Texas Priced LinkedIn's Ghost Jobs at $17.8B. Sourcing Changes Now.

On July 14, 2026, Texas Attorney General Ken Paxton opened a formal investigation into LinkedIn for selling Premium subscriptions against a listings inventory that independent studies put at 18 to 27 percent ghost jobs. The Civil Investigative Demand names LinkedIn's roughly $17.8 billion in FY2025 revenue as the material harm. If you run inbound hiring in 2026, a state regulator has just told you your top-of-funnel is a paid product with a documented counterfeit rate.

This is not a refund story. It is the first time the top of the funnel has been priced as a defective good, and it lands the same quarter Qualigence and ERE published pieces arguing LinkedIn's Hiring Assistant deepens dependence rather than fixing sourcing. The takeaway for founders, recruiters, and engineering leaders is simple: stop treating LinkedIn as a channel and start treating it as a signal source.

What Texas actually filed on July 14, 2026

Paxton issued a Civil Investigative Demand, not a lawsuit, compelling LinkedIn to hand over documents, data, and internal communications about how Premium is marketed against listing quality. The theory of the case is that Premium buyers were never told "a significant percentage of job postings may be inactive, unfilled, or otherwise not representative of genuine hiring opportunities."

Three things make this different from prior LinkedIn complaints:

• It targets Premium as a paid-subscription product, not a bulletin board. That framing makes Recruiter licenses (the enterprise SKU inside the same $17.8B revenue line) the next obvious extension.
• It cites independent studies from ResumeUp.AI, Greenhouse, LiveCareer, and Clarify Capital, so LinkedIn cannot dismiss the ghost-jobs number as an outlier.
• LinkedIn is a Microsoft subsidiary reported at a $26.2B valuation. The parent is deep-pocketed enough that a settlement would land as a template, not a footnote.

No lawsuit has been filed. The CID is the discovery phase. California's March 2025 disclosure law and Ontario's Working for Workers Act (in force January 2026) already require employers to state whether a posting reflects an actual vacancy, giving Paxton a policy backstop if the investigation moves to enforcement.

The 27% ghost-jobs number, and why sector matters

LinkedIn's US listings run 27.4% ghost jobs by the ResumeUp.AI methodology, and a March 2025 LiveCareer survey of 918 HR pros found 93% admit to posting them - 45% regularly, 48% occasionally. Ghost jobs are listings that do not reflect an actual, currently fillable vacancy, defined by ResumeUp as postings older than 30 days relative to typical hiring timelines.

The headline number hides a sector spread that matters if you are hiring:

City-level variance is real too. Los Angeles hits 30.5%, Philadelphia 30.1%, Indianapolis 27.8%, San Francisco 26.0%, and New York carries the highest raw volume at 23,000 ghost listings. Seattle bottoms out at 16.6%. If you are sourcing software engineers in New York or LA, roughly one in three listings you scrape is noise.

441:1

Why "AI-Assisted Search" does not fix this

LinkedIn's AI-Assisted Search translates natural language into the same underlying Boolean filters, so the AI improves the front end without changing the machine underneath. That is Qualigence's line, and it is the sharpest critique in market. Layering a chatbot on the same filter graph does not add candidates; it just makes the same graph faster to query.

LinkedIn's own Hiring Assistant pilot metrics - 4+ hours saved per role, 62% fewer profiles reviewed, 69% higher InMail acceptance - are company-supplied and not independently verified, per ERE. Even taken at face value, they measure efficiency inside a funnel that a state AG just called defective. Cutting profiles reviewed by 62% inside a 27%-ghost input is not obviously a win.

There is also a revenue tension. Hiring Assistant hit a ~$450M annualized run-rate by April 2026 (The Information), or roughly 2.5% of LinkedIn's $17.8B revenue. If listing quality is deteriorating and LinkedIn is monetizing an AI layer designed to navigate its own noise, the platform is charging both sides more to solve a problem it created. Qualigence calls this "deepening dependence." Paxton calls it a potential consumer-protection issue.

A 30-day-old repost is a ghost job for applicants and a lead for outbound sourcers.

The real math: 441 engineers per sourcer

The bottleneck is not ghost jobs. It is that there are roughly 441 US software engineers for every US sourcer, and adding AI to LinkedIn's search box does not change ratio math. In Refolk's index, the US carries 554,933 software engineers against 1,258 people holding Technical Sourcer, Talent Sourcer, or Sourcing Recruiter titles. Top employers of that scarce sourcer pool include Rippling, Verkada, Anthropic, EvolutionIQ, Fetch Rewards, Gartner, and Zipline, concentrated in the SF Bay Area.

Two implications:

1. Inbound cannot scale even if listings were clean. At 441:1, a sourcer cannot manually qualify enough applicants per req to make LinkedIn's funnel work, ghost jobs or not. Every hour a sourcer spends chasing a stale listing is an hour not spent on outbound.
2. The companies already buying sourcers heavily - Anthropic, Rippling, Verkada - are the same ones investing outside LinkedIn's funnel. They read the same JOLTS data you do.

For a state-specific example, Texas alone has 930 SWEs in Refolk's index concentrated in Austin (10 top-employed) and Dallas (2), with Meta, UT Austin, Visa, Amazon, and LiveRamp as top employers. That is a small, sourceable, named pool. It is not a "search LinkedIn for Texas engineers" problem. It is a lookup problem.

Outbound sourcing 2026: read job boards as intent data

Treat every LinkedIn listing as a demand signal, not a distribution channel. A 30-day-old repost is a ghost job if you are an applicant. If you are an outbound sourcer, it tells you which team is struggling, which skills they are missing, and who inside the company currently holds that title. That is intent data, and it is exactly the wedge Refolk exploits: describe the person in plain English and get a ranked shortlist across GitHub, LinkedIn, and the open web.

The workflow that replaces the broken inbound funnel:

1. Scrape the listing for signals, not candidates. Job title, tech stack keywords (Rust, CUDA, Terraform), team size hints, and whether the req is a backfill or growth. Ignore the "apply" button.
2. Identify the incumbent. Who currently holds this title at the posting company? Who left in the last 12 months? These are the two highest-signal cohorts for a competitor's outbound.
3. Build the lookalike pool from the incumbent's peer graph. Not "people with title X at company Y." People who worked with the incumbent, contributed to the same repos, published in the same conferences.
4. Message from the specific, not the generic. Reference the actual project, PR, or paper. InMail acceptance rates rise when the sender proves they read something before hitting send.
5. Feed rejects back as signal. "Not interested, but talk to my old teammate at Stripe" is worth more than 40 cold InMails.

This is the shape of outbound sourcing 2026: listings become a leading indicator, and the actual sourcing happens in the peer graph. Gem, hireEZ, and Ashby's ATS data all point in this direction. What Refolk adds is the plain-English query layer that collapses the "translate hiring manager brief into Boolean" step Qualigence identified as LinkedIn's core weakness.

LinkedIn Recruiter alternatives, ranked by what they replace

LinkedIn Recruiter is three products in a trench coat: a search index, an outreach tool, and a CRM. You do not have to replace all three at once. The Texas investigation gives you cover to unbundle.

The point is not that LinkedIn is worthless. The point is that at 27% ghost, at $17.8B in revenue, and at 441:1 SWE-to-sourcer ratios, LinkedIn Recruiter should be a verification layer at the end of an outbound workflow, not the workflow itself.

What to do this quarter

Move budget from Recruiter seats to outbound tooling and one senior sourcer, in that order. Paxton's CID is a leading indicator. Even if LinkedIn settles quietly, the disclosure precedent from California and Ontario means listing quality will be a public metric within 18 months, and any team still running inbound-first will be paying more for less.

A concrete 90-day plan:

• Week 1 to 2: Audit your last 20 hires. What fraction came from LinkedIn Easy Apply vs. outbound? If inbound is above 40%, you are exposed.
• Week 3 to 4: Cut one Recruiter seat. Reallocate to a sourcing tool that queries across sources. Refolk is the plain-English option; hireEZ is the traditional one.
• Week 5 to 8: Rebuild your top three reqs as outbound campaigns. Use the incumbent-and-alumni method above. Track reply rate, not send volume.
• Week 9 to 12: Measure cost per qualified conversation, not cost per hire. If it drops, kill another Recruiter seat.

If you want a concrete starting point, ask Refolk for the peer graph of a role you are currently trying to fill and compare the shortlist to what LinkedIn Recruiter surfaces for the same brief. The delta is the ghost-job tax you have been paying.

FAQ

Is the Texas investigation likely to result in a lawsuit against LinkedIn?

The CID is a discovery instrument, not a complaint, and no lawsuit has been filed. The precedent that matters most is not the fine but the disclosure requirement. California's March 2025 law and Ontario's Working for Workers Act (January 2026) already require employers to state whether a posting reflects an actual vacancy, and a Texas settlement would likely extend that framing to the platform itself. Recruiter licenses are the next obvious target because the same $17.8B revenue argument applies.

How do I tell a ghost job from a real one on LinkedIn?

Three heuristics from the ResumeUp.AI and Greenhouse studies: posting age over 30 days on a typical white-collar role, listings that reappear every 30 to 45 days with identical copy, and postings from companies where the same req has been open across multiple recruiters' pipelines. Tech runs about 48% ghost by the JOLTS-derived analysis, so in software specifically, assume a listing is stale until proven fresh. The safer move is to skip the freshness question and use the listing as an intent signal for outbound instead.

What does outbound sourcing in 2026 actually mean in practice?

It means treating job boards as demand signals rather than distribution channels, and sourcing candidates from the peer graph of incumbents and recent alumni rather than from applications. The mechanics: identify who currently holds the role at the posting company, find their coworkers and collaborators through GitHub, publications, and shared work history, and reach out with specific references to their work. Tools like Refolk collapse the "translate brief into search" step into a plain-English query, which is the exact friction Qualigence identified as LinkedIn's structural weakness.

Should I cancel my LinkedIn Recruiter seats?

Not all of them, and not immediately. Recruiter is still useful as a verification layer for confirming current employment and reaching people who explicitly open-to-work. But if inbound is above 40% of your hires or if you are paying for InMail credits that burn on ghost-req teams, cut at least one seat this quarter and redirect the budget to outbound tooling and a senior sourcer. The 441:1 SWE-to-sourcer ratio in Refolk's index tells you where the leverage actually is.