--- title: The AI That Fact-Checks Every Job Posting description: See which AI models we train ourselves, where our job data comes from, and why ghost-posting, pay and visa answers come from records, not guesses. type: product url: https://www.foundrole.com/how-it-works date: 2026-08-15T16:29:34Z og_description: Most job boards resell whatever they are given. We pull openings straight from employer hiring systems, run our own AI models over them, then check each one against public records before you see it. og_image: https://www.foundrole.com/img/pages/q5solw/how-it-works.png?v=1 breadcrumbs: - label: Home url: https://www.foundrole.com/ --- FoundRole checks every job posting before you open it: whether the listing looks real, how its pay compares to wages employers actually filed, and whether that employer sponsors visas. The openings come straight from employer hiring systems rather than resold feeds, and we run our own fine-tuned AI models over the whole index — while the verdicts themselves are computed from public records, so the answer is the same whether you browse the site or ask ChatGPT. ## Frequently Asked Questions ### Where do FoundRole job listings come from? Openings are imported directly from the hiring systems employers publish to, not bought as a resold aggregator feed. Over 1.8 million positions are indexed this way and refreshed hourly, so a posting that an employer closes stops showing up instead of lingering for weeks. ### How do you know if a job posting is real? Each posting is scored against eight independent checks — things like whether the pay and seniority match comparable openings for that role, and how long the listing has been recycled. Postings that fail several checks are graded and pushed down the list rather than deleted, because a weak signal is not proof and you should be able to judge for yourself. ### Where do your salary numbers come from? Pay comparisons are anchored to wages employers filed with the US Department of Labor, not to self-reported salary surveys. When a posting hides its range, we estimate one from those same records and label it as an estimate. ### How do you know which employers sponsor visas? Visa sponsorship is read from public US immigration filings, which name the employer, the role and the wage paid. That means the answer is a matter of record rather than a guess — and you get it before you reach the sponsorship question buried in an application form. ### How does the AI job match work? Upload a resume and our own AI models read it into a structured profile, then score each posting against what it actually asks for: which requirements you already meet, which skills are missing, and how much the gap matters. It matches meaning rather than counting keywords, so a skill you describe in your own words still counts. ### Does FoundRole use AI, or is it just a keyword search? AI is the engine, not a label. We fine-tune our own neural models on job-market data and run inference across the entire index continuously, so a search understands that two differently worded roles are the same job. What we deliberately do not do is let a chatbot decide the verdicts: real or ghost, pay and sponsorship are computed from public records, so they cannot be hallucinated and do not change between two runs. ### Can I use FoundRole inside ChatGPT or Claude? Yes. Connect FoundRole once and your assistant can search openings, read the same checks and manage your tracker from the chat. It is free and needs no API key. Interactive panels render inline in assistants that support them, and everything degrades to plain text where they do not. ### What does this cost? The fact-checks are free on every posting, along with job search, the tracker, the resume tools and the assistant connection. Pro is $20 a month and adds depth: the full read on why a job fits, screening a whole list at once, and sponsorship history year by year. # The AI That Fact-Checks Every Job Posting Job boards ask you to trust the listing. We would rather show our work: where the openings come from, which AI models we train and run ourselves, and why the verdicts you rely on are computed from records instead of guessed by a chatbot. ## Where the jobs come from Most job sites buy the same recycled feed and pass it on, which is why the same dead posting follows you across five of them: an aggregator buys a feed, resells it onward, and nobody rechecks it. We read the hiring systems employers publish to themselves — re-read every hour, checked before you see it, so a role the employer closed is gone here. 1. **Read from the source** — we connect to the systems employers post their own openings in, so a listing arrives the way the employer wrote it: title, pay, location and all. 2. **Refresh every hour** — when an employer closes a role it stops showing up here, instead of lingering for weeks the way recycled feeds do. 3. **Check before you see it** — each posting is matched to the employer behind it and checked against public records. Anything we cannot verify is labelled, not quietly dropped. ## What we check Four questions you would otherwise research yourself, one posting at a time — an hour of digging most people skip, then pay for later. All four are answered free, before you open the job. **Is this posting real?** Every posting is scored against eight independent checks — whether the pay and seniority line up with comparable openings for that role, how thin the description is, how long the listing has been recycled. Postings that fail several are graded and pushed down rather than deleted, because a weak signal is not proof. **Does it pay what the market pays?** Pay is compared to wages employers actually filed with the US Department of Labor — a legal record, not a self-reported survey. When a posting hides its range, we estimate one from the same records and say plainly that it is an estimate. **Does this employer sponsor visas?** Read from public US immigration filings, which name the employer, the role and the wage paid. You get the answer before you reach the sponsorship question buried at the end of an application form. **Is this job worth your time?** Upload a resume and our AI reads it into a real profile, then scores each posting against what it actually asks for — which requirements you already meet, which skills are missing, and how much that gap matters. Not a keyword count: the model matches meaning, so a skill you describe in your own words still counts. ## The AI underneath AI is the engine here, not a label on the box: we train our own models on job-market data and run inference across the whole index, continuously. - **Our own fine-tuned models** — the same role is written a hundred ways across employers. We fine-tune our own neural language models on job-market data, so a search understands that two differently worded roles are the same job, and that a third only looks similar. - **AI matching that reads meaning** — your experience is read into a structured profile and scored against what each posting actually asks for, in a shared semantic space. That is how a match knows the difference between a skill you have and a keyword you used once. - **Inference at index scale** — classifying, de-duplicating and scoring millions of postings runs around the clock on infrastructure we operate. **Where we deliberately don't use a chatbot.** Ask a chatbot whether a job is a ghost posting and it will always produce an answer — sounding just as certain when it is wrong. That is a hallucination you would act on, with a week of your search attached. So the verdicts you rely on are computed from records instead, and the same job gives the same answer twice. ## AI assistants FoundRole is AI-native in the literal sense: it is built as something an AI agent can operate directly, not a website with a chatbot bolted to the corner. Connect it once in ChatGPT, Claude or any assistant that speaks the same protocol, and the search and tracker answer from the chat: - Search openings with the same checks you see on the site - Read why a posting looks real, what it pays, whether the employer sponsors - Save jobs to your tracker and move them between stages - Set follow-up reminders and manage job alerts Where the assistant supports it, results come back as interactive panels. Where it does not, the same answers arrive as plain text. Free on our side — no API key, no plan required. ## Your data - **Your resume stays yours** — read to build your profile and score your matches. Not sent to employers, sold, or turned into a candidate listing. - **Your search stays private** — the notes, the salaries, the rejections are private to your account. - **You are not the product** — we make money from Pro subscriptions, not from selling your data to recruiters. That is why the free tier can afford to be honest about a bad posting. [Read the privacy policy](https://www.foundrole.com/privacy-policy) ## Where to check us We are a small team, so "trust us" is worth little on its own. These are places outside our control where FoundRole is on the record. - **[AWS Activate](https://aws.amazon.com/startups/showcase/startup-details/a0c5b2d2-2b02-4060-980a-61130a22284b)** (cloud provider) — we build on AWS and are part of its startup program, with a profile in the public showcase. - **[SaaSHub](https://www.saashub.com/found-role)** (software directory) — anyone can leave a review there, including a bad one. - **[Crunchbase](https://www.crunchbase.com/organization/found-role)** (company database) — the reference database investors and journalists check when they want to know who a company is. - **[F6S](https://www.f6s.com/company/foundrole)** (startup network) — our company profile and what we are building are on the record. ## Pricing The fact-checks are free on every posting, along with job search, the tracker, the resume tools and the assistant connection. Pro is $20/month and adds depth: the full read on why a job fits, screening a whole list at once, and sponsorship history year by year. --- ## Learn More - [Job Search](https://www.foundrole.com/jobs) — browse the index with fact-checks on every posting - [AI Job Search MCP](https://www.foundrole.com/ai-search-mcp) — connect ChatGPT, Claude or Cursor - [Visa Sponsors](https://www.foundrole.com/h1b-sponsors) — which employers actually sponsor - [Visa Salaries](https://www.foundrole.com/h1b-salaries) — what employers filed, by role and year - [Pricing](https://www.foundrole.com/pricing) — what is free and what Pro adds - [About Us](https://www.foundrole.com/about-us) — who builds this and why