one by hand.
The dream-candidate profile
Goes past the resume: what this person does all day, what they've already built, and the proof an A-player leaves behind in public.
The sharper this profile, the better every downstream search gets.
The platform map
The 3 to 5 places where the best people in this field post their work, verified by live research rather than memory.
For an AI developer that might be GitHub, X, and Hugging Face. For a social media manager, Instagram, TikTok, and LinkedIn.
The qualification scorecard
5 to 7 public signals that separate an A-player from an average one, with what "great" looks like for each.
All 25 candidates get scored against it, and you can reuse it on inbound applicants too.
The shortlist table
25 real people ranked by score, each with profile links, the best piece of public work, and one line on the fit.
The top 5 come with a note on how to approach each one.
⚡ 4 Things That Will Trip You Up
Trip-Up #1 — Computer keeps burning credits after a silent failure
A blocked site or an expired login can send Computer into a retry loop.
It keeps spawning sub-agents without telling you, and each one costs credits.
Fix: Check the progress panel during the run. If it repeats the same step, stop the task and restart with that platform removed.
Trip-Up #2 — LinkedIn and Instagram hide most profiles behind a login
Both platforms only see public data, so locked-down sites return thinner results.
Perplexity gets further than Claude here, and it still hits the gates.
Fix: Rerun the thin platform with Claude for Chrome, which browses in your logged-in session, or ask where else this talent pool posts.
Trip-Up #3 — Research drains your Claude usage limits fast
One Research run pulls dozens of sources, so it counts against your plan limits faster than a normal chat.
Fix: Run the Scout once and save the outputs. Rerun it only when the role changes.
Trip-Up #4 — The shortlist can name the wrong person
Two people with the same name can get merged into one candidate.
It happens most with common names and thin public profiles.
Fix: The Hunt prompt forces the agent to verify every link during the run. Still open your top 5 yourself — the agent's "verified" is a filter, not a guarantee.
Send one outreach to a verified A-player who wasn't on your radar this morning. The Scout prompt (Step 3) builds your hunting kit, the Hunt (Step 6) surfaces the candidates, and Step 10 delivers the first message.
✅ You're done when
📌 The bottom line
A-players aren't on job boards. Used right, Claude Research and Perplexity Computer become your headhunter. A job description goes in. 25 verified candidates come out, plus a first message built for the person, not the template. Less scrolling. More closing.
