# Research prompt library

Research prompts tested in Autumn (https://www.autumn.ai) in October 2026. Each was run and kept because its rows held up against their sources. Interactive version: https://www.autumn.ai/resources/prompt-library. JSON: https://www.autumn.ai/resources/prompt-library/prompts.json

Every prompt has three parts: a task (what, how many, the window), rules (what counts as proof and what to leave blank) and columns (the shape of the table).

## One-line prompts

For one person or company you already know, a type word and a name is enough.

### Profile a person

You know who it is and want everything public about them in one record. (Sales, Recruiting and experts, Investors)

Looks in: LinkedIn, Personal sites, Talks and podcasts, Press.

```text
Profile: Karri Saarinen, Linear
```

Change: person.

From testing: Add the company after the name when the name is common. A long brief listing the fields came back no richer than this one line.

Link: https://www.autumn.ai/resources/prompt-library#person-profile

### Profile a company

Before a call, a deal or a partnership: one record with the company's people, money, product and news. (Sales, Founders and market, Investors)

Looks in: Company site, Funding announcements, Press, Job posts.

```text
Profile: Linear (linear.app)
```

Change: company.

From testing: Put the domain in brackets so it researches the right company. Our longer brief asking for each field by name came back with fewer fields than this.

Link: https://www.autumn.ai/resources/prompt-library#company-profile

### Research an account from its domain

A signup or a target account where all you have is the domain. (Sales)

Looks in: Company site, Funding announcements, Press, Job posts.

```text
Account: ramp.com
```

Change: domain.

From testing: A bare domain is enough. It resolves the company first, then builds the same full record as a company profile.

Link: https://www.autumn.ai/resources/prompt-library#account-research

### Find a company's tech stack

You sell to teams that run a given tool, or you need to know what a company already pays for. (Sales, Founders and market)

Looks in: DNS records, Subprocessor lists, Job posts, Site code.

```text
Tech stack: ramp.com
```

Change: domain.

From testing: It reads DNS records, subdomains, subprocessor lists, job posts and the site's own code, so tools show up with proof, not a vendor's guess. Versions appear only where the site's code exposes them.

Link: https://www.autumn.ai/resources/prompt-library#tech-stack

### Map who reports to whom

You need the real buying path or a team's structure, not a flat list of titles. (Sales, Recruiting and experts)

Looks in: LinkedIn, Team pages, Press.

```text
Org chart: Ramp (ramp.com)
```

Change: company.

From testing: Ask for the org chart, not for "reports to" lines in job posts. Most posts never name a manager; the org chart prompt builds the tree from profiles, team pages and press, with evidence for each line.

Link: https://www.autumn.ai/resources/prompt-library#org-chart

## Prompts with rules

## Sales

Inbound, accounts, buyers and timing.

### Research inbound signups

New signups arrive as bare emails and you want to know who they are before you reply. (Sales)

Looks in: Company sites, LinkedIn, GitHub, Press.

```text
Research these inbound signups. For each email, identify the person and their current company. Keep exactly one row per input, in this order, and never merge or drop rows:
jane@company.com
sam.lee92@gmail.com
alex@startup.io

Rules:
- Identify the person only from public evidence that ties this email, or the name and company it implies, to them (company site, LinkedIn, GitHub, personal site, press).
- For a work email, confirm they work at that domain's company today.
- Keep job_title verbatim from their current profile. company_domain is the employer's root domain.
- Never guess a LinkedIn URL or a name. Leave a cell blank rather than guess.

Columns: input_email, name, job_title, company, company_domain, company_headcount, company_stage, linkedin_url, location, why_it_matters, source_urls.
```

Change: signup emails, one per line.

What the rules caught in our run:

- Tied each address to its person through a public profile at the same company, and confirmed they work there today.
- Noticed one founder's current title had moved to a new venture and returned the current role, not the one the email domain implied.
- Wrote a one-line why_it_matters for each signup from their role and the company's stage and size.

From testing: Gmail and other personal addresses work too: add the name from your signup form if you have it, since it is the biggest help in resolving a personal address.

Link: https://www.autumn.ai/resources/prompt-library#inbound-signups

### Find work emails without guessing

You need a work email you can defend, and a bounce or a wrong person costs more than a blank. (Sales)

Looks in: Company sites, Press releases, GitHub commits, Talk pages.

```text
Find the work email for each of these people. Keep exactly one row per person, in this order:
1. Karri Saarinen, Linear (linear.app)
2. Zach Lloyd, Warp (warp.dev)
3. Amjad Masad, Replit (replit.com)

Rules:
- Return an email only when (a) a public source shows this exact address (company site, press release, personal site, GitHub commit, talk or conference page), or (b) the company's email pattern is confirmed by a public address of another employee at the same domain, cited.
- Label how you know: email_basis is "published", "pattern-confirmed" or "not found".
- Never return an unconfirmed guess. Leave work_email blank instead.
- Confirm the person still works at that company today; if they don't, say where they are now in notes.

Columns: name, company, work_email, email_basis, evidence_url, notes.
```

Change: people, one per line.

What the rules caught in our run:

- Turned down an address that appeared only on a data vendor's page and left the cell blank.
- Noticed one person had left the company months earlier and named their new employer.
- Found a CEO's address on the company's older mail domain, not the product domain, from a post he wrote himself.

From testing: The email_basis column is what makes this usable: you can send to "published" right away and review "pattern-confirmed".

Link: https://www.autumn.ai/resources/prompt-library#work-email

### Find the buying committee at an account

You are working one account and need everyone who will weigh in, not just the top title. (Sales)

Looks in: LinkedIn, Company site, Talks and posts.

```text
Buying committee: who at Notion (notion.so) would evaluate and approve buying a sales prospecting and data enrichment tool? Find the current people in sales leadership, revenue operations, sales operations / GTM engineering, and growth.

Rules:
- Each person currently holds the role today, shown by a public source dated in the last 12 months.
- No former employees. Leave a person out rather than guess.

Columns: name, title, team, seniority, linkedin_url, likely_role_in_purchase, evidence_url.
```

Change: account; what you sell; teams.

What the rules caught in our run:

- Flagged one person as unclear because their profile address didn't match their name, instead of counting them.

Link: https://www.autumn.ai/resources/prompt-library#buying-committee

### Find companies hiring for a role

A new hire for a specific role means a team is about to buy, build or change something you care about. (Sales, Founders and market)

Looks in: Careers pages, Greenhouse, Lever, Ashby.

```text
Find 20 B2B software companies with 50 to 500 employees that posted a job for a GTM engineer, RevOps engineer or growth engineer in the last 30 days.

Rules:
- The job post is on the company's own careers page or ATS (Greenhouse, Lever, Ashby, Workable), dated within the window, and still open; cite it.
- Exclude staffing agencies and recruiters.

Columns: company, domain, headcount, job_title, posted_date, job_url, hiring_manager_if_named, why_it_matters.
```

Change: how many; company size; roles; days.

From testing: Asking for the post on the company's own careers page or ATS keeps out reposts on job boards and agency listings.

Link: https://www.autumn.ai/resources/prompt-library#hiring-signals

### Check for leadership changes at your accounts

A new leader in the seat you sell to is the best reason to reach out, and the worst one to miss. (Sales)

Looks in: Company posts, LinkedIn, Press.

```text
Executive change check. For each account, tell me who holds the CTO (or the most senior engineering leader if there is no CTO) role today, whether the same person held it six months ago, and if not, who they replaced and when. Keep one row per account, in this order:
linear.app
vercel.com
notion.so
figma.com

Rules:
- Current holder confirmed by a source dated in the last 12 months.
- Any change backed by an announcement or a dated profile change.
- Leave a cell blank rather than guess.

Columns: account, current_leader, current_title, in_role_since, changed_in_last_6_months, previous_leader, change_evidence_url, linkedin_url.
```

Change: role; accounts, one per line.

What the rules caught in our run:

- Left two seats blank where the previous leader had announced an exit but no successor was public, instead of naming the next most senior person.

From testing: Ask for in_role_since. A start date inside your window is the trigger; a long tenure tells you to look elsewhere for a reason to reach out.

Link: https://www.autumn.ai/resources/prompt-library#leader-change

## Founders and market

Competitors, new companies, lookalikes.

### Track everything a competitor publishes

You want one dated list of a competitor's launches, posts and events instead of checking five channels. (Founders and market)

Looks in: Blog and changelog, LinkedIn, X, Webinars.

```text
Track everything Clay (clay.com) published in the last 30 days: blog posts, changelog entries, product launches, case studies, webinars and posts from the company's LinkedIn and X accounts.

Rules:
- One row per item, dated inside the window, with a direct link.
- Skip reposts and job posts.

Columns: date, channel, title, type, one_line_summary, url.
```

Change: competitor; days.

From testing: Run it again on the same task each month and ask for what's new since the last run.

Link: https://www.autumn.ai/resources/prompt-library#competitor-tracking

### Find startups the week they incorporate

You want new companies before they announce: founders to back, hire or sell to first. (Founders and market, Investors, Sales)

Looks in: Delaware registry, Company sites, LinkedIn.

```text
Find 15 companies incorporated in Delaware between 2026-10-16 and 2026-10-30 that look like software or AI startups. For each one, resolve the company behind the filing: its live website or domain, what it does in one sentence, and its founders with their LinkedIn URLs and previous employer.

Rules:
- The entity appears in the Delaware registry with an incorporation date inside the window; cite the registry or a filing source.
- Only list a founder when a public source ties that person to this specific company.
- Leave a cell blank rather than guess. A shorter verified list beats a padded one.

Columns: entity_name, file_number, incorporation_date, domain, one_line_description, founders, founder_linkedin_urls, founder_previous_employers, source_urls.
```

Change: how many; from; to; what kind.

What the rules caught in our run:

- Left founders blank for companies whose site and profiles named no one, instead of attaching a lookalike name.

Link: https://www.autumn.ai/resources/prompt-library#new-delaware-companies

### Find lookalikes of a company

You have one great customer, competitor or portfolio company and want more like it. (Founders and market, Sales, Investors)

Looks in: Company sites, Funding announcements, Press.

```text
Lookalikes: find 15 companies most similar to Linear (linear.app).

First, write down what defines Linear (linear.app) from its own site: who buys it, how it sells (self-serve, free tier, sales-assisted), its product category and its stage. Then find companies that match on at least three of those, founded after 2018, 30 to 500 employees.

Rules:
- similarity_reason names the specific shared traits for that company with a fact from its own site or press, never a generic sentence reused across rows.
- Cite source_url for every row.
- Exclude the company itself and companies over 1,000 employees.

Columns: company, domain, one_line_description, founded, headcount, last_funding_round, shared_traits, similarity_reason, source_url.
```

Change: how many; company; filters.

From testing: Make it state the traits first. Our first version, without that step, gave every row the same generic reason.

Link: https://www.autumn.ai/resources/prompt-library#lookalikes

### Map a market

You need every company in a category with who funds them, in one table you can sort. (Founders and market, Investors)

Looks in: Company sites, Funding announcements, Filings.

```text
Market map: find every company building observability, evaluation or monitoring tooling for AI agents and LLM applications.

Rules:
- The company's own site shows a live product for this use (not a services firm or a feature inside a broad suite); cite it.
- Include funding only from an announcement or filing.

Columns: company, domain, one_line_description, segment, founded, founders, total_raised, last_round, lead_investors, latest_announcement, source_url.
```

Change: category.

What the rules caught in our run:

- Dropped companies whose sites showed a pivot, consulting work or no live product, and named which ones.
- Filled funding only where an announcement stated it, and marked the rest absent rather than estimating.

From testing: Give it segments to sort into. The segment column turns a long list into a map you can read.

Link: https://www.autumn.ai/resources/prompt-library#market-map

## Recruiting and experts

People who don't show up in the usual searches.

### Find students building in public

You want builders before they start a company or show up in recruiting searches. (Recruiting and experts, Investors)

Looks in: GitHub, Hackathon results, Papers, University pages.

```text
Find 15 university students (undergraduate or graduate, 18 or older) who are building notable software or AI projects in public but have not started a funded company yet.

Rules:
- Show the project with a link (GitHub repo, launched product, paper or demo) and a traction signal (stars, users, citations, press or a hackathon win).
- Confirm they are currently a student.
- Exclude anyone whose company has raised money.

Columns: name, university, program_and_year, project, project_url, traction_signal, github_url, x_or_linkedin_url, source_url.
```

Change: how many; field.

What the rules caught in our run:

- Marked "no funded company" as unclear on every row rather than claiming it, since no source can prove a negative.

From testing: Expect honest gaps here: a missing funding announcement isn't proof of no funding, so check the unclear cells before you reach out.

Link: https://www.autumn.ai/resources/prompt-library#students-building

### Find experts who have done the job

You need people for expert calls or advisory roles who have actually run the thing, not just written about it. (Recruiting and experts)

Looks in: LinkedIn, Talks and podcasts, Blog posts.

```text
Find 20 people who ran payments infrastructure (engineering or product) at a company processing over a billion dollars a year, are no longer in that role, and have written or spoken publicly about it.

Rules:
- Show the past role and company with a source.
- Link at least one talk, post, podcast or paper by them on the subject.
- Confirm their current role so we know they have moved on.

Columns: name, past_role, past_company, years_in_role, current_role, current_company, public_work_title, public_work_url, linkedin_url.
```

Change: how many; what they ran; at what scale.

From testing: Asking for public work does two jobs: it proves they did the work, and it gives you something specific to mention when you reach out.

Link: https://www.autumn.ai/resources/prompt-library#expert-sourcing

## Investors

Who is building, who is funding, who just left.

### Spot people leaving to start companies

You want to meet founders in the weeks after they leave, before the round is announced. (Investors, Recruiting and experts)

Looks in: LinkedIn, X, Personal sites, Incorporations.

```text
Find people who left OpenAI, Anthropic or Google DeepMind in the last 90 days and show signs of starting a company.

Rules:
- A dated public source (LinkedIn, X, personal site, press) shows they left within the window.
- At least one founding signal, cited: a new company or domain, an incorporation, a profile change to "stealth" or "founder", a funding announcement, or a co-founder pairing.
- Exclude people who simply joined another existing company.

Columns: name, left_company, role_there, left_date, founding_signal, new_company_or_stealth, co_founders, linkedin_url, evidence_url.
```

Change: companies; days.

What the rules caught in our run:

- Dropped people who left before the window, even when they were well known.
- Excluded people who joined someone else's startup as an early employee, since that isn't founding.
- Marked one departure unclear because the profile named the parent company, not the lab.

From testing: The founding-signal rule does the filtering. Without it you get everyone who changed jobs.

Link: https://www.autumn.ai/resources/prompt-library#lab-departures

### Rank who leads rounds in a category

You're raising, or tracking competition for deals, and need who actually leads, not who participates. (Investors, Founders and market)

Looks in: Funding announcements, Filings, Company posts.

```text
List the investors who led the most seed or Series A rounds in developer infrastructure and developer tools companies in the last 12 months.

Rules:
- Count only rounds where the investor is named as lead in the announcement, a filing or the company's own post; cite each round.
- Rank investors by number of led rounds.

Columns: investor, led_rounds_count, companies_led, notable_co_investors, partner_who_led_if_named, source_urls.
```

Change: rounds; category; months.

From testing: Counting only rounds where the announcement names a lead is what makes the ranking mean something; participants in a round don't count.

Link: https://www.autumn.ai/resources/prompt-library#lead-investors

## Risk and KYC

Check that a company is what it says.

### Verify a business

Onboarding a merchant, vendor or customer and you need each claim checked against a source. (Risk and KYC, Sales)

Looks in: State registries, Domain records, Company site, Founder profiles.

```text
Verify this business: Example Labs, Inc. (examplelabs.ai), Delaware file number 1234567.

Check each item against a public source and say whether it lines up: incorporation (state, date, entity type), registered agent, the people listed as founders or officers and whether their public profiles support those roles, operating address, domain registration date and age, whether the website describes a real product, and any public funding or press. Flag anything that does not line up or cannot be confirmed.

Rules:
- One row per check; status is "confirmed", "mismatch" or "not found".
- Never mark a check confirmed without a source.

Columns: check, finding, status, evidence_url.
```

Change: business.

What the rules caught in our run:

- Flagged that the legal name on the company's own site differed from the registry entity, instead of assuming they matched.
- Marked the registered agent and street address "not found" rather than filling them from a lookalike company.
- Confirmed both founders against the company's About page and their own profiles, and dated the domain from its registration record.

From testing: A mismatch isn't a verdict. It's the list of things to ask the business about before you approve it.

Link: https://www.autumn.ai/resources/prompt-library#verify-business