Eric Siu's AI Outbound Agent: Does a $15 Scan of 120,000 Followers Add Up?
By Matt Cullerton, founder of Mavric and Apso
In a Marketing School clip posted by @levelingupofficial, Eric Siu describes an agent he calls "GrokBot" that runs his outbound: it watches who engages with a competitor, sorts his own followers, and scans another account's followers for recruiting. The headline number, $15 to scan 120,000 followers, conflicts with X's published API pricing, which puts reading that many of another user's followers at $1,200. The clip shows no invoice, data source or agent output, so its numbers are the agent's own estimates relayed second-hand.
What does the AI outbound agent do?
The clip's cards are titled "Autonomous ABM: AI does your outbound." The speaker describes three jobs:
- Competitor watch. "whoever engages with Profound's posts on LinkedIn or their ads, if they're ICP fit, let me know." A card shows "4" with "ICP-fit people engaged. Here's the angle to reach out." and "DRAFTED, READY TO SEND".
- Sorting his own followers. He asks for "M&A targets", "recruit targets" and "ICP fit", and the agent reports "you have 581 people that are ICP fit."
- Scanning someone else's followers. He asks it to "do it on Lily Ray's X profile" and cover "all her 120,000 followers."
The card footers all read "SINGLE BRAIN", which the clip does not explain. "GrokBot" is not documented anywhere I could find, so whether it uses xAI's Grok model or only borrows the name is unknown.
Does it really cost $15 to scan 120,000 followers?
The agent's answer in the clip is "if you remove less than 100 followers, that will cut it down from $25 to $15." The card reads "$15 to scan 120,000 followers." with "$25 BEFORE FILTERING OUT TINY ACCOUNTS".
X's developer pricing page lists reading another user's followers at $0.010 per resource returned. It lists a tenth of that, $0.001 per resource, only for "owned reads", meaning the authenticated user's own followers on an app that user owns. Lily Ray's followers are not Eric Siu's followers.
At $0.010 each, reading 120,000 followers through the official pay-per-use API costs $1,200, before any language model scores them. A $25 figure would match the owned-read rate only for about 25,000 accounts at $0.001.
That does not prove the clip wrong. A third-party scraper, a cached dataset, another data vendor or a flat-fee plan could cost less, and the agent may have quoted its own estimate rather than a billed charge. The X page also says prices are subject to change and shows no last-updated date. If you plan to budget for this, price your data source first.
The internal math is consistent. Going from $25 to $15 is a 40% drop, which implies cost scales with the number of accounts scanned after small accounts are filtered out. The clip does not say what share of followers the filter removed.
How reliable is the "581 ICP-fit followers" number?
Three gaps limit it:
- ICP, the ideal customer profile, is never defined in the audio. The card's "CMOS AND AGENCIES THE RIGHT SIZE" comes from the description, not the speech.
- Nobody says how many of the 581 a person reviewed or how many were wrong.
- The description says the work "would have taken weeks by hand", but the audio does not make that claim.
The description's own advice is to "keep a human approving messages until the hit rate is proven." The clip never states a hit rate.
Can an agent see who engages with a competitor's LinkedIn ads?
The clip shows no evidence of how. The audio says "these four people engaged X amount of times", leaving the count as a placeholder. Reactions and comments on public posts are visible, but I did not verify whether LinkedIn exposes who engages with a competitor's ads. Profound, the competitor named, is a real company that sells AI-search visibility tracking.
What happened when the agent started outreach?
After a single "Okay, scan it.", the speaker says "it started reaching out to people for recruits. And they're like, I'm interested, but how did you get my information?"
The speech does not say whether a person approved each message. That sits awkwardly next to the description's advice to keep human approval in place. The clip shows no message, reply or send log.
The recipients were contacted because they followed someone else's public account. The speaker closes with "It's going to be a world of more and more abundance, just more." The description draws a different lesson: "If an agent sends outreach for you, it has to reach the right person with a useful reason, or you've just automated spam." The clip's own evidence supports the second reading, since at least one recipient asked where their details came from.
Where is the full episode?
The description credits "Marketing School with Eric Siu and Neil Patel." I could not find the source episode. Episode #2923 from February 26, 2025 covers Grok, but its page mentions none of ABM, Profound, Lily Ray, 581 or a $15 scan, so it is not this clip.
Bottom line
The workflow is a real pattern: watch competitor engagement, classify followers, and draft outreach. The $15-for-120,000 figure does not match X's official API rate for another user's followers, which works out to $1,200. The 581-person count has no definition or accuracy check, and the "how did you get my information?" reply shows why human approval on outreach matters.
Sources
- X API pricing, X developer documentation: $0.010 per resource for another user's followers, $0.001 for owned reads, and the "subject to change" notice.
- Using Grok for infinite interns and who will really win in an AI world, Marketing School #2923: a Grok episode checked and ruled out as this clip's source.
- Profound Hits $1.8B on $180M Series D From Sequoia, Kleiner, AI Weekly: secondary coverage of Profound as an AI-search marketing company, seen only as a search result.
- TikTok post by @levelingupofficial: the clip reviewed here.
Published October 29, 2026. Based on a TikTok by @levelingupofficial. Full video notes (markdown)