Accounts

Writing effective prompts

The quality of your AI enrichment depends almost entirely on how well you write your prompt. A great prompt produces consistent, useful output across hundreds of rows. A vague one wastes credits and gives you data you can't use. This guide covers the principles and patterns that work.

Core principles

Be specific

Tell the AI exactly what you want. Specify the format, length, tone, and structure of the output.

Bad:

Write something about this person.

Good:

Write a 2-sentence personalized connection request for {{firstName}}, who works as {{jobTitle}} at {{companyName}}. Use a professional but friendly tone. Reference their role specifically.

Provide context

Give the AI enough background to produce relevant output. Explain who the audience is and how the output will be used.

Bad:

Score this lead.

Good:

You are a B2B sales analyst. Score this lead from 1-10 based on fit for our project management software.
Consider: {{jobTitle}} (decision-maker titles score higher), {{companySize}} (50-500 employees is ideal),
and {{industry}} (tech and professional services are best fit). Return only the number and a one-sentence justification.

Use column variables

Reference other columns with {{columnName}} to personalize each row's output. The more relevant context you provide, the better the results.

Summarize {{companyName}} in one paragraph for a sales rep preparing for a call.
They are in the {{industry}} industry, have {{employeeCount}} employees,
and {{firstName}} {{lastName}} works there as {{jobTitle}}.

Set the output format

Explicitly state what the output should look like. This keeps things consistent across all rows.

Return your response in this exact format:
Score: [1-10]
Reason: [one sentence]
Suggested Action: [one of: Reach Out, Nurture, Skip]

Keep it conversational when needed

For outreach messages, tell the AI to write like a human, not a robot.

Write as if you're a real person sending a quick LinkedIn message. No corporate jargon.
Don't start with "I hope this message finds you well." Keep it under 50 words.

Common prompt patterns

Lead scoring

Rate this lead from 1 to 10 for fit with a [your product type] solution.

Data points to evaluate:
- Job Title: {{jobTitle}} (higher score for VP, Director, Head of, C-suite)
- Company Size: {{employeeCount}} (ideal range: 50-1000)
- Industry: {{industry}} (best fit: technology, SaaS, professional services)

Return format:
Score: [number]
Reason: [one sentence explaining the score]

Personalized outreach message

Write a 3-sentence cold email opening to {{firstName}}.
- They are {{jobTitle}} at {{companyName}}.
- Reference their industry ({{industry}}) naturally.
- Our product helps teams automate [describe your value prop].
- Tone: conversational, not salesy. No buzzwords.
- Do NOT start with "I noticed that..." or "I came across your profile."

Company summary

Write a 2-3 sentence summary of {{companyName}} for a sales rep preparing for a discovery call.
Include what the company does, their target market, and any notable details.
Industry: {{industry}}. Size: {{employeeCount}} employees.
Focus on information relevant to selling [your product type].

Classification

Classify {{companyName}} into exactly ONE of these categories based on their industry and size:
- Enterprise (1000+ employees, established industry)
- Mid-Market (100-999 employees)
- SMB (under 100 employees)
- Startup (under 50 employees, technology/innovation focused)

Industry: {{industry}}
Employees: {{employeeCount}}

Return only the category name, nothing else.

Sentiment analysis

Analyze the sentiment of this LinkedIn headline: "{{linkedinHeadline}}"

Classify as: Positive, Neutral, or Negative.
Then note if it suggests they are: Job Seeking, Actively Building, Thought Leader, or Standard Professional.

Format:
Sentiment: [Positive/Neutral/Negative]
Profile Type: [category]

Testing and iterating

  1. Run your prompt on 5 rows and read every result carefully.
  2. Check if the format is the same across all results. Are some outputs too long or too short?
  3. Test edge cases. What happens when a column variable is empty? Does the AI handle missing data gracefully?
  4. Adjust your prompt based on what you see, then test again. Repeat until the output is consistently good.
  5. If the AI occasionally goes off track, add explicit constraints: "Do NOT include...", "Always start with...", "Maximum 50 words."

Common mistakes

Asking for too much in one prompt

Problem: Cramming multiple tasks into a single prompt produces messy, inconsistent output.

Fix: Create separate AI columns for each task. One column for scoring, another for messaging, another for classification.

Using wrong column names

Problem: Typing {{first_name}} when the column is actually named {{firstName}} produces empty values.

Fix: Check your exact column names in the table header before writing the prompt. Column variable names are case-sensitive and must match exactly.

Being vague about tone

Problem: Saying "write a nice message" gives you wildly different results for each row.

Fix: Be explicit. "Professional but warm. Short sentences. No exclamation marks. Read like a peer-to-peer message, not a sales pitch."

Not specifying output length

Problem: Some rows get 3 words, others get 3 paragraphs.

Fix: State the expected length clearly. "2-3 sentences", "under 50 words", "exactly one paragraph."

Skipping the test step

Problem: Running on 500 rows with an untested prompt, then discovering the output is unusable.

Fix: Always test on 5-10 rows first. The small credit investment saves you from wasting a large batch.

Not handling empty data

Problem: When a referenced column is empty, the AI generates awkward or incorrect output.

Fix: Add instructions for handling missing data: "If {{companyName}} is empty, skip the company reference and focus on their role."