Your CV isn't there to get you hired
It's there to stop you being binned in seven seconds

Mechanism: screening is an elimination task, not a selection task. Facing two hundred applications, the only workable strategy is to shrink the pile fast, so the reader is hunting for a reason to discard, not for highlights. Two things follow. If the top three lines give no reason to keep you, nothing below them was ever read. And effort of reading is itself a reason to discard — the brain mistakes ease of processing for quality, so jargon and cramped layout get filed as "this person hasn't thought it through either".

• Hard evidence · fluency gets misread as liking: Winkielman & Cacioppo (2001, JPSP), Mind at ease puts a smile on the face: the more fluently something is processed, the stronger the EMG activity over the zygomaticus (smiling) muscle — and that faint positive affect gets misattributed to whatever is being judged. Physiological measures across several experiments; solid.
• Peer-reviewed · complicating your language costs you marks: Oppenheimer (2006, Applied Cognitive Psychology): replacing words in the same passage with needlessly complex ones significantly lowered readers' ratings of the author's intelligence. "Leveraging holistic user-mindshare enablement" doesn't read as expert. It reads as suspicious.
• Hard evidence · the machine itself is biased: Bertrand & Mullainathan (2004, American Economic Review), a field experiment sending roughly 5,000 CVs: white-sounding names got about 50% more callbacks than African-American-sounding ones, and the return to a higher-quality CV was larger for the white-sounding names. This isn't here to demoralise you: polishing the page improves your conditional odds, it does not repair the system — and it happens to explain why the channel where someone vouches for you pays off most for the people the system underrates.

Boundary: landing-page logic applies where a human actually reads. Keyword screening at a large employer is a different machine (words first, people second), so lift the exact terms from the job description rather than inventing synonyms. Small companies, referrals and mail sent straight to the hiring manager are where this pays best. And don't let "results" slide into puffery — every number has to survive one question: how was that calculated?

Two mechanisms. Social: talking about yourself is genuinely rewarding at the neural level, so letting them talk makes the conversation feel better to them — and they will later file that feeling under "we clicked". Predictive: interviewers' confidence in their ability to read people vastly exceeds what unstructured conversation actually delivers, which means you cannot count on being discovered through chat. You have to put something evaluable into the room yourself.

• Hard evidence · self-disclosure is intrinsically rewarding: Tamir & Mitchell (2012, PNAS): talking about oneself raised activity in the nucleus accumbens and ventral tegmental area — the mesolimbic dopamine reward pathway — along with medial prefrontal cortex; in companion behavioural studies people gave up real money for the chance to disclose about themselves. fMRI plus behaviour; solid.
• Peer-reviewed · the illusion of the unstructured interview: Dana, Dawes & Peterson (2013, Judgment and Decision Making): even when interviewees were instructed to answer at random, interviewers reported feeling they had learned about the person — and predictions made with interview information were worse than predictions from the objective record alone.
• Meta-analytic, with a correction history · structure beats chat: Sackett et al. (2022, Journal of Applied Psychology) re-estimated the validity of selection methods: structured interviews around .42, unstructured around .19, work samples around .33. Note this is a revision of the classic Schmidt & Hunter (1998) figures and the absolute values move with the correction method — read the ranking, not the decimals.

Boundary: turning the questions round is diagnosis, not a coup. Two at the start and one at the end is plenty; keep it up all hour and they'll feel they're the ones being interviewed. Some stages (mass graduate rounds, an HR screen) leave no room to diagnose at all — clear the keywords there and save the diagnosis for the hiring manager's round. And demo only the step you can genuinely do; improvising through one you can't is worse than not demoing.

Three mechanisms. One, your BATNA — the best alternative you actually have if this falls through — sets your floor; you never state it, but it decides where you can stop. With no alternative, the honest move is to admit you're in the weaker position and put your chips on non-cash terms. Two, the first number spoken becomes an anchor, and adjustment away from it is usually insufficient. Three, this is a repeated game: you'll be working alongside this person next week, so winning too visibly gets collected later — which is why the framing is "how do we close this" rather than "here's what I want".

• Hard evidence · the first mover does better: Galinsky & Mussweiler (2001, JPSP): across several negotiation experiments, the side making the first offer ended up with significantly better outcomes, because that number anchored the range everyone searched afterwards. (An anchor works by changing which information the other side retrieves, not by convincing them — the mechanism is unpacked in the piece on anchoring and concessions.)
• Hard evidence · the social cost of negotiating isn't evenly distributed: Bowles, Babcock & Lai (2007, OBHDP): evaluators penalised women who initiated salary negotiations more heavily, being less willing to work with them, with male evaluators penalising most. This isn't an argument for not negotiating; it's a warning that the cost lands unfairly. Framing the ask around the value of the role rather than personal want reduces the backlash — it does not remove it.
• Hard evidence · change the rule and the gap changes: Leibbrandt & List (2015, Management Science), a large field experiment in real hiring: when the advert explicitly said wages were negotiable, the gender gap in whether people negotiated largely disappeared and in places reversed. The difference comes far more from ambiguity about whether asking is allowed than from disposition.

Boundary: don't negotiate before an offer exists — you'd only be handing over a free reason to cut you. Don't invent a competing offer; it's a small world and a repeated game, and being caught costs for years. And more practically: you can only push if you're genuinely prepared to walk. If you know you'd accept regardless, put the effort into the non-cash terms — level, reporting line, starting budget, a review date — which often shape where you are in two years more than a few thousand does.

Mechanism: a referral is a costly signal — the referrer is staking their own standing inside the company, and a bad call gets remembered. Precisely because it costs a careless recommender more than an honest one, employers take it seriously. It also spares the employer the most expensive step of all: working out which of two hundred applications is worth an hour. Doing a small piece of work before you ask runs on the same logic — someone willing to pay a cost up front is no longer filed alongside everyone declaring how passionate they are.

• Hard evidence · referred workers really are better, not just better-connected: Burks, Cowgill, Hoffman & Housman (2015, Quarterly Journal of Economics), using data across several firms spanning call centres, trucking and high-skill roles: referred candidates are more likely to be hired, less likely to quit, and more profitable to the employer overall; the authors attribute this mainly to better matching and information rather than favouritism. Large samples, peer-reviewed.
• Peer-reviewed · referrals are a deliberate investment: Fernandez, Castilla & Moore (2000, American Journal of Sociology) tracked a call centre's entire hiring process and found the returns from referrals were large enough that the referral bonus was a sound investment for the firm.
• The honest other side: referrals amplify existing network inequality — networks are homophilous, so people outside the circle get shut out systematically, which points at the same problem as the CV field experiment above (identical documents, different names, half the callbacks). This is exactly why "manufacture weak ties and do the work first" pays off most for people starting with no network: you have no ready-made channel, so you build one.

Boundary: doing work first is not free labour. A one-off small thing is a signal; a continuing stream of unpaid output is being taken advantage of — if they take the work and only want more of it, stop. And don't perform being in demand: inventing competing interest or pretending to be busy costs you the only genuinely valuable thing you have the moment it's spotted. As for polishing the sentence other people use to describe you, the piece on selling yourself as the meta-skill takes that apart.