Most “personalized” LinkedIn outreach is still a template with a first-name token and a company mention. Prospects recognize it quickly, and reply rates reflect that.
PhantomBuster’s AI LinkedIn Message Writer generates messages from real profile context, but treat it as one step in a repeatable outreach workflow—not a standalone shortcut. Effective AI outreach is a workflow problem, not just a writing problem.
Message quality depends on strong input data, clear prompt constraints, and a review step before anything is sent. Layer your workflow first, then scale once it’s stable (Brian Moran, Product).
This guide explains what data to provide, how to write reliable prompts, and how to scale responsibly.
What does the AI LinkedIn Message Writer do—and what doesn’t it do?
The generation layer, not the sending layer
PhantomBuster’s AI LinkedIn Message Writer reads your lead data and generates a unique message for each prospect. It adds the output to a new ai_message column in your results file.
You’ll map the ai_message column as the message body in PhantomBuster’s sending automations. It doesn’t send messages. Use PhantomBuster’s LinkedIn Auto Connect, LinkedIn Outreach Flow, or LinkedIn Message Sender automations to deliver approved messages as part of the same workflow. This separation is intentional.
It creates a clear checkpoint between message generation and execution, giving you time to review quality, correct errors, and confirm that each message fits the campaign.
What inputs does the AI use?
The model receives only the columns you select as inputs. If those columns contain limited or generic information, the messages will be generic too.
Choose the higher-capability model for complex constraints. Run a 50-lead A/B test comparing reply rate and error rate per credit before rolling out. In a reliable outreach system, PhantomBuster’s AI LinkedIn Message Writer sits in the middle of the workflow:
- Define the audience
- Export leads
- Extract and enrich profile data
- Generate messages
- Review the output
- Send
What data should you feed the AI?
Which LinkedIn fields matter most for first-touch messages?
PhantomBuster’s LinkedIn Profile Scraper extracts more than 47 data points, but not every field improves a first-touch message. For first-touch messages, prioritize:
- Headline, which often reveals the prospect’s focus, responsibilities, or positioning
- Current job title and company, which help confirm relevance
- Location, when regional context genuinely matters
- Recent experience entries, including the two most recent roles
Add PhantomBuster’s AI LinkedIn Profile Enricher step to your flow to generate a concise career summary or custom signals the Message Writer can use for sharper hooks. Treat it like a mail merge with better inputs: richer fields → sharper hooks → higher reply rate.
Which engagement signals drive higher intent?
PhantomBuster’s LinkedIn Post Commenters Export automation captures each prospect’s comment text. This gives the AI a specific opinion, question, or topic to reference instead of producing a vague line such as “I saw you engaged with this post.”
PhantomBuster’s LinkedIn Post Likers Export adds reaction data. Without comment text, personalization is limited to lighter context (e.g., topic relevance). This is the minimal data → message flow:
- PhantomBuster’s LinkedIn Search Export or Sales Navigator Search Export
- PhantomBuster’s LinkedIn Profile Scraper
- PhantomBuster’s AI LinkedIn Profile Enricher (optional)
- PhantomBuster’s AI LinkedIn Message Writer
Start with the Profile Scraper (and optional Enricher). Use the Post Commenters Export or Activity Extractor for higher-intent personalization when available.
| Data source | What it adds | Personalization value |
|---|---|---|
| PhantomBuster — LinkedIn Profile Scraper | Headline, title, company, location, recent roles | Core relevance signals |
| PhantomBuster — AI LinkedIn Profile Enricher | Summarized career context and custom outputs | Deeper hooks and framing |
| PhantomBuster — LinkedIn Post Commenters Export | Actual comment text | Specific, high-intent references |
| PhantomBuster — LinkedIn Activity Extractor | Recent topics and engagement data | Activity-based conversation starters |
How to write prompts that produce usable messages
What prompt structure scales across large lists?
A prompt that remains reliable across a large list needs three elements:
- Role definition: Tell the AI which perspective or persona to adopt.
- Task specification: State exactly what it should produce.
- Constraints: Define the tone, length, source boundaries, and prohibited content.
These instructions reduce ambiguity. The model is more likely to remain consistent when “good” is clearly defined and the boundaries are explicit.
Prompts for connection requests: short-form
Character limits differ by account type. Check your limit and set the maximum length in your prompt. State the maximum length. Without a clear limit, generated messages may be truncated when sent.
Example prompt for connection requests:
You are a sales representative at [Company]. Write a LinkedIn connection request note to the prospect. Use only the provided data. Keep the message under 180 characters. Be conversational, not salesy. Do not pitch or request a meeting. Reference their role or company only when relevant. Do not use exclamation points.
Example output:
Hi Sarah, saw you’re leading ops at Acme. Open to connecting to swap notes on ops workflows?
Prompts for follow-up messages: longer-form
Messages to first-degree LinkedIn connections allow more space, but that does not mean you should use it all. Focused follow-ups are easier to read and answer than long introductions. Specify the tone, length range, and desired next step.
State whether the message should end with a soft question, resource offer, or meeting request. Soft questions lower response friction and improve reply rate on first touch compared with immediate meeting asks.
Example prompt for follow-up messages:
Write a follow-up LinkedIn message to a new connection. Reference their headline and one recent role. Keep the message between 80 and 150 words. Be helpful and direct, not pushy. End with one soft question rather than a meeting request. Do not invent details that are not included in the input data.
Prompt discipline matters: The AI cannot personalize messages using fields you did not extract. Confirm which columns your upstream automations populate before writing the prompt.
The full workflow: from search to send
Step 1: Build the lead list
Use PhantomBuster’s LinkedIn Search Export or Sales Navigator Search Export to collect prospects who match your ideal customer profile. Define the audience before increasing volume.
Use filters such as seniority, geography, company size, function, or industry to keep the list relevant. Results flow into the PhantomBuster Leads page. Review and remove duplicates before enrichment.
Step 2: Enrich with profile data
Run PhantomBuster’s LinkedIn Profile Scraper on the exported profiles to collect the context required for personalization. Review the output before continuing. Check for missing headlines, outdated roles, company mismatches, emojis in names, or fields that would sound unnatural inside a message.
For more context, use PhantomBuster’s AI LinkedIn Profile Enricher to summarize experience or identify a relevant signal. Only create fields that directly support the campaign.
Step 3: Generate the messages
Add the cleaned and enriched file to PhantomBuster’s AI LinkedIn Message Writer. Select only the columns the model needs. Too little context creates generic messages, while too many unrelated fields can distract the model.
Minimal starter set: headline, title, company, last 1–2 roles
Upgrade set: recent comment text, activity topics Test on 30–50 leads. Check length compliance, factual accuracy, repeated openings, and tone consistency before scaling. Once the prompt works consistently, run it across the full list.
Step 4: Review the output
Do not send generated messages without checking them. Review a representative sample from every audience segment. Look for invented claims, awkward references, repeated phrasing, incorrect names, overly familiar language, and calls to action that feel too aggressive.
Pay particular attention to sparse profiles. When input data is incomplete, the safest output may be a simple, relevant message rather than forced personalization. Teams that review a 10–20% sample before sending keep quality high without slowing volume.
Step 5: Send with a PhantomBuster automation
After approval, connect the output to the appropriate sending automation. Use PhantomBuster’s LinkedIn Auto Connect for connection notes and LinkedIn Outreach Flow or LinkedIn Message Sender for follow-ups.
Map your ai_message column as the message source. Start with a controlled batch. Monitor acceptance rates, replies, message quality, and any signs of account friction (captchas, unusual verification prompts, or throttling) before increasing activity.
LinkedIn doesn’t publish fixed limits. Ramp volume gradually, keep message patterns varied, and monitor acceptance and reply trends before scaling. Avoid sudden jumps in activity, especially on accounts with limited previous usage. Build the workflow in layers, then increase volume gradually.
Frequently asked questions
Does the AI LinkedIn Message Writer send messages automatically?
No. PhantomBuster’s AI LinkedIn Message Writer generates message text and adds it to an ai_message column in your results file. You need PhantomBuster’s LinkedIn Auto Connect, LinkedIn Outreach Flow, or LinkedIn Message Sender to send the messages.
Can the AI reference a prospect’s recent LinkedIn activity?
Yes, but only when that activity has been extracted and included in the selected input columns. The model cannot access information that is not provided. Use PhantomBuster’s LinkedIn Post Commenters Export or LinkedIn Activity Extractor to capture this data upstream.
Should I review every generated message?
Review the complete output for small campaigns and a representative sample for larger ones. Check high-value accounts, sparse profiles, unusual data, and messages generated from new prompts. A 10–20% sample keeps quality high without slowing volume.
What columns should I select as inputs for best personalization?
Start with headline, title, company, and the last 1–2 roles. Add comment text or activity topics for higher-intent personalization. Avoid unrelated fields that don’t support the campaign—too many columns can dilute the model’s focus.
How do I A/B test prompts and models in PhantomBuster?
Split your list into two segments. Run one version of the prompt (or one model) on segment A and the alternate version on segment B. Compare reply rate, error rate, and tone consistency across 30–50 leads per variant before scaling the winner.
What’s the safest daily sending ramp for a new account?
Start with 10–15 connection requests per day for the first week. Monitor acceptance rate and account friction. If stable, increase by 5–10 per day each week until you reach a steady-state volume that maintains high acceptance and reply rates.
How do I map the ai_message column to LinkedIn Outreach Flow?
In PhantomBuster’s LinkedIn Outreach Flow settings, select the ai_message column as your message body source. The automation will read each row’s generated message and send it to the corresponding prospect. Test with 3–5 leads before scaling.
What should I do when a profile has minimal data?
Write a prompt that defaults to a simple, relevant message when input data is sparse. Avoid forcing personalization when only generic fields are available. A short, direct connection note performs better than a message filled with invented context.
Conclusion
The workflow is straightforward: export leads, enrich profiles, generate messages, review output, then send. PhantomBuster’s AI LinkedIn Message Writer works best when you feed it strong input data and write prompts that define clear constraints.
Quality outreach starts with the data layer. Richer fields create sharper hooks, which improve reply rates. Review a sample before scaling, ramp volume gradually, and monitor reply trends to refine the system over time.
Learn more about automating LinkedIn outreach or Start your free trial