An AI career coach,
available right now.

Career is not a restricted category - no policy stops a general assistant answering. The problem is different and in some ways more frustrating: the answers are permitted, plentiful, and useless.

Ask about negotiating a salary and you get advice that is entirely true and that you already knew. Research the market. Know your worth. Do not name a number first. Nobody disagrees with any of it, and nobody knows what to type into the document at nine on a Tuesday evening.

That happens because a general model is answering the general question. Without your city, your level, your actual accomplishments and the specific posting in front of you, there is no version of the answer that is not generic - and it will not ask for those, because asking is a behaviour that has to be configured rather than a fact to be recalled.

The other half is that career decisions are regional and current. Compensation varies enormously by market and moves faster than training data, so the numbers in a general answer are directional at best.

Drafting, mostly. Getting from a blank page to a first version of a cover letter, a follow-up email or a resignation is genuinely faster with one, and the blank page is where most of these stall.

Reformatting works well - turning a rambling description of what you did into structured bullet points, or tightening something that runs long.

Explaining conventions works. What a typical interview loop looks like, what an ATS does, what vesting means, how notice periods generally work.

Where it falls down is anything requiring your specifics to be applied. Rewriting your resume against a particular posting, deciding whether a number is right for your market and level, or working out whether a counter-offer is worth taking. Those need the inputs to be gathered first, which is a different interaction.

And anything with a lawyer in it - severance, a non-compete, a release of claims - is a genuinely different category regardless of the tool.

If there is a legal document - severance, non-compete, a release of claims, anything alleging discrimination - get an employment attorney. An hour against a document that permanently gives up your right to bring a claim is not a close call, and many will review a severance agreement on a flat fee.

If you need the number, use the data sources rather than any model: Levels.fyi, Glassdoor, and the BLS. Compensation is regional and current, and no assistant's recall is a substitute for looking it up.

If you need to prepare - the resume against this posting, the answers to the six questions that actually decide the interview, the sentence you will say when they ask for a number - that is where a specialist copilot fits, because it starts by asking for the posting, the city and the level rather than answering without them.

The reliable pattern is that generic advice is free everywhere and specific preparation is what changes an outcome. The gap between the two is entirely in whether the inputs were gathered.

The folklore says a robot rejects you. The reality is duller and more actionable.

An applicant tracking system parses your document into structured fields - titles, dates, skills - and stores it. Recruiters then search and filter that data. Nothing automatically rejects most applicants; what happens is that a badly parsed resume never surfaces in the search, which produces the same outcome by a different route.

That reframes the practical advice entirely. Standard section headings parse reliably. Multi-column layouts and text inside tables or images frequently do not. A PDF exported from a design tool can lose its text layer completely, at which point the document is a picture as far as the system is concerned.

Then a human spends somewhere between six and thirty seconds on it, scanning for a match to the role. The top third of the first page does almost all the work, which means a summary that could describe anyone is wasting the most valuable space on the page.

The vocabulary point is real and widely overstated. Using the words the posting uses matters because those are the words being searched. Stuffing keywords into a skills block does not help and reads badly to the person who sees it next.

Most of the anxiety about salary conversations comes from not knowing the structure, so it is worth laying out.

The first move is usually a request for expectations. Several states now prohibit asking for salary history, which is worth knowing because it changes what you are obliged to answer. A range framed around the role and the market, rather than around what you currently earn, is a complete answer.

The second is the offer, which is rarely the top of what is available. Companies expect negotiation and budget for it, and the frequently cited finding is that most people never ask at all - while most who do get something.

The third is the part people forget entirely. Base salary is one component. Signing bonus, equity, vesting schedule, review timing, title, remote arrangements and start date are all negotiable, and several are easier for a company to move on than base. If base is genuinely capped, the conversation is not over.

Two things reliably go wrong. Naming a number and softening it in the same breath - adding "but I'm flexible" before anyone has responded - concedes the negotiation before it starts. And treating the recruiter as an opponent, when their actual incentive is usually to close the hire.

Practise the sentence until it is boring. Then say it and stop talking, which is harder than it sounds and is most of the technique.