Research And Automation Playbooks

Competitor Landscape

• Good for: strategy, sales, and market positioning
• Best setup: search tools plus structured synthesis
• Example prompt:

Competitor landscape

Map the competitive landscape for [market]. Compare positioning, pricing cues, target customers, recent launches, and where the whitespace seems to be. End with a crisp strategic takeaway.

Academic Synthesis

• Good for: research briefs and literature scans
• Best setup: scholar search plus clear synthesis format
• Example prompt:

Academic synthesis

Find the most relevant recent academic work on [topic] and synthesize it into a brief for a non-academic team. Separate established findings from emerging or weak evidence.

Daily Social Monitor

• Good for: trend tracking and recurring social intelligence
• Best setup: social tools plus scheduled tasks
• Example prompt:

Social monitor

Set up a daily workflow that checks X, Reddit, YouTube, TikTok, and Instagram for discussion around [topic]. Return the top narratives, unusual spikes, and any posts worth escalating.

Weekly Executive Digest

• Good for: recurring summaries for leadership
• Best setup: connected data sources, scheduled tasks, and markdown/report outputs
• Example prompt:

Executive digest

Create a weekly executive digest that combines [sources]. Keep it to one page: wins, risks, notable changes, and the few items leadership should actually pay attention to.

News Sweep

• Good for: catching up on a company, market, or story without wading through duplicate coverage
• Best setup: Google News, which filters by recency anywhere from the last hour to the last year
• Example prompt:

News sweep

Pull the last week of news on [company/market], deduplicate the stories, and tell me what actually changed.

Find The Right Person

• Good for: outreach lists, sourcing prospects, and verifying an email before you hit send
• Best setup: Hunter — it is built in, nothing to connect
• Example prompts:

Find and verify an email

Find and verify an email address for [name] at [company].

Build a prospect sheet

Discover companies like [description] and enrich the top 10 into a Google Sheet I can work.

One caveat worth knowing: Hunter here is discovery and enrichment only — eight lookup tools for finding people, companies, and email addresses. It is not a CRM; lead lists land in a spreadsheet you own (connect My GSheets under Settings → Tools & Data → Data Sources for the sheet handoff).

Product & Market Scan

• Good for: competitive teardowns, pricing checks, and reading how a category presents itself
• Best setup: Amazon Search and Amazon Product for the listings, Google Images for the visual read
• Example prompts:

Category teardown

Search Amazon for [product category] under [price], pick the five best-reviewed listings, and tear down how they position themselves — naming, price points, claims, and the themes that keep coming up in reviews.

Single-product teardown

Here is a competitor's Amazon link: [URL]. Pull the full listing — price, rating, feature bullets, variants — and compare it point by point against [our product].

Visual market scan

Search Google Images for [product category] packaging and branding, and describe the visual conventions — colours, layouts, claims — we would have to stand out against.

Amazon Search targets your local storefront automatically and can sort by price, average review, newest arrivals, or bestsellers; Amazon Product pulls one listing's full details from a link or product code.

Demand Curve Before The Deck

Every other research tool tells you what got published. Google Trends tells you what people are searching for — so before you argue a market is growing, you can show the curve.

The recipe runs in three passes, and you can ask for all three in one prompt:

1. Interest over time for your term (and up to four rivals, compared on one scale) — the curve, the peak and its date, and whether the second half is up or down on the first.
2. Interest by region — which countries, states, or metros want it most, so the go-to-market has a first market rather than a guess.
3. Rising queries — the long-tail terms growing fastest around your term, which is where demand shows up before the coverage does.

Every time-series and region result also lands as a CSV in the session, so Code Executor can regress it and Report Generator or Presentation Create can cite it without a second search.

• Good for: market sizing narratives, launch timing, category-entry decisions, and sanity-checking hype
• Best setup: Google Trends for the demand read, then Report Generator or Presentation Create for the output
• Example prompts:

Demand curve with rivals

Compare search interest in [our category] against [rival 1] and [rival 2] over the past 5 years. Tell me which is growing, when each peaked, and whether the trend is still intact — then chart it.

Where to launch first

Which countries search most for [product]? Then break the top country down by state and tell me where to launch first.

Seasonality and timing

Show me 5 years of search interest for [product] and tell me what month it peaks, how sharp the peak is, and when we should start the campaign to land ahead of it.

Long tail before the campaign

What are the rising and breakout searches around [term] in the last 3 months? Group them into themes and flag the ones worth building content for.

Two things to hold onto when you read the answer. The numbers are a 0–100 index relative to that exact comparison, never searches per month — a term scoring 8 next to a giant may score 100 on its own. And Trends defaults to worldwide, unlike Google News and Google Search, so name a country when you want one.

Sizing a market from an index

Trends can't size a market on its own — it has no units. It gets there by anchoring: you supply one number whose absolute value you already know, and Trends carries the ratio.

1. Pick an anchor — your own analytics figure is best, a public company's reported segment revenue next, a cited published stat last.
2. Measure the ratio in one call. Anchor term and target term in the same comma-separated query, same window and region, so both sit on one scale. If the target reads 0–2 against the anchor, the ratio is noise — pick a smaller anchor.
3. Convert to demand. Searches aren't buyers: multiply by a conversion rate and an average order value, and name the source of each. Ask for live price points rather than a guessed one.
4. Split by geography. Within a single region call the values are comparable, so the regional shares are a fair way to divide the total into per-country numbers.
5. Calibrate if a public comparable exists — regress its reported quarterly revenue on its own Trends curve, report the R², and apply that relationship. A fitted number is an estimate; the same arithmetic without one is a guess with decimals.
6. Publish a range, not a point, with the assumptions in a spreadsheet you can edit.

Anchored market sizing

Size the market for [product]. Anchor it against [known term or company] where I know the real number is [figure, source], get the search-interest ratio in a single comparison, then convert with a conversion rate and a median price from live listings. Give me a low/base/high range and put the assumptions in a spreadsheet I can change.

Calibrate against a public comparable

Pull [public company]'s quarterly revenue for the last 5 years and its Google Trends curve over the same period, line them up, and tell me how tightly search interest tracks their revenue. If the relationship holds, use it to estimate [private competitor or category].

Ask for the chain in the output — anchor, source, ratio, each multiplier, and the range. "The market is $340m" isn't checkable; "$180–420m, from [anchor] at [figure] × 0.34 search ratio × 2.1% conversion × $89 median price across 40 listings" is.

What to pair it with

Trends answers how much attention, and when. Everything else supplies a dimension it structurally can't:

Pair with	What you get
Google News	Why the curve moved — a spike with coverage behind it is an event, one without is usually noise
Company Financials / Stock Analyzer	Calibration — fit search interest to reported revenue on a public comparable
Google Shopping / Amazon Search	Price and supply — at what price, from how many sellers, with what review velocity
LinkedIn Jobs	Competitor commitment — rising demand plus rivals hiring is a contested market
Google Maps	Physical supply density, to find high-demand, low-supply geographies
Reddit / YouTube	The language buyers actually use, which is what the copy should mirror
Code Executor / Excel Ops	The arithmetic and the editable model, run on the CSV Trends already saved
Task Scheduler	A proprietary demand history — re-run weekly and append

Two rules of thumb: run Trends first, because it's one fast call that decides whether the rest of the work is worth doing; and never let it be the last word — no recommendation should rest on the index alone.

Who Is Hiring, And For What

• Good for: pay research, hiring-market reads, and a company's plans off its open roles — including as an alt-data signal for diligence, market sizing and competitor tracking
• Best setup: Google Jobs for the listings, LinkedIn Jobs alongside it for depth, Code Executor or Excel Ops to work the saved CSV
• Example prompts:

Shortlist live openings

Find [role] jobs in [city] posted in the last week, and give me the ones with a published salary first — title, company, pay, and the apply link.

What the role actually pays

Pull 30 [role] openings in [city], then tell me the salary range companies are actually advertising and which requirements come up in most postings.

Read a competitor's roadmap

What is [company] hiring for right now? Group the openings by team and tell me what that says about where they are investing.

Filters go in the sentence — "remote", "part time", "internship", "no degree", "since yesterday" — because Google Jobs parses them from the query rather than from separate fields. Every search saves a CSV into the session carrying the complete description text for each posting, so filtering 30 roles by requirement is one follow-up question, not 30 more searches. Name a city to anchor the search; if Google does not recognise the place, Alfrada OS retries with it written into the query and tells you when the results ended up country-wide. For LinkedIn-only listings with applicant counts, ask for LinkedIn Jobs instead. If you are the one job hunting rather than researching the market, read the next section — the flow is different.

Open roles as alt data

A company's job postings are it stating, in public and at its own expense, where the next money goes. They lead reported headcount and operating costs by a quarter or two, name a product or geographic push before the announcement, and show a retreat as postings quietly disappearing. For a private company — an acquisition target, a competitor with no filings, a market with no published research — this is often the only continuous operating disclosure that exists.

What a target is really building

Pull [company]'s live job postings with full descriptions. Break them down by function, seniority and location, pull out the technologies and certifications the requirements name, and tell me what that says about what they are building and where they are expanding.

Competitor commitment to a market

Is anyone actually staffing for [market or product category]? Search live postings for the roles that market needs across [region], tell me which companies are hiring for it, in which cities, at what pay — and whether that makes this a contested market or an opening.

Weekly hiring tracker

Every Monday, run the same job searches for [companies] and append the results to a workbook: role counts by function, how many are new since last week, and the salary bands. I want a hiring trend I own, not a snapshot.

Two things make this trustworthy rather than a number you liked the look of:

It is a sample, not a census. Google Jobs returns up to 30 listings a call and LinkedIn Jobs up to 100 — neither is "how many openings exist", so a raw count is never headcount growth. What compares cleanly run to run is composition (function, seniority, location mix), flow (postings that are new since last week, matched on their apply links), and pay. Ask for the query set to be held fixed; changing the wording rebases the series and the change reads as news.

Job data lies in five specific ways, and Alfrada OS handles each: one role syndicated to eight boards is one role, not eight; an evergreen req reposted monthly looks permanently fresh; agencies list roles under their own name; retail and logistics hire on a calendar, so compare like-for-like periods; and a company that stops posting has frozen hiring — that absence is a finding, not a blank.

The first run is not empty history, either: postings carry their age, so bucketing what is live today by posting date reconstructs roughly the last month of hiring flow before you have collected anything.

Find A Job, And Fit Your CV To It

• Good for: job hunting end to end — finding roles, seeing how you actually score against each one, and fixing the CV and LinkedIn profile before you apply
• Best setup: your CV uploaded to the session (or your LinkedIn URL), Google Jobs and LinkedIn Jobs for the roles, Doc Patch to edit the CV in place
• Example prompts:

The whole job hunt

Here's my CV. Find me [role] jobs in [location or remote], score each one against my background, and tell me which are worth applying to and what to change first.

Score one posting

How well does my CV match this role? Show me the hard requirements I miss, the exact keywords the posting uses that my CV doesn't, and whether I should apply as-is or fix it first.

Rewrite for a target role

Tailor my CV for [role] at [company] using only experience I actually have, keep my formatting, and give me the LinkedIn headline, About opener and skills to match.

Alfrada OS starts by looking for what you have already given it: a CV or resume uploaded in this session or any past one (History Searcher reads PDF, DOCX and TXT in full), anything stored in memory about your role, location and seniority, and your public LinkedIn profile if you paste the URL. It asks — once — only for what is genuinely missing, and the questions that change the answer: target title and seniority, location and remote preference, contract type, salary floor, and work authorisation. That last one silently disqualifies half a shortlist, so it always gets asked.

Then it searches both boards: Google Jobs for coverage across LinkedIn, Indeed, company career pages and job boards, and LinkedIn Jobs for LinkedIn's own listings and its structured filters (work type, contract type, experience level, posted-within). Roles appearing on both are the ones being pushed hardest. Duplicates are merged before anything gets scored.

How the ATS scoring works

An applicant tracking system flattens your CV to text and matches it against the posting, so the score is built the same way — and the arithmetic is shown, never just a number:

• Hard requirements (40) — years, licence or degree where it is genuinely required, named tools, domain. One miss caps a role near 60; two makes it a long shot, and it says so.
• Keyword coverage (25) — ATS matching is literal, not semantic. "Kubernetes" and "k8s" are different strings. The list of terms the posting uses that your CV never states in those words is the actionable part of the whole exercise.
• Title and seniority alignment (15) — a most-recent title three levels away from the posting gets filtered before a human reads it.
• Recency and progression (10) — is the relevant experience current, and does the trajectory point here.
• Parse safety (10) — this scores the file, not you: multi-column layouts, tables, text living inside a logo, contact details in the header, and image-exported PDFs are the standard ways a strong CV reaches the ATS as gibberish.

What comes back

A ranked shortlist (a spreadsheet once it passes ten roles), the two or three gaps costing you the most points per role, an edited CV, and paste-ready LinkedIn text — headline, About opener, and the skills recruiters filter on. An uploaded .docx is patched in place so your formatting survives; a PDF can't be patched, so you get a rebuilt version and are told it is a rebuild.

One rule it will not bend: missing keywords go in only where you genuinely have the experience. If a posting wants something you have never done, that is reported as a real gap rather than written into your skills section — an ATS-optimised claim you can't defend fails at the first screen, and the CV is the document you get interviewed against.

Deep Research With Smith

Smith is Alfrada's autonomous deep-research agent: it forges an independent worker that goes off, researches, and reports back — several workers in parallel if the job splits. You pick how hard it digs:

• Morpheus — quick lookup, around 13 tool calls

• Trinity — standard research, around 33

• Neo — exhaustive deep dive, up to 65

• Good for: questions that need dozens of searches and real synthesis, not one pass

• Best setup: Smith, with the effort level named right in your prompt

• Example prompt:

Neo deep dive

Run a Neo deep dive on [market], with sources for every claim.

Landscape Scan In Swarm Mode

A landscape has many independent branches — competitors, pricing, regulation, technology — and researching them one at a time is slow. Swarm mode fans the work out to parallel agents, each taking a branch.

Switch it on from the chat input → model selector → Mode row → Swarm. It is available on every plan, including the free Community plan. One thing to know: in Swarm the model and Effort pickers don't apply — Swarm orchestrates its own models per agent. See Swarm for how it works.

• Good for: broad landscapes with many independent branches
• Best setup: Mode set to Swarm; describe the branches and the merge you want
• Example prompt:

Swarm landscape scan

Map the [market] landscape — competitors, pricing, regulation, and recent funding. Split the branches across agents and merge everything into one brief with a strategic takeaway.

Same-Day Follow-Through

• Good for: fast-moving stories you want re-checked without keeping the tab open
• Best setup: the Wake-Up Timer — Alfrada OS wakes itself up in this same conversation and picks the work back up
• Example prompt:

Wake-up follow-through

Wake up in 2 hours, re-check the [topic] thread, and update the brief if anything moved.

For recurring schedules rather than a one-off nudge, use Automations.

Ship The Findings

Research that stays in the chat window dies there. Two good endings: a polished document to circulate, or a living page you keep watching.

• Good for: turning a finished investigation into something you can hand over or keep open
• Best setup: Report Generator for a PDF; HTML Dashboard for a living readout
• Example prompts:

PDF research report

Write the research up as a polished PDF report — executive summary, findings, and a sources appendix.

Living landscape dashboard

Turn the landscape into a live HTML dashboard — key players, signals to watch, and latest news — and refresh it whenever I ask.

For everything dashboards can do — auto-refresh, exports, theming — see the Dashboards playbook.

Standing Social Monitoring

The Daily Social Monitor workflow above is the manual version: you ask, Alfrada OS reports. For always-on monitoring that messages you — an agent that watches your feeds on a schedule and pings you on WhatsApp when something needs attention — see Argus On WhatsApp. Argus is available on the Pro Max and Black plans.

Under the hood — tool identifiers

• google_news — Google News (recency filters last_hour, last_day, last_week, last_month, last_year)
• hunter_* — the eight Hunter tools: hunter_domain_search, hunter_email_finder, hunter_email_verifier, hunter_email_count, hunter_discover_companies, hunter_people_enrichment, hunter_company_enrichment, hunter_combined_enrichment
• amazon_search, amazon_product — Amazon Search / Amazon Product
• google_images — Google Images
• agent_smith — Smith (effort levels morpheus/trinity/neo, aliases low/medium/high; step budgets 13/33/65)
• wake_me — Wake-Up Timer (same-conversation self-wake-ups)
• report_generator — Report Generator (PDF/DOCX)
• html_dashboard — HTML Dashboard (live page; exports to PDF/PNG)