Jeff Bullas

This is one of the most critical decisions you’ll make.

Short Answer: The most important factors are your long-term content format, your need for compliant audience data, and your future monetisation strategy.

Choosing a host is a foundational business decision, not just a technical one, as it dictates the future capabilities of your show.

You must first evaluate your audio content format; a simple, 30-minute mono podcast has vastly different storage and bandwidth needs than a 90-minute, high-production stereo audio format, so your provider’s plan must align with your show’s publishing ambitions without punishing you for growth. The second factor is the analytics format, which must be IAB-certified. This ensures the text-based data and graphs you receive are accurate and standardised, giving you a true picture of your content’s performance. The third factor is your monetisation format; you must select a host that supports the revenue streams you plan to use, whether that’s a premium subscription audio format, a dynamic ad insertion format, or simple text links in your show notes. The most common mistake is choosing a “free” host, only to discover later that their terms lock you out of these critical monetisation formats, costing you far more in the long run.

Cheers,
Jeff

This is the most critical strategic decision you’ll make.

Short Answer: Your niche is too broad if you can’t describe your ideal listener in one sentence, and it’s too specific if you can’t brainstorm 50 episode titles right now.

Finding this balance is the key to creating a sustainable show that can actually build a dedicated audience.

The problem with a broad text format like “a business podcast” is that it targets no one, so it’s impossible to create content that feels specific and essential. This is a listener definition problem. The test is to write a single sentence that describes your target listener, and if it includes the word “everyone” or “people,” your niche is too broad. On the other hand, a niche is too specific when it’s a content volume problem. This is where your chosen text format is so narrow that you will run out of audio content after a dozen episodes. The test here is to sit down and write fifty potential episode titles. If you can’t, you haven’t given yourself a sustainable format for long-term production. The goal is to define a specific audience format, not a specific topic format, which gives you the freedom to explore many topics through the lens of that one audience.

Cheers,
Jeff

Great. You’ve got the lever: triage + KPIs. Now let’s turn it into a 90-minute build you can copy, with ready-to-paste templates for Notion and Obsidian, a bulletproof triage prompt, and a weekly review that keeps PARA clean without busywork.

Do / Do not

• Do: capture everything to a single Triage bucket, then classify.
• Do: enforce one field on every Project: Next Action.
• Do: link, don’t duplicate (relations or backlinks).
• Do not: exceed 3 tags; if you need more, your structure is unclear.
• Do not: migrate everything; seed with current work only.

What you’ll need

• Pick one tool for 2 weeks: Notion or Obsidian.
• Any AI chat assistant.
• Your top 5 Projects, 5 Areas, 10 Resources.

90-minute setup

1. Create your buckets (10 minutes)
   • Notion: one database per bucket (Projects, Areas, Resources) plus an Archive checkbox or a fourth database. Add a simple Triage view that shows items with empty Next Action.
   • Obsidian: folders P, A, R, Archive, and one Triage folder for all new notes.
2. Paste these lean templates (25 minutes)
   • Notion properties (everywhere): Title, Summary (text, 1–2 lines), Next Action (text), Tags (multi-select, max 3).
     • Projects also: Status (select: Planned, Active, Waiting, Done), Due Date (date), Area (relation), Last Reviewed (date), Review Cadence (select: Weekly, Monthly), Review Due (formula: dateAdd(Last Reviewed, 7 for Weekly or 30 for Monthly)).
     • Formula hints: Review Due = if(prop(“Review Cadence”) == “Weekly”, dateAdd(prop(“Last Reviewed”), 7, “days”), dateAdd(prop(“Last Reviewed”), 30, “days”)); Review Status = if(empty(prop(“Next Action”)), “Needs Action”, if(prop(“Review Due”) <= now(), “Review”, “OK”)).
     • Views: Triage (filter Next Action is empty), Active (Status = Active), Stale (Status = Active AND Review Status = Review).
   • Obsidian YAML frontmatter (copy these into four template files):
     • Project: type: project; summary: ; next_action: ; status: Planned|Active|Waiting|Done; area: ; tags: []
     • Area: type: area; summary: ; review_cadence: weekly|monthly; owner: ; tags: []
     • Resource: type: resource; summary: ; tags: []
     • Archive: type: archive; reason: ; tags: []
   • Body sections (all tools): Projects = Purpose, Scope, Links, Next Action, Notes. Areas = Purpose, Standards, Links. Resources = Topic, Best links, Notes. Archive = Reason, Links.
3. Build the AI triage lane (10 minutes)
   • Commit: all new notes land in Triage. AI classifies and proposes the Next Action before you move the note.
4. Seed the system (25 minutes)
   • Take your top 3 Projects, 5 Areas, 10 Resources. For each, run the triage prompt below. Paste results into Notion properties or Obsidian YAML + body, then move out of Triage.
5. Operationalize the weekly review (20 minutes)
   • Book a recurring 20–30 minute slot.
   • Create a saved view/search: items with empty Next Action and items due for review (Notion filter or Obsidian search: next_action: is empty OR folder:Triage).
   • Run the Weekly Review prompt; clear the Triage and “Needs Action” views to zero.

Copy-paste AI prompts

• PARA Triage (Notion or Obsidian): You are my PARA triage assistant. Return two parts. Part A: Bucket (Project|Area|Resource|Archive); Title (5–9 words, verb-first if Project); One-sentence Summary (≤25 words); Next Action (imperative, ≤15 words); Up to 3 Tags; Suggested Area (if relevant). Part B: Output format. If I say NOTION, list Property: Value lines matching my database names. If I say OBSIDIAN, output YAML frontmatter using keys: type, summary, next_action, tags, plus any relevant keys, then a short body under my template headings. Keep it concise. Input: [paste note or dump text]
• Weekly PARA Review: Act as my PARA reviewer. I’ll paste changed items since last week with short context. For each: 1) confirm or correct PARA bucket, 2) propose a tighter Title if needed, 3) generate one realistic Next Action, 4) flag missing links (Areas/Resources), 5) suggest archive if stalled. Finish with a 5-item Weekly Focus list and 3 items to archive.
• Bulk Migration Helper: I will paste several notes separated by — lines. For each, output either NOTION properties or OBSIDIAN YAML + a 2-line body, plus one Next Action. Preserve unique facts; avoid duplicate tags; keep summaries ≤25 words.

Worked example — Obsidian

• YAML for a new project note (paste, then fill blanks): type: project; summary: Launch campaign for updated product; next_action: Draft launch checklist; status: Active; area: Marketing; tags: [launch, q3]
• Body sections: Purpose: clarify objectives and scope. Scope: channels, budget, timeline. Links: link to press list and checklist notes. Next Action: Draft launch checklist. Notes: capture decisions.
• Add backlinks to two Resources (press list, checklist). Don’t copy content.

Insider tricks

• Name projects verb-first: “Ship v2 website,” not “Website v2.” It clarifies Next Action instantly.
• In Notion, add a “New Project” page template that pre-fills sections and an empty Next Action so you can’t forget it.
• In Obsidian, create a saved search: path:Triage OR next_action:s*$ to show anything unclassified or missing a Next Action.

KPIs (weekly)

• % Projects with a Next Action (aim 90%+).
• Average search time (target under 60 seconds).
• Time from capture to classified (under 24 hours).
• Weekly review done (yes/no; target 4/4).

Mistakes and quick fixes

• Too many properties. Fix: cap at five; move the rest into body text.
• Resources duplicated across projects. Fix: link once; reference everywhere.
• Stalled projects. Fix: no Next Action = set Status to Waiting or Archive.

7-day plan

1. Day 1: Stand up buckets and templates; set Triage as the default landing spot.
2. Day 2: Migrate top 3 Projects using the Triage prompt.
3. Day 3: Migrate 10 Resources; link, don’t duplicate.
4. Day 4: Create views/searches for Triage and “Needs Action.”
5. Day 5: First Weekly Review; archive two low-value items.
6. Days 6–7: Measure KPIs, trim tags to ≤3, tighten templates.

Pick your tool (Notion or Obsidian), and I’ll tailor the exact database fields or YAML templates to your setup next.

Build a tiny learning engine this week. Keep one topic, one outcome, and let AI handle the heavy lifting: a tight reading list, clean summaries, and flashcards you’ll actually remember.

Why this works: You capture highlights, AI turns them into testable cards, and spaced repetition converts that into long-term memory. Small, steady sessions beat marathon reading.

What you’ll set up (once):

• AI chat tool
• Spreadsheet with columns: Item, Type, Time, Why, Priority, Status
• SRS app (Anki, Quizlet, or RemNote)
• Note/highlight tool (Kindle, phone notes, Evernote)

Step-by-step

1. Pick one target. Topic + outcome in one line (e.g., “Understand core negotiation moves well enough to use 2 at work in 6 weeks”).
2. Get a laser-focused reading list. Use the prompt below to ask AI for 6 items (2 books, 2 articles, 2 podcasts/talks) ordered by impact for your goal. Push for short, practical picks.
3. Plan your sessions. 3 short blocks/week (25–40 minutes). For each block, aim for 1 chapter or 30 minutes of audio.
4. Capture highlights the right way. 5–10 bullets per session, one idea each, written in your words. Avoid pasting large passages.
5. Create summaries + flashcards with guardrails. Paste highlights into AI and use the flashcard prompt. Expect 6–10 cards per session: half cloze (fill‑in), half Q&A (application).
6. Import once, review daily. Export the AI’s CSV and import to your SRS app. Do 10–20 minutes daily. Edit 3–5 weak cards each week.
7. Track the right numbers. New cards/week: 15–40. Recall rate: 70–90%. If recall drops below 60%, slow down card creation.

Premium shortcut: Pre-commit your card quality. Ask AI for short, clear fronts, single-deletion clozes, and one-sentence “why it matters” notes. This trims review time by ~20–30%.

Copy-paste prompts

• Reading list builder (use as-is):“I have [6 weeks] with [3 hours/week]. Learning style: [prefer audio on weekdays, 1 short book on weekends]. Topic: [TOPIC]. Outcome: [SPECIFIC OUTCOME]. Propose a 6‑item list ordered by impact: 2 books (ideally <250 pages), 2 articles, 2 podcasts/talks. For each, give: Title, Type, Estimated time, Why it matters (1 line), First 3 pages/sections to sample. Then end with: ‘If you only do one, start with: [X]’.”
• Highlights → summary + cards (CSV, ready to import):“I’m studying [TOPIC] to achieve [OUTCOME]. Here are my highlights (1 idea per line): [PASTE HIGHLIGHTS]. Create: 1) a 2‑sentence summary, 2) 8 flashcards total: 4 cloze, 4 Q&A. Rules: one idea per card; Front ≤ 18 words; Back ≤ 25 words; plain language. Cloze: one deletion only, show the full sentence with {{c1::deletion}}. Include an Extra field with a 6–12 word memory hook. Tags: [topic];[source];[easy|medium|hard]. Output only CSV with header: Type,Front,Back,Extra,Tags; quote every field with double quotes; replace internal double quotes with single quotes.”
• Fix bad cards (weekly tidy):“These cards are hard or confusing: [PASTE 5–10 CARDS IN CSV]. Improve by: splitting multi‑idea cards, simplifying wording, adding context, or switching fact cards to application. Return only CSV with the same header. Tag each revised card with ‘revise’.”
• Application drills (cement understanding):“Based on these highlights: [PASTE], write 3 brief scenarios (2–3 sentences) where I’d apply the ideas at work. After each, add one Q&A card that asks me to choose the best action and explain why (≤25 words). Return as CSV with Type,Front,Back,Extra,Tags.”

Worked example (behavioral economics)

• Highlight: “People prefer smaller immediate rewards over larger delayed ones.”
• Cloze: Front: “Preferring smaller immediate rewards over larger delayed ones is {{c1::present bias}}.” Back: “Present bias (hyperbolic discounting).” Extra: “Now feels bigger than later.” Tags: behavioral;chapter1;easy
• Q&A: Front: “Why choose $10 today over $100 next month?” Back: “Present bias—immediate rewards feel overweighted; use commitment devices.” Extra: “Make future easier than now.” Tags: behavioral;chapter1;medium
• Application: Front: “One workplace fix for present bias?” Back: “Default automatic savings or scheduled purchases.” Extra: “Make good default.” Tags: behavioral;chapter1;medium

Common mistakes and simple fixes

• Too many tiny cards. Fix: 3–6 solid cards per section; merge duplicates.
• Verbatim, context‑free clozes. Fix: full sentence with one cloze; add a short Extra note.
• Fronts that are mini‑essays. Fix: ≤18 words on the front; one idea only.
• Skipping reviews. Fix: 10–20 minutes after coffee or commute; protect that slot.
• Recall stuck below 60%. Fix: add more application cards, slow new cards for a week.

What to expect

• Day 2: A tight reading list that respects your time.
• Day 4: First 15–25 cards imported and reviewing smoothly.
• Week 2: Faster recall, lighter editing, clearer tags.

7‑day starter plan

1. Day 1: State your topic + outcome. Run the reading list prompt. Log items in your sheet.
2. Day 2: Read/listen 30–40 minutes. Capture 5–10 highlights.
3. Day 3: Run the highlights → cards prompt. Import the CSV into your SRS.
4. Day 4: Do 10–20 minutes of reviews. Edit 3 weak cards.
5. Day 5: Another short read/listen. Add 6–8 new cards.
6. Day 6: Run the “Fix bad cards” prompt on your hards.
7. Day 7: Light review + quick reflection: What one tweak improves next week?

Insider tip: Tag three ways—core (must know), useful (nice to know), nice (optional). Study core daily, useful every other day, nice weekly. It keeps energy where it counts.

Start small. One topic, one outcome, a few highlights. The AI does the grunt work; you keep the judgment. That’s how your reading turns into knowledge you can use.

Spot on — the 2-pass workflow is the signal. Your preflight and A/B/C prioritization turn AI drafts into a calm, schedule-ready shoot. Let’s add one more layer: a “board pack” that the AI can produce in one go — beats, shots, coverage, storyboard prompts, CSV, and risks — so you move from idea to call sheet without retyping.

Why this works: AI is great at structured outputs. Give it a tight brief, a clear priority system, and a coverage rule, and it will return a usable storyboard and shot list you can refine with your DP in minutes.

What you’ll need

• One-paragraph script or treatment
• Two-sentence objective (message + mood)
• Beats with rough seconds per beat
• 3 mood keywords or 3 reference images
• Constraints: budget band, camera/lenses, locations, max crew

Step-by-step (preflight + board pack)

1. Preflight 10: Purpose (what must the viewer feel/do), People (how many on camera), Places (where/time of day). Note three A-priority shots.
2. Beat grid: 3–8 beats with seconds. Tag must-have moments with an asterisk.
3. Director pass: Ask for two visual options per beat (emotion, framing, blocking).
4. Pick A-shots: Lock your three must-haves. Everything else becomes B or C.
5. Production pass: Translate chosen visuals into specs (INT/EXT, framing, move, lens suggestion, simple gear, minutes per shot).
6. Coverage rule: Use a “coverage triad” where it matters — Wide (context), Mid (action), Close (emotion). You can drop C shots if time tightens.
7. Export: Ask AI for a CSV-friendly block to paste into your spreadsheet.
8. Storyboard prompts: Generate simple image prompts for 6–12 frames you’ll show stakeholders.
9. Risk + buffer: Flag time sinks, add ~20% buffer to any complex shot, and list one fallback angle per A-shot.

Copy-paste AI prompt: Commercial Board Pack (director + production in one)

Project: “[TITLE]”. Objective (2 sentences): “[message + mood]”. Script (one paragraph): “[paste]”. Beats with durations: [Beat 1 – 5s; Beat 2 – 7s; Beat 3 – 8s]. Mood keywords: [e.g., warm, energetic, modern]. Constraints: budget [low/med/high], camera [model], lenses [list], locations [list], max crew [#].
Output as a board pack:
1) Beat summary: intent, emotion, and key action for each beat.
2) Shot list per beat with priority A/B/C: number, INT/EXT, action, framing (W/M/CU), camera move, approx duration (sec), and one visual reference keyword.
3) Production notes per shot: lens suggestion, tripod/handheld, minimal gear, estimated minutes to shoot, and any dependency (talent, prop, light).
4) Coverage triad for each must-have moment: Wide (context), Mid (function), Close (emotion). If time is tight, suggest which to drop first.
5) Storyboard image prompts: one sentence per shot describing the frame for a simple sketch (include subject, light, color mood, and composition cue).
6) CSV block for spreadsheet: Shot #, Priority, Beat, INT/EXT, Framing, Move, Duration (sec), Lens, Gear, Est. Minutes, Dependency.
7) Risks + contingency: list top 3 time-risk shots and propose a fallback angle for each. Add a 20% time buffer to complex shots and show the adjusted total minutes.
Keep language simple and non-technical.

Worked micro-example (15s, 3 beats)

• Objective: Show a local gym as friendly and energizing. Mood: warm, motivating, real-people.
• Beats: 1) Arrival smile – 5s, 2) Quick workout montage – 7s, 3) Post-workout glow + CTA – 3s.
• A-shots: A1 warm entrance wide; A2 close-up effort moment; A3 satisfied smile + logo.

1. Shot 1 (A) — INT, 5s: Wide lobby as member enters; move: gentle push-in; visual: morning warmth. Prod: 24–28mm, tripod/slider, 10 min.
2. Shot 2 (B) — INT, 3s: Mid treadmill feet; move: slight pan; visual: rhythm. Prod: 35–50mm, tripod, 6 min.
3. Shot 3 (A) — INT, 3s: Close-up effort face; move: static; visual: determination. Prod: 85mm, handheld, 8 min.
4. Shot 4 (C) — INT, 3s: Dumbbell rack insert; move: rack focus; visual: sleek metal. Prod: 50mm, tripod, 6 min.
5. Shot 5 (A) — INT, 4s: Mid smile with logo wall; move: slow dolly out; visual: achievement. Prod: 35mm, slider, 10 min.

Coverage triad note: Beat 2 uses Mid (feet), Close (effort), optional Wide (floor patch); drop C first if timing slips. Total est. shooting minutes (incl. setup): ~40–45 with 20% buffer.

Insider tricks

• Same-lights alternate: Ask AI for a “same-light alt” for each A-shot (new angle, same key light) — gives you an instant backup without re-lighting.
• Transition anchors: Request 1–2 neutral cutaways (hands, signage) to smooth edits and salvage continuity.
• Shot economy ratio: Plan for 60% A, 30% B, 10% C time allocation; it keeps focus on what sells the story.

Common mistakes & fixes

• Vague objectives — Fix: force two sentences before anything else.
• Too many moves — Fix: cap to one purposeful move per A-shot; static for B/C.
• No durations — Fix: seconds per beat + minutes per shot, always.
• Skipping fallbacks — Fix: require one alternate for each A-shot.

3-day action plan

1. Day 1: Preflight 10 + beat grid; pick three A-shots.
2. Day 2: Run the Board Pack prompt; review with DP; adjust priorities and minutes.
3. Day 3: Generate 6–12 storyboard frames; export CSV to your schedule; lock 48–72 hours before shoot.

What to expect: a first board pack in 20–40 minutes, then one review pass to tighten lens/gear and a final pass to lock risks and buffer. The result is a storyboard and shot list your crew can follow without guesswork.

Final nudge: Use AI to structure, prioritize, and timebox. Keep human judgment for taste, blocking, and safety. Get your A-shots first, and let B/C expand only if time allows.

Nice point — that 5-minute win is exactly the kind of quick data point that turns interest into action. A short, repeatable test removes the guesswork and gives you something defensible to show stakeholders.

Quick context

Do this like an experiment: small sample, clear KPI, conservative assumptions. The goal is a reliable signal, not perfection.

What you’ll need

• A single workflow (e.g., client report, sales follow-up, expense reconciliation).
• Baseline data: stopwatch 10–20 tasks or 1–2 weeks of logs.
• Hourly value or opportunity cost (use the lowest realistic rate).
• Tool cost estimates and an hourly rate for setup/training.
• 3–4 sample items for the 5-minute test and 20 for a short pilot.

Step-by-step — do this

1. Pick the task and define 1–2 KPIs (time per task, error minutes, revenue impact).
2. Run the 5-minute test on 3 recent items: time yourself, then run the AI and record time + quality.
3. If the test looks promising, run a matched 20-task pilot and capture the same metrics.
4. Calculate raw savings: (baseline mins – AI mins) × tasks/year ÷ 60 × $/hr + direct revenue change.
5. Apply a conservative overhead (15–25%) and run a ±20% sensitivity check.
6. Report: assumptions, calculations, adjusted benefit, first-year cost, first-year ROI = (benefit – cost) / cost.

Copy-paste prompts

5-minute summary prompt (use on a 10–15 minute report)

“You are an assistant that converts raw project notes into a one-page client report. Given the notes: [paste notes], produce: 1) a 3-sentence executive summary, 2) 5 bullet-point highlights, 3) 2 recommended next steps with owner and deadline. Keep language clear and non-technical.”

ROI estimator prompt (use after you’ve collected numbers)

“You are an ROI analyst. Given: baseline average time per task = [X minutes], sample size = [N], hourly value = [$Y], error/rework minutes per task = [Z], AI pilot average time per task = [A minutes], pilot error minutes = [B], annual tool cost = [$C], setup hours = [H] at [$rate/hr], and conservative overhead = [P%]. Calculate: 1) annual time saved (hours), 2) annual monetary value of time saved, 3) adjusted value after overhead, 4) total first-year cost, and 5) first-year ROI as (adjusted value – cost)/cost. Show calculations and list assumptions.”

Example (concise)

Baseline: 6 hrs/week on client reports. AI pilot: 2 hrs/week. Time saved = 4 hrs × $120/hr = $480/week → $24,960/year. Tool cost = $2,400/year + setup $600. Apply 20% overhead → adjusted benefit ≈ $19,968. First-year ROI ≈ (19,968 – 3,000) / 3,000 ≈ 5.66x.

Mistakes & fixes

• Mistake: Optimistic hourly rate. Fix: use the lowest plausible $/hr or opportunity cost.
• Mistake: Too short a pilot. Fix: minimum 20 tasks or 2 weeks.
• Mistake: Ignoring quality. Fix: convert fewer errors into rework minutes or lost revenue.

7-day action plan

1. Day 1: Pick workflow and KPIs.
2. Days 2–3: Collect baseline (10–20 tasks).
3. Day 4: Run the 5-minute test on 3 items.
4. Day 5: Run 20-task pilot if test looks good.
5. Day 6: Run the ROI estimator prompt and apply overhead + sensitivity.
6. Day 7: Prepare a one-page brief and decide scale/iterate/stop.

Small, fast experiments win. Get a real data point, be conservative, and iterate — that’s how you turn AI curiosity into trusted ROI.

Hook: You don’t have to read every new paper — you need a fast system that finds the few that matter. AI can do the heavy lifting so you spend your energy on decisions, not triage.

Quick correction: don’t rely only on very narrow keywords. That can miss relevant cross-disciplinary work. Use a small set of focused keywords plus author/journal follows and an occasional broader semantic search or citation-tracking run.

What you’ll need:

• A list of 5–10 priority topics/keywords and 5–10 key authors or journals.
• An aggregator (RSS reader or alert-capable database) or tools that can push new paper metadata into your workflow.
• An AI assistant that can read abstracts or PDFs and return short triage/summaries (many cloud LLM tools or your reference manager plugins).
• A place to store/tag papers (reference manager, note app, or simple folders).

Step-by-step setup:

1. Define focus: write 5–10 phrases (topics + methods + populations). Keep one or two broader phrases for discovery runs.
2. Create alerts: set saved searches on 2–3 sources (PubMed, arXiv, Scopus, or journal alerts). If a source lacks RSS, use the site’s saved search or an aggregator that supports APIs.
3. Automate intake: route new results to an RSS reader, email folder, or Zapier/Make flow that posts titles+abstracts to your AI tool or note app.
4. AI triage: use the copy-paste prompt below to ask the AI for a one-line verdict and a 3-bullet summary per abstract. Tag as Immediate / Maybe / Skip.
5. Summaries & actions: for Immediate items, run a second prompt to generate a 1-paragraph takeaway and the practical action (read full methods, contact author, add to review, etc.).
6. Weekly review: spend 30–90 minutes on the Immediate folder. Move items, update keywords, and archive summaries.

Copy‑paste AI prompt — triage (use exactly):

“Read this abstract and give me: 1) a one-line relevance verdict for my priorities (diabetes clinical trials; machine learning in medical imaging), using: Read Now / Maybe / Skip; 2) three short bullets: key finding, method, and sample size; 3) one sentence: why I should care (clinical or research implication). Keep it under 70 words.”

Copy‑paste AI prompt — summary + action:

“For this paper (title + abstract + link), write 1 short paragraph summarizing the main result and 1 sentence that states the next practical action I should take (e.g., read methods, replicate analysis, cite in review, contact author). Add 3 tags from this list: [Clinical, Methods, ML, Small-N, RCT, Preprint].”

Example (what to expect):

• Input: abstract → Output: “Read Now. ML model reduced false positives in chest X‑rays; CNN on 10k images; multicenter. Action: review methods for bias (1 paragraph). Tag: ML, Methods.”

Mistakes to avoid & fixes:

• Relying only on abstracts — fix: flag high-priority papers for full-text checks focusing on methods and sample size.
• Too many narrow alerts — fix: add periodic broad searches and track high-citation authors to catch cross-field work.
• Manual copy/paste overload — fix: automate with integrations (RSS→AI) or reference manager plugins.

3-step action plan (next 2 hours):

1. Write your 5–10 focus phrases and pick 5 authors/journals.
2. Create alerts on one primary source and route results into an RSS or email folder.
3. Use the triage prompt on 10 recent abstracts and tag them — adjust keywords based on results.

Closing reminder: start small, automate quickly, and iterate. The goal is a steady, manageable stream of high-value papers — not inbox zero.