A job search has two separate problems, and it’s worth naming them separately.
The first is finding the roles worth your attention: watching every board and alert, checking each new posting against what you actually want, and getting hundreds of them down to a handful. That’s a pipeline problem, and it’s what Fitcheck does.
The second starts once you have that handful. Is the company stable? Has the team churned three times this year? What’s the stack really? None of that is in the posting, and no scoring rule can infer it, because the posting itself is mostly boilerplate. Answering it means research, which is slow, and worth doing properly on a few roles rather than badly on forty.
That second job is what an AI is genuinely excellent at, so we put it one click away.
The Ask AI button
Open any job on JobWeb Board and there’s an Ask AI button sitting above the description, next to the copy and share controls. Click it and you get a short menu: ChatGPT, Claude, Grok, Perplexity, Meta AI.
Pick the one you use and it opens in a new tab with the posting link and a full research prompt already written. Nothing to copy, nothing to paste, and no account with us required, so it works whether you’re signed in or just passing through. It runs on your AI rather than ours, which also means your job search isn’t routed through a third party on the way.
What it actually asks for
The prompt isn’t “tell me about this job.” It asks for a structured due-diligence report and, importantly, tells the model not to make things up:
- Company health: funding history, financial state, runway, stage and risk profile
- Market: total addressable market with a cited source and year, competitors, differentiation
- Role details: reporting line, backfill and turnover signals, likely salary range with a confidence level
- Tech stack: inferred from postings, the engineering blog, public tooling profiles
- Employee sentiment: patterns across Glassdoor, Reddit, Blind
- Red flags: layoffs, leadership churn, funding gaps, long time-to-fill, litigation, negative press
It also instructs the model to cite references, separate opinion from fact, and decline to guess rather than fabricate. That last part matters: an unprompted “research this company” often returns confident nonsense, and the difference between a useful report and a plausible one is whether you asked for sources.
Where the pipeline ends and the research starts
It’s worth being precise about what Fitcheck does and doesn’t do, because the two halves are easy to confuse.
Fitcheck is a pipeline, and in one sense an unusually good search engine, because its data set is everyone else’s: every job alert you already receive, from any board that sends them, forwarded in and read for you, plus our own board, plus a shared pool of postings that other people’s alerts surfaced. Each one gets resolved back to the employer’s own listing and checked against the title, pay, location, and work mode you specified, coming back as Apply, Review, or Skip with the reasons attached. It runs continuously, so roles reach you whether or not you went looking. What it does not do is form an opinion about the company. It has no view on funding runway, leadership churn, or whether the stack is a decade old, because none of that is in a job posting to be checked against anything.
The AI report is the other half. It goes wide on exactly those questions, on one role at a time, using sources that have nothing to do with the posting.
Notice what we didn’t do there. Every product you use is currently having an AI bolted to its side and a higher price attached, whether or not the thing underneath got any better. We could have done exactly that: run a model over every posting, call it JobWeb Intelligence, add a tier to the pricing page, and let the marketing carry it.
Instead the button hands the work to an assistant you already have, on your own account, at whatever you’re already paying for it, which for most people is nothing at all. The research is better that way regardless, because it lands in a chat where you can push back on a claim, ask the obvious follow-up question, and click through the sources it cites rather than taking a generated summary on faith.
Neither substitutes for the other, and running them in the wrong order wastes your evening. Research forty postings by hand and you’ll burn a week to find that thirty of them never matched your criteria in the first place. Filter first, then go deep on what survives.
The button inside Fitcheck reflects that division: because the role already carries a score against your criteria, the prompt hands those criteria to the model too, and asks it to name what the posting fails to address rather than quietly assuming it’s satisfied. A posting that never mentions remote work isn’t remote-friendly, and an assistant that knows what you asked for will say so.
When to use it
Not on everything. Twenty AI reports is the same time sink as twenty job postings, just automated.
Use it on the roles you’re seriously considering, after the posting has cleared your basic bar. The research is most valuable at exactly the moment you’re about to spend an evening tailoring a resume, because that’s when discovering a hiring freeze or a leadership exodus saves you the most.
For the deeper version, including how to bring your own resume into the comparison and what to do with what you find, we wrote the full method in Stop applying blind: use AI to vet companies first.
Why this is a button and not a feature
We could have built a research feature, run a model over every posting, and charged for it. It would be worse. You already have an assistant that does this well, and the report is better when it runs on your account with your history and your follow-up questions. What an assistant can’t do is the continuous part: watching every source and evaluating everything that arrives, day after day, so there’s a short list waiting when you sit down. That’s the half worth building, and the half worth paying for.
So: Fitcheck to decide which three roles are worth an evening, the AI report to find out what those three companies are actually like. Different tools, different questions, and the order matters.