Eightfold's Secret 0-5 Score: The 45-Minute LinkedIn Fix

If you have applied to Microsoft, PayPal, Starbucks, or Morgan Stanley in the last year, an Eightfold model may have already scored you 0 to 5 before a recruiter opened your file. The January 2026 class action against the vendor claims that score was built from scraped LinkedIn and GitHub data you never handed over. The fix that actually moves the number is not your resume. It is the shadow profile the model is still indexing while you read this.

What the Eightfold AI lawsuit actually alleges

The Eightfold AI lawsuit, filed January 20, 2026 by Erin Kistler and Sruti Bhaumik, alleges the platform secretly generates a 0-to-5 "likelihood of success" score on applicants using scraped public data, without the disclosure and dispute rights the Fair Credit Reporting Act requires. It is not a bias case. It is a privacy case, and that distinction is what makes it dangerous for every AI screening vendor, not just Eightfold.

The named details from the complaint and reporting:

• Filed in California, removed to the Northern District of California as No. 3:26-cv-1768, still in the pleading stage as of late April 2026.
• Brought by former EEOC chair Jenny R. Yang, the nonprofit Towards Justice, and Outten & Golden LLP.
• Alleges Eightfold pulls from LinkedIn, GitHub, job boards, location data, and tracking cookies to score candidates applying to Microsoft, PayPal, Starbucks, Morgan Stanley, Chevron, and Bayer.
• Claims the AI uses "more than 1.5 billion global data points" and scraped personal data on over one billion workers to feed the ranking.
• Statutory damages sought: $100 to $1,000 per violation under federal FCRA, up to $10,000 per violation under California's ICRAA, plus punitive damages.
• Eightfold's public response: it operates only on candidate-provided and customer-provided data and does not actively scrape social media.

The plaintiffs are not arguing the score was wrong about them. They are arguing the score existed at all, in secret, and that they were denied the FCRA-mandated right to see and dispute it. That legal theory travels. Every sourcing tool that quietly builds a candidate profile from public web data (hireEZ, Fetcher, SeekOut, LinkedIn Recruiter's own inferences) sits on the same fault line.

How the Match Score is actually built

The Match Score is a three-step pipeline that turns your public footprint into a single 0-to-5 number, and roughly none of it reads your uploaded PDF the way you think it does. The complaint lays out the mechanism in unusual detail.

1. Semantic similarity. An LLM compares the job description to your candidate profile (the aggregated shadow record, not just the resume you attached).
2. Feature extraction. The AI pulls skill overlap, title progression, seniority fit, industry similarity, and comparison against "ideal candidates" and the hiring manager's own profile.
3. Calibrated ranking. Those features get blended into the proprietary 0-5 score, and low-ranked candidates are discarded before a human sees the file.

Read step 2 twice. Skill overlap, title progression, seniority fit, industry similarity: every one of those signals is inferred from your LinkedIn history, not from the resume you tailored last night. If your LinkedIn still lists a 2019 title as "current," the model treats you as under-leveled today. If your headline says "Open to opportunities" instead of the exact role family you want, semantic similarity drops. The resume is a downstream artifact. The shadow profile is the input.

The complaint also alleges Eightfold's reports describe personality traits like "team player" and "introvert" and predict future job titles from data the candidate never provided. That is inference on top of inference, and none of it lives inside the PDF you control.

Why your resume rewrite is the wrong first move

Rewriting the resume matters, but it lands second. The model has already indexed years of your LinkedIn changes and, if you are an engineer, your GitHub commit graph. A new PDF cannot retroactively edit that index. What it can do is agree with a cleaned-up shadow profile so the two signals reinforce each other during semantic scoring. That is the exact work Refolk does after the cleanup: paste the posting, get your resume rewritten to match, and see the fit score before you send.

The 45-minute LinkedIn optimization for ATS shadow profiles

The highest-leverage cleanup is a 45-minute pass on LinkedIn that changes what any Eightfold-style scraper feeds back into the score. You are not gaming the model. You are removing the stale signals that make the model wrong about you.

Work in this order. The order matters because later steps depend on earlier ones being consistent.

Minutes 0 to 10: Headline, About, and location

• Rewrite your headline as exact target-role keywords plus one differentiator, not "Open to opportunities." Semantic similarity is a keyword-heavy step.
• Rewrite the first two lines of your About to name the same role family, seniority, and industry your target postings use. The model compares "industry similarity" and "seniority fit."
• Fix the location to the city you will actually work in or from. A stale "San Francisco Bay Area" on a remote worker in Austin creates a phantom market-fit gap.

Minutes 10 to 25: Title progression

• Confirm every job has clean start and end dates. Gaps and overlaps make "title progression" look erratic.
• Standardize titles to the industry-recognizable version. "Growth Alchemist" reads as a novelty; "Senior Growth Marketing Manager" reads as a level the model can rank.
• If you were promoted internally, split the entries. One "Software Engineer II to Senior Software Engineer" row hides the progression the model is looking for.

Minutes 25 to 35: Skills, endorsements, and Featured

• Reorder your Skills so the top three match the top three skills in the job posting you care most about. Skill overlap is a discrete extracted feature.
• Delete skills you no longer want to be ranked against. Every stale skill is a vector the model can pull toward the wrong role family.
• Pin two or three Featured items that show recent, on-topic work. This is public and scrapeable.

Minutes 35 to 45: Privacy, activity, and adjacent surfaces

• Under Settings, turn off "Profile discovery using email address" and "Profile discovery using phone number" if you do not want dormant records tying back to you.
• Review your public activity feed. Old comments and reactions are indexed too.
• For engineers: update your GitHub bio, pin three current repos, and hide personal forks that skew your language mix. The complaint names GitHub explicitly.
• If your employer uses Eightfold internally, remember the Digital Twin product launched in May 2025 pulls from email, messaging, CRM, and project tools. Your public Jira boards and Slack Connect bios count.

That is the whole pass. Set a timer.

What FCRA hiring AI rules would change if plaintiffs win

If the Kistler plaintiffs win on the theory that Eightfold's report is a "consumer report," every AI screening vendor selling into US employers has to give applicants notice, a copy of the report, and a dispute channel before an adverse action. That is a structural shift, not a tweak.

The mechanics of why this matters to you as an applicant:

• Notice. You would receive written disclosure that an AI-generated report was used, which vendor generated it, and what data classes fed it.
• Access. You could request the report itself, including inferred traits like "team player" or predicted future titles.
• Dispute. You could challenge inaccuracies (wrong title, wrong tenure, wrong skill inference) and force a re-score.
• Damages. Failure to comply carries $100 to $1,000 per federal violation and up to $10,000 per California violation, which is why a class action is economically viable even without proven bias.

Until the pleading is resolved, none of that is guaranteed. The rational move for a job seeker is to assume the score exists, assume it pulls from public data, and clean the inputs.

Eightfold is smaller inside recruiting than the headlines suggest

Eightfold has a large lawsuit profile and a much smaller footprint inside recruiting teams than the coverage implies, which changes how you should think about cleanup advice. In Refolk's index of professional profiles, only 15 US recruiters and TA specialists publicly list "Eightfold" as a skill, spread across employers like Waymo, HelloFresh, Lam Research, Freeport-McMoRan, and TriNet. For comparison, 183 US recruiters list "Workday Recruiting," roughly a 12x gap.

The takeaway is not that Eightfold is small and safe to ignore. It is that Eightfold is bought at the CHRO or CIO level and often invisible to the line recruiters you talk to. That is exactly why candidates never hear the vendor's name in adverse-action notices, and why the cleanup above generalizes to every peer tool doing the same thing under a different logo. A LinkedIn LinkedIn Recruiter inference, a hireEZ sourced list, a SeekOut candidate score: same inputs, same fix.

FAQ

Does cleaning LinkedIn actually change an Eightfold score?

If Eightfold's public denial is accurate and it only uses candidate-provided and customer-provided data, then a locked-down LinkedIn changes nothing. If the plaintiffs' allegations are accurate and the platform scrapes LinkedIn, GitHub, and job boards, then LinkedIn is the single largest input you control. Until the case is resolved, the asymmetric bet is to clean it. The downside is 45 minutes; the upside is a more accurate shadow profile at every vendor that scrapes public data.

Do I need to worry about this if I am not applying to Microsoft or PayPal?

Yes, for two reasons. The complaint names Microsoft, PayPal, Starbucks, Morgan Stanley, Chevron, and Bayer, but a recent LinkedIn study found 93% of recruiters plan to increase AI use in 2026. Sourcing tools that build inferred candidate profiles from public data are now the default across mid-market and enterprise hiring, not the exception. The cleanup pass targets the input, not the vendor.

What is the Eightfold Digital Twin and does it affect me if I am employed?

Digital Twin is an Eightfold product launched in May 2025 that captures employee data across email, messaging, CRM, and project management tools to score internal mobility candidates. If your employer runs it, the same shadow-profile logic applies to your promotion and internal-transfer applications, and the surfaces to clean expand to your Slack bio, your GitHub commit messages under your work email, and public project boards.

Should I still tailor my resume to each posting?

Yes, and more carefully than before. A shadow score is going to run whether you tailor or not, so you want the one document you fully control to score high on the features you can influence: exact skill overlap with the posting, standardized titles, clear seniority markers. Refolk rewrites your resume against the posting and returns a fit score so you can see, before hitting Apply, whether your resume and your cleaned-up LinkedIn tell the same story.