Inbound LinkedIn leads are easy to collect, but slow to qualify. If you treat every inbound profile the same, your response time slips and your best-fit leads wait in the same queue as everyone else. A lot of advice online puts all the weight on prompt writing. In practice, the prompt is only one component.<\/p>\n
If you paste messy profile text into ChatGPT and ask for a 0 to 100 score, you usually get a number with no consistent rubric, no evidence trail, and no routing action you can run in a real workflow. Reliable AI lead scoring<\/a> starts with structured profile data, a clear ICP rubric, evidence-based reasoning, and outputs designed for routing.<\/p>\n Below you’ll find six copy-paste prompts by use case, plus a simple workflow that connects enrichment, scoring, and routing into something you can run every day.<\/p>\n Pasting unstructured profile text into a prompt creates inconsistent inputs. The model sees different field orders, missing sections, and formatting noise each time. Without normalized fields like job title, company, industry, and location, the model is not able to score against your rubric.<\/p>\n A headline might show up first for one profile and buried later for another, which changes what the model “pays attention” to. The fix usually isn’t a fancier prompt. It’s better data. Extract the profile into consistent fields before the model ever sees it, then scoring becomes much more predictable.<\/p>\n A 0 to 100 score without evidence is hard to use. You can’t review it, calibrate it, or explain why you skipped a lead. When prompts don’t require the model to quote from the profile, it fills gaps with assumptions. The resulting output can sound reasonable, but might have little to do with the data you provided. The fix is simple: require quoted evidence. If the model can’t find support in the profile fields you gave it, it should return “Needs review” instead of guessing.<\/p>\n A score needs to map to an action like reply now, nurture, ignore, or review. You need outputs that link to a queue, a CRM field, or a Sheets column. Generic prompts often return paragraphs. That forces manual re-entry and breaks automation. Ask for action-ready fields (status, reason, next_step, optional JSON) that map to your PhantomBuster \u2192 Sheets\/CRM handoff. The goal isn’t insight, it’s execution.<\/p>\n Translate your ICP into explicit criteria:<\/strong> Must-haves, disqualifiers, and score bands. If you don’t define “qualified,” the model will use its own assumptions. Be specific about fields you can extract:<\/strong> Target titles, industries, company size ranges, geographies, and keywords that signal relevant responsibility. Example: “Tier 1 requires VP-level Sales or RevOps titles at SaaS companies with 50 to 500 employees in North America, plus a mention of ‘outbound,’ ‘pipeline,’ or ‘sales automation’ in recent experience.”<\/p>\n Add a line in every prompt that supports the scoring:<\/strong> “Quote specific phrases from the profile fields provided.” This makes scores auditable. You can see whether the model is using strong signals or stretching weak ones.<\/p>\n Define the output structure:<\/strong> Status label, one-sentence reason, next-step recommendation, and optionally JSON for ingestion. Strict formatting keeps outputs machine-readable and prevents the model from drifting into prose.<\/p>\n Not every lead is a clear yes or no. Your prompt should support a “Needs review” outcome for mixed signals, missing fields, or ambiguity. This keeps AI in the assist role.<\/p>\n “Automation should amplify good behavior, not replace judgment.” \u2014 PhantomBuster Product Expert, Brian Moran<\/a><\/p><\/blockquote>\n Prompt design checklist:<\/strong><\/p>\n Use case:<\/strong> You receive dozens of inbound connections daily and need a fast output to decide who gets a reply first. Prompt:<\/strong> You are a sales triage assistant. Evaluate this LinkedIn profile against our target audience: B2B SaaS founders and sales leaders at companies with 10 to 100 employees. Profile data<\/strong> (structured fields):<\/p>\n Output format<\/strong> (one line only): STATUS: [Hot\/Warm\/Cold\/Needs_Review] | REASON: [one sentence, cite a short quote if possible] Status definitions:<\/strong><\/p>\n Key design choice:<\/strong> Minimal output for speed. This works as a first filter before you spend time on deeper scoring.<\/p><\/blockquote>\n Use case:<\/strong> You need a full breakdown you can review, calibrate, or share internally. Prompt:<\/strong> You are a lead qualification analyst. Score this LinkedIn profile against our ICP using ONLY the data provided. Do not infer company size, budget, or intent unless it is explicitly stated. ICP definition:<\/strong><\/p>\n Profile data<\/strong> (structured fields):<\/p>\n Scoring rubric:<\/strong><\/p>\n Output format:<\/strong> SCORE: [number] STATUS: [Qualified\/Unqualified\/Needs review] JUSTIFICATION (3 bullets, each includes a short direct quote):<\/p>\n RED FLAGS: [bullets, or “None found in provided fields”] GREEN FLAGS: [bullets, or “None found in provided fields”] NEXT_STEP: [reply now \/ ask 1 clarifying question \/ nurture \/ ignore]<\/p>\n Key design choice:<\/strong> The evidence requirement makes scoring reviewable and reduces guesswork.<\/p><\/blockquote>\n Use case:<\/strong> You need a split view: can they drive a purchase, and do they likely feel the pain. Prompt:<\/strong> You are an enterprise sales analyst. Score this profile on two dimensions: Authority and Need. Use ONLY the data provided. If evidence is missing, say “Unknown” and route to Needs review. Your solution<\/strong> (one sentence): [describe your solution] Profile data<\/strong> (structured fields):<\/p>\n Authority score (1 to 10): Consider: seniority, scope, team ownership, budget responsibility. Return a score AND a short direct quote as evidence, or “Unknown” if not stated. Need score<\/strong> (1 to 10): Consider: explicit problems, goals, keywords tied to your solution. Return a score AND a short direct quote as evidence, or “Unknown” if not stated. Output format:<\/strong> AUTHORITY: [1-10 or Unknown] | EVIDENCE: “[quote or Unknown]” NEED: [1-10 or Unknown] | EVIDENCE: “[quote or Unknown]” STATUS: [Hot\/Warm\/Cold\/Needs_Review] NEXT_STEP: [reply now \/ ask clarifying question \/ nurture \/ ignore]<\/p>\n Key design choice:<\/strong> Two-axis scoring separates “can they buy” from “do they need it,” which helps prioritize follow-up in longer sales cycles.<\/p><\/blockquote>\n Use case:<\/strong> You want scoring plus a draft reply, without turning it into a pitch. Prompt:<\/strong> You are a sales assistant. First, score this lead against our target persona. Then, only if the lead is a fit, draft a short icebreaker. Persona:<\/strong> Sales leaders at tech companies with 20 to 200 employees who care about pipeline efficiency. Profile data<\/strong> (structured fields):<\/p>\n Step 1:<\/strong> Score (1 to 5 stars)<\/p>\n Step 2:<\/strong> If score is 3 stars or higher, write an icebreaker message:<\/p>\n Output format:<\/strong> SCORE: [stars] EVIDENCE: “[short quote used for the decision]” ICEBREAKER (only if 3+ stars): [message]<\/p>\n Key design choice:<\/strong> The icebreaker stays professional and light. The quote requirement keeps the personalization anchored to real profile details. For more on crafting effective outreach, see how AI personalization can improve LinkedIn reply rates<\/a>.<\/p><\/blockquote>\n Use case:<\/strong> You score leads via Zapier, Make, or a script, and need clean fields that map to your CRM or Sheets. Prompt:<\/strong> You are a data formatter. Evaluate this LinkedIn profile against our target audience and output ONLY valid JSON. No markdown. No explanation. Use empty strings for missing string fields; use null only for numeric or boolean fields where explicitly marked nullable. Do not guess. Target:<\/strong> Marketing and sales leaders at B2B SaaS companies. Profile data<\/strong> (structured fields):<\/p>\n Required JSON schema: { “lead_name”: “string”, “profile_url”: “string”, “current_company”: “string”, “current_title”: “string”, “industry”: “string”, “location”: “string”, “icp_score_1_to_100”: number\u00a0| null, “qualification_status”: “Qualified” | “Unqualified” | “Needs review”, “is_decision_maker”: boolean\u00a0| null, “primary_reason”: “string”, “evidence_quotes”: [“string”, “string”] } Output JSON only.<\/p>\n Key design choice:<\/strong> Strict JSON keeps the workflow reliable. Clear null handling avoids type conflicts in downstream automations.<\/p><\/blockquote>\n Use case:<\/strong> You want the model to call out mixed signals instead of forcing a yes or no. Prompt:<\/strong> You are a lead triage assistant. Classify this profile into one of three categories: Qualified, Unqualified, or Needs review. ICP:<\/strong> Sales and marketing leaders at B2B companies with 50 to 500 employees. Profile data<\/strong> (structured fields):<\/p>\n Classification rules:<\/strong><\/p>\n Output format:<\/strong> STATUS: [Qualified\/Unqualified\/Needs review] REASON: [one sentence, include one short quote if possible] MISSING_FIELDS: [comma-separated list, or “None”] NEXT_STEP: [reply now \/ ask 1 clarifying question \/ nurture \/ ignore]<\/p>\n Key design choice:<\/strong> The review lane reduces false positives and keeps human judgment where the data is incomplete.<\/p><\/blockquote>\nWhy most AI lead scoring fails before the prompt runs<\/h2>\n
Why does raw profile text create inconsistent inputs?<\/h3>\n
Why an opaque 0\u2013100 score fails in sales operations<\/h3>\n
What happens when scores lack routing rules?<\/h3>\n
What’s the minimum prompt design that ops teams can trust?<\/h2>\n
Define your ICP rubric as a scoring contract<\/h3>\n
Require quoted evidence from specific fields<\/h3>\n
Specify an output format you can route<\/h3>\n
Add a “Needs review” lane for borderline cases<\/h3>\n
\n
Copy-paste prompts by operational use case<\/h2>\n
Prompt 1: Quick triage for high-volume inbound<\/h3>\n
\n
\n
Prompt 2: Detailed ICP scoring with evidence<\/h3>\n
\n
\n
\n
\n
Prompt 3: Authority and need scoring for enterprise cycles<\/h3>\n
\n
Prompt 4: Score the lead, then write an icebreaker<\/h3>\n
\n
\n
\n
Prompt 5: JSON output for automation pipelines<\/h3>\n
\n
Prompt 6: Borderline lead classification with a review lane<\/h3>\n
\n
\n