There are many ways to write faster. Here we offer eight tricks; use none, one, or all of them, depending on your needs (see the decision tree below). Each trick is a two-minute read. Every trick is optional: start from whatever you already have, skip everything before your entry point.

Where are you? (each leaf links to a trick)

have topic? ─no─▶ (1) (2)      │ grid team; map it to venue
    │yes                       │
have abstract? ─no─▶ (3)       │ read above citation knee
    │yes                       │
have full draft? ─no─▶ (4) (5) │ grammar expands; un-slop it
    │yes                       │
reviews back? ─no─▶ (6) (8)    │ critics, then audit claims
    │yes                       │
    ▼                          │
(6) (5) (8)                    │ critics replay real reviews;
                               │ rewrite; audit response letter
                               │
at any point:                  │
  more than one author? ─▶ (0) │ every author picks a rung
  dividing up sections? ─▶ (1) │ regrid: authors × sections
  machine did the work? ─▶ (7) │ pause; check the knobs

Where will your team write — before it writes anything?
Write together — eleven rungs of shared writing, from Gemini inside a Google Doc to agents that open PRs overnight. Every author picks their own rung; git merges the difference.

Want to know what your team is actually good at?
Know thyself — “If you're a bit better than most at a particular kind of analysis or a specific domain, choose research topics where you can use that” (Herb Simon). Self-report lies. Grid the team from its own papers: an LLM reads 1,200 paper records in minutes and plays triad interviewee.

An example: a repertory grid showing the overlap of authors' research areas and a field's topic areas. LLMs can generate such grids automatically, in a minute or two. For more, see 1 and 2.

Have strengths — but where should you aim them?
Map team to field — score the venue's areas on the same grid as the people. Centroid, maximin, and blends nominate the topic; blind spots get priced.

Drowning in 10,000 search hits?
Read above the knee — filter by venue inside the query, read above the citation knee, snowball both directions, and treat the download rate as a finding.

Staring at a blank page?
Write to a grammar — a paper parses under a small grammar: intro, lit review, methods, threats, conclusion. Disagree? Edit the productions, regenerate the sample; the experiment costs minutes.

Does your draft smell like AI slop?
Un-slop the prose — keep the STE skeleton (short, active, one meaning per word); subtract the engagement tells (hooks, applause paragraphs). Ship the ban list as a linter, then rewrite the missing bridges by hand.

Want Reviewer 2's verdict before Reviewer 2 gets it?
Assemble critics — an LLM plays Reviewer 2 using the venue's own review criteria; a triage prompt splits findings into machine fixes and human decisions.

Afraid the machine will run away with your paper?
Pause and reflect — every computed step ends in a boxed call-out: the questions a reviewer should ask, and the exact knob to edit so the rest follows the human, not the machine.

Will your claims survive an audit?
Keep the receipts — every citation carries a clickable DOI; claims wait for their statistics (significance + effect size); the conclusion retells the intro with numbers attached; title and abstract must recode with zero flips.

Just want words to paste?
✎ Sample prompts — the triad interview, the lit search, the STE rewrite (with permission for the occasional flourish), learn-a-grammar, Reviewer 2, triage, the zero-flip gate.

Mastered the eight? Make your own.
★ Advanced tricks — don't write a grammar, learn one from papers you admire; venue-tune it; parse drafts against it; learn the anti-grammar from slop.