<- all tokdocs

Using AI to Network on LinkedIn: Referrals, DISC and the DM Rules

By , founder of Mavric and Apso

View on TikTok ->

A TikTok from @angusthenontechnical shows a workflow where Claude reads your LinkedIn connections, sorts people into four writing styles, and drafts networking messages aimed at getting a referral. Referrals do convert to hires far more often than job board applications. The video ends with "Go have your AI DM people," which is the one step LinkedIn's User Agreement prohibits.

Do referrals really get more people hired?

Yes. The video's evidence is a Google AI Overview, which is a summary of social posts, so recruiting funnel data is a better check. The 2019 Jobvite Recruiting Benchmark Report publishes a source-of-hire table from applicant tracking records:

  • Referrals: 29,458 hires from 376,594 applications, a 7.82% hire rate.
  • Job boards: 45,928 hires from 7,060,680 applications, a 0.65% hire rate.
  • Career sites: a 1.39% hire rate.

Referrals made up 2.63% of applications and 12.26% of hires. That is a 12-fold gap in hire rate between referrals and job boards.

A version of this statistic often repeated online says referrals are about 7% of applicants and 40% of hires. Jobvite's table does not support it. The direction is right, but the 40% figure is more than three times Jobvite's measured hire share.

How do you export your LinkedIn connections?

The voiceover says your AI "will download" your connections. In the video, a person does it. The screen shows LinkedIn's "Download my data" settings page, and the prompt begins "Start with my Connections.csv and use Clay to expand the list."

LinkedIn's help page says a request for a specific data type arrives by email "within minutes," while the larger archive arrives "within 24 hours," and the download link stays live for 72 hours. The Connections file contains "the first and last name, public profile URL, email address, company, position, and connection date." The order is a manual export first, then the AI analysis.

The video's tool picker lists four ways to build the list: Clay ("Infrastructure to get any data, run agentic workflows, and launch GTM plays"), Apollo ("240M+ contacts and 30M+ accounts"), LinkedIn Sales Navigator, and your own connections. The prompt depends on the last one.

What are the four networking archetypes?

The creator describes "the competent salesy dudes, the influencer-ish fun people, like nerdy introverted people, like the steadfast team-loving people." The on-screen cards label them Driver, Influencer, Steady and Analyst, and each card is stamped D, I, S or C.

That is the DISC model. William Moulton Marston set out the four traits in his 1928 book Emotions of Normal People, using the words Dominance, Inducement, Submission and Compliance. Walter Clarke built the first self-assessment on the theory in 1956. Marston presented it as a theory of emotional expression, not an outreach tool. Modern commercial versions rename D and C as Driver and Analyst, as the video's cards do.

In the video, a language model assigns these types from someone's public LinkedIn writing, and in one case reports "medium confidence." The prompt does ask the model to record "confidence, exact evidence and source links" and to use "insufficient evidence" when needed, and the flowchart includes an "Unclear? Review" branch.

Is it against LinkedIn's rules to have AI send DMs?

Yes. LinkedIn's User Agreement, section 8.2, says members agree not to "Use bots or other unauthorized automated methods to access the Services, add or download contacts, send or redirect messages, create, comment on, like, share, or re-share posts, or otherwise drive inauthentic engagement." It also bans scraping tools, browser plugins and add-ons. LinkedIn's help page on prohibited software says members who use such tools "risk having their accounts restricted or shut down."

The workflow on screen stays on the allowed side. Its flowchart ends at "Draft + review," the model reports "This draft is still unsent," and every message screenshot shows an empty message box. The AI drafts and a person sends. The voiceover's closing line drops that distinction.

The four example DM threads in the video all carry the same "AUG 3 · 8:11pm" timestamp and stock-style portraits, which marks them as staged examples.

What the AI got right

The session runs on Claude Opus 5.5 at medium effort, which matches Anthropic's model listing (claude-opus-5-5, default effort medium). It uses Exa, which describes itself as "The world's data for AI agents," for web research. Partway through, the model corrects itself: "Correction first: four people are outside the 90-day window, not two," with the date math checking out.

The sample draft it produces references Jockey, which TwelveLabs describes as "the first video intelligence AI agent," and the person it targets is credited by name on the TwelveLabs engineering blog. The draft named a real product at a real employer. Another person on the list had "posted three of Union.ai's open roles himself," and Union.ai is a real AI runtime company.

Is LinkedIn mostly AI agents?

The video's description says LinkedIn is "basically the first social media platform mostly consisting ai agents." LinkedIn's Community Report for July 1 to December 31, 2025 lists "over 1.3 billion members" and says "99.7% of the fake accounts were stopped proactively, before a member report." It publishes no figure about the share of accounts or activity that are AI agents.

Bottom line

Referrals convert to hires at about 12 times the rate of job board applications in Jobvite's data, so networking for a referral is sound advice. Exporting your connections and having AI research and draft messages is a reasonable workflow. Send the messages yourself, because LinkedIn's User Agreement bans automated messaging and can restrict accounts that use it.

Sources

Published October 7, 2026. Based on a TikTok by @angusthenontechnical. Full video notes (markdown)