Research and Learning skill 02 of 13

Desk Research SKILL.md

The Desk Research skill makes Claude, Codex or another AI agent find and use existing information well. It turns sub-questions into search strategies, uses operators and specialist databases, finds primary sources, logs every fact with its source and date, traces citations and triangulates findings.

desk-research/SKILL.md, version 1.0.0, 1,506 words.

Desk research means finding and using information that already exists: reports, statistics, company filings, regulations, academic papers, news, websites and databases. It is the fastest and cheapest research method, yet many people search badly: one vague query, the first few results, no record of sources and conclusions drawn from SEO-driven blog posts. Skilled researchers search more precisely, go to primary sources, use specialist databases, read laterally to check credibility and keep a clean record of everything they find.

This skill runs effective desk research: turning questions into search strategies, using search operators and specialist sources, finding primary data, reading efficiently, recording sources and notes, triangulating findings and summarising what is known, unknown and uncertain.

Operating principles

  1. Start from the question. Search for answers to sub-questions, not for the topic in general.
  2. Go to the source. Prefer original data, filings, laws and studies over articles about them.
  3. Search precisely. Use operators, specific terms and specialist databases.
  4. Record as you go. Source, date, key fact and page or quote for everything you might use.
  5. Triangulate. Important facts need more than one independent source.
  6. Know when to stop. Stop when sub-questions are answered to the agreed standard.

Source types

TypeExamplesGood for
Official statisticsNational statistics offices, World Bank, OECD, central banksPopulation, economy, sector data
Regulation and governmentLaws, regulators' guidance, consultationsRules, compliance, policy changes
Company sourcesAnnual reports, investor presentations, filings, pricing pagesStrategy, financials, products
Research and academiaJournals, working papers, university reportsEvidence, methods, effects
Industry reportsAnalysts, trade associationsMarket sizing, trends (check methods)
News and trade pressReputable outlets, specialist publicationsRecent events, interviews
Communities and reviewsForums, review sites, social mediaCustomer language, complaints (biased samples)
DatasetsOpen data portals, APIs, your own dataAnalysis and verification

Search operators (Google and many engines)

OperatorUseExample
"exact phrase"Exact wording"net revenue retention"
site:Search one site or domain typesite:gov.uk apprenticeship levy
filetype:Find documentsfiletype:pdf annual report 2025 fintech
- (minus)Exclude wordsjaguar -car
ORAlternatives(survey OR study) remote work productivity
intitle:Words in the titleintitle:"market report" edtech Nigeria
before: / after:Date limits (where supported)data privacy fines after:2025-01-01

Check current operator support, as search engines change features over time.

When to use, and when not to

Use it for market and competitor research, background before interviews, regulations, statistics, best practices and fact-finding for reports and decisions.

Don't use it when the answer does not exist in published sources (use interviews, surveys or experiments) or as a substitute for legal or financial advice.

Inputs

  • The research plan and sub-questions.
  • Known sources and databases available (including paid ones).
  • Regions, languages and time periods in scope.
  • Evidence standards and deadline.

Raw SKILL.md file

.md

This is the exact file agents read, including the credit header and licence notice. Save it as desk-research/SKILL.md.

---
name: desk-research
description: "Run effective desk research: search terms from sub-questions, source types from official statistics to filings and studies, search operators, primary sources, skimming and reading, lateral credibility checks, a research log, tracing citations, triangulation and summaries of what is known, uncertain and unknown. Use whenever someone needs to research a topic online, find statistics or market data, search for regulations or company information, trace a statistic to its source or build a research log."
license: CC-BY-4.0
metadata:
  author: Afeez Adebayo
  author_url: https://afeezadebayo.com
  author_linkedin: https://www.linkedin.com/in/afeez-adebayo/
  author_x: https://x.com/za_icon
  source: https://afeezadebayo.com/skills-md/productivity/research-learning/desk-research/
  category: productivity
  subcategory: research-learning
  version: "1.0.0"
  created: "2026-10-10"
  updated: "2026-10-14"
---

# Desk Research

Desk research means finding and using information that already exists: reports, statistics, company filings, regulations, academic papers, news, websites and databases. It is the fastest and cheapest research method, yet many people search badly: one vague query, the first few results, no record of sources and conclusions drawn from SEO-driven blog posts. Skilled researchers search more precisely, go to primary sources, use specialist databases, read laterally to check credibility and keep a clean record of everything they find.

This skill runs effective desk research: turning questions into search strategies, using search operators and specialist sources, finding primary data, reading efficiently, recording sources and notes, triangulating findings and summarising what is known, unknown and uncertain.

## Operating principles
1. **Start from the question.** Search for answers to sub-questions, not for the topic in general.
2. **Go to the source.** Prefer original data, filings, laws and studies over articles about them.
3. **Search precisely.** Use operators, specific terms and specialist databases.
4. **Record as you go.** Source, date, key fact and page or quote for everything you might use.
5. **Triangulate.** Important facts need more than one independent source.
6. **Know when to stop.** Stop when sub-questions are answered to the agreed standard.

## Source types
| Type | Examples | Good for |
|---|---|---|
| Official statistics | National statistics offices, World Bank, OECD, central banks | Population, economy, sector data |
| Regulation and government | Laws, regulators' guidance, consultations | Rules, compliance, policy changes |
| Company sources | Annual reports, investor presentations, filings, pricing pages | Strategy, financials, products |
| Research and academia | Journals, working papers, university reports | Evidence, methods, effects |
| Industry reports | Analysts, trade associations | Market sizing, trends (check methods) |
| News and trade press | Reputable outlets, specialist publications | Recent events, interviews |
| Communities and reviews | Forums, review sites, social media | Customer language, complaints (biased samples) |
| Datasets | Open data portals, APIs, your own data | Analysis and verification |

## Search operators (Google and many engines)
| Operator | Use | Example |
|---|---|---|
| "exact phrase" | Exact wording | "net revenue retention" |
| site: | Search one site or domain type | site:gov.uk apprenticeship levy |
| filetype: | Find documents | filetype:pdf annual report 2025 fintech |
| - (minus) | Exclude words | jaguar -car |
| OR | Alternatives | (survey OR study) remote work productivity |
| intitle: | Words in the title | intitle:"market report" edtech Nigeria |
| before: / after: | Date limits (where supported) | data privacy fines after:2025-01-01 |
Check current operator support, as search engines change features over time.

## When to use, and when not to
Use it for market and competitor research, background before interviews, regulations, statistics, best practices and fact-finding for reports and decisions.

Don't use it when the answer does not exist in published sources (use interviews, surveys or experiments) or as a substitute for legal or financial advice.

## Inputs
- The research plan and sub-questions.
- Known sources and databases available (including paid ones).
- Regions, languages and time periods in scope.
- Evidence standards and deadline.

## Workflow

### Step 1. Turn sub-questions into search terms
List key terms, synonyms, technical terms, acronyms and the words experts use.

### Step 2. Choose source types
For each sub-question, which types of source are most likely to hold the answer?

### Step 3. Search systematically
Start broad to learn the vocabulary, then narrow with operators and specialist sources. Search official and primary sources directly.

### Step 4. Skim, then read
Skim titles, summaries and headings; read in full only sources likely to answer a sub-question.

### Step 5. Check credibility
Read laterally: what do other sources say about this source? See `source-evaluation`.

### Step 6. Record findings
One line per fact: claim, number, source, date, link, page, confidence.

### Step 7. Follow citations
Trace statistics back to their original source; use reference lists to find better sources.

### Step 8. Triangulate
Compare sources on key facts; note agreements, conflicts and gaps.

### Step 9. Summarise
For each sub-question: what we know, how confident we are, what conflicts and what is still unknown.

## Research log template
```
| # | Sub-question | Claim or data | Source | Date | Link | Page or quote | Type | Confidence |
| 1 | Market size | 4,200 secondary schools in Lagos | State education ministry | 2024 | ... | p. 12 | Primary | High |
| 2 | Market size | 38% use paid digital tools | Industry survey | 2025 | ... | Table 3 | Secondary, n=300 | Medium |
```

## Finding primary sources
- Search the publisher of a statistic, not the article quoting it.
- Look for "methodology", "technical notes" or "about this data" pages.
- Use regulators' and statistics offices' own search tools.
- For companies, use investor relations pages and official filings.
- For research, use Google Scholar, PubMed, SSRN or arXiv, and check whether a paper is peer-reviewed or a preprint.

## Specialist databases worth knowing
| Need | Where to look |
|---|---|
| Company filings (UK) | Companies House |
| Company filings (US) | SEC EDGAR |
| Academic research | Google Scholar, PubMed, SSRN, arXiv, Semantic Scholar |
| Global statistics | World Bank Open Data, OECD, UN data portals |
| Patents | Google Patents, national patent offices |
| Web archives | Internet Archive Wayback Machine for past versions of pages |
| Search trends | Google Trends for relative interest over time |

## Using AI search carefully
AI search and research tools can speed up finding sources, but they can also cite pages that do not say what is claimed. Open every cited source for key facts, check dates and record the original source rather than the AI summary. See `ai-assisted-research`.

Save the log beside the final report.

## Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Data is old | Use it with the date stated; look for newer proxies |
| Sources conflict | Compare methods and dates; report both with reasons |
| Non-English sources | Search in local languages; translate carefully; note translation |
| Paywalled reports | Look for executive summaries, press releases, author talks or library access |
| Only blog posts available | Treat as low confidence; trace claims back or seek other methods |
| Too much information | Return to sub-questions; stop collecting what does not answer them |

## Handoffs to related skills
- Planning: `research-planning`
- Credibility checks: `source-evaluation`
- Checking specific claims: `fact-checking`
- AI research tools: `ai-assisted-research`
- Academic sources: `literature-review`
- Synthesis: `research-synthesis`

## Worked example
A marketing manager needed the size of the digital learning market in Lagos secondary schools. Generic searches returned blog posts quoting a figure with no source. Using the skill, she searched the state education ministry and national statistics sites for school counts, found an industry survey with a disclosed sample for adoption rates, used company pricing pages for typical spend and traced the blog figure back to a 2019 report that measured something different. The log recorded each source with dates and confidence, and the final estimate gave a range with clear assumptions instead of a single unsupported number.

## Common mistakes
- One vague search and the first page of results.
- Quoting statistics without finding their origin.
- Mixing old and new data without dates.
- No research log.
- Collecting interesting facts that do not answer the question.

## Output format
```
1. Search strategy per sub-question
2. Research log
3. Findings by sub-question with confidence
4. Conflicts and gaps
5. Sources list
```

## Self-check before you answer
- Were searches tied to sub-questions?
- Were primary sources found for key facts?
- Is every claim logged with source and date?
- Are confidence, conflicts and gaps stated?

---
Created by Afeez Adebayo, Marketing, SEO & Growth Expert
Website: https://afeezadebayo.com
LinkedIn: https://www.linkedin.com/in/afeez-adebayo/
X: https://x.com/za_icon
Source: https://afeezadebayo.com/skills-md/productivity/research-learning/desk-research/

Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and adapt this skill, including for commercial use, as long as you credit Afeez Adebayo and link to the source URL above. Keep this notice in any copy or adapted version.

Workflow

Step 1. Turn sub-questions into search terms

List key terms, synonyms, technical terms, acronyms and the words experts use.

Step 2. Choose source types

For each sub-question, which types of source are most likely to hold the answer?

Step 3. Search systematically

Start broad to learn the vocabulary, then narrow with operators and specialist sources. Search official and primary sources directly.

Step 4. Skim, then read

Skim titles, summaries and headings; read in full only sources likely to answer a sub-question.

Step 5. Check credibility

Read laterally: what do other sources say about this source? See source-evaluation.

Step 6. Record findings

One line per fact: claim, number, source, date, link, page, confidence.

Step 7. Follow citations

Trace statistics back to their original source; use reference lists to find better sources.

Step 8. Triangulate

Compare sources on key facts; note agreements, conflicts and gaps.

Step 9. Summarise

For each sub-question: what we know, how confident we are, what conflicts and what is still unknown.

Research log template

| # | Sub-question | Claim or data | Source | Date | Link | Page or quote | Type | Confidence |
| 1 | Market size | 4,200 secondary schools in Lagos | State education ministry | 2024 | ... | p. 12 | Primary | High |
| 2 | Market size | 38% use paid digital tools | Industry survey | 2025 | ... | Table 3 | Secondary, n=300 | Medium |

Finding primary sources

  • Search the publisher of a statistic, not the article quoting it.
  • Look for "methodology", "technical notes" or "about this data" pages.
  • Use regulators' and statistics offices' own search tools.
  • For companies, use investor relations pages and official filings.
  • For research, use Google Scholar, PubMed, SSRN or arXiv, and check whether a paper is peer-reviewed or a preprint.

Specialist databases worth knowing

NeedWhere to look
Company filings (UK)Companies House
Company filings (US)SEC EDGAR
Academic researchGoogle Scholar, PubMed, SSRN, arXiv, Semantic Scholar
Global statisticsWorld Bank Open Data, OECD, UN data portals
PatentsGoogle Patents, national patent offices
Web archivesInternet Archive Wayback Machine for past versions of pages
Search trendsGoogle Trends for relative interest over time

Using AI search carefully

AI search and research tools can speed up finding sources, but they can also cite pages that do not say what is claimed. Open every cited source for key facts, check dates and record the original source rather than the AI summary. See ai-assisted-research.

Save the log beside the final report.

Edge cases and judgement calls

SituationWhat to do
Data is oldUse it with the date stated; look for newer proxies
Sources conflictCompare methods and dates; report both with reasons
Non-English sourcesSearch in local languages; translate carefully; note translation
Paywalled reportsLook for executive summaries, press releases, author talks or library access
Only blog posts availableTreat as low confidence; trace claims back or seek other methods
Too much informationReturn to sub-questions; stop collecting what does not answer them
  • Planning: research-planning
  • Credibility checks: source-evaluation
  • Checking specific claims: fact-checking
  • AI research tools: ai-assisted-research
  • Academic sources: literature-review
  • Synthesis: research-synthesis

Worked example

A marketing manager needed the size of the digital learning market in Lagos secondary schools. Generic searches returned blog posts quoting a figure with no source. Using the skill, she searched the state education ministry and national statistics sites for school counts, found an industry survey with a disclosed sample for adoption rates, used company pricing pages for typical spend and traced the blog figure back to a 2019 report that measured something different. The log recorded each source with dates and confidence, and the final estimate gave a range with clear assumptions instead of a single unsupported number.

Common mistakes

  • One vague search and the first page of results.
  • Quoting statistics without finding their origin.
  • Mixing old and new data without dates.
  • No research log.
  • Collecting interesting facts that do not answer the question.

Output format

1. Search strategy per sub-question
2. Research log
3. Findings by sub-question with confidence
4. Conflicts and gaps
5. Sources list

Self-check before you answer

  • Were searches tied to sub-questions?
  • Were primary sources found for key facts?
  • Is every claim logged with source and date?
  • Are confidence, conflicts and gaps stated?

When to use this skill, and when not to

Use it for market and competitor research, regulations, statistics, best practices and fact-finding for reports.

One vague search and the first page of results lead to unsupported claims; precise searching and primary sources fix that.

Example prompts

  • “Find the original source of this statistic.”
  • “Research the market size for edtech in Lagos.”
  • “Build a research log for this project.”

Example output

The skill returns a fixed structure. A shortened example for an imaginary site:

Searches: per sub-question with operators
Primary sources: ministry data, filings, surveys with methods
Traced: a blog figure back to an old, different report
Output: range estimate with assumptions and confidence

Every fact is logged with its source and date, and important facts come from primary sources.

What the agent needs from you

It needs the research plan, available databases, regions, languages, periods and evidence standards.

How this skill hands off to related skills

How to run Desk Research step by step

  1. Install the skill as desk-research/SKILL.md.
  2. Turn sub-questions into search terms.
  3. Choose source types.
  4. Search systematically.
  5. Skim, then read.
  6. Check credibility.

Situations this skill handles

Desk Research handles these situations:

SituationWhat the skill does
Data is oldUse it with the date stated; look for newer proxies
Sources conflictCompare methods and dates; report both with reasons
Non-English sourcesSearch in local languages; translate carefully; note translation
Paywalled reportsLook for executive summaries, press releases, author talks or library access
Only blog posts availableTreat as low confidence; trace claims back or seek other methods
Too much informationReturn to sub-questions; stop collecting what does not answer them

What makes this skill different

Most online research stops at the first results. This skill searches precisely, goes to primary sources and records everything so findings can be checked.

Practical tips from real projects

  • Start from the question.
  • Go to the source.
  • Search precisely.
  • Record as you go.

A worked example: A marketing manager needed the size of the digital learning market in Lagos secondary schools. Generic searches returned blog posts quoting a figure with no source. Using the skill, she searched the state education ministry and national statistics sites for school counts, found an industry survey with a disclosed sample for adoption rates, used company pricing pages for typical spend and traced the blog figure back to a 2019 report that measured something different. The log recorded each source with dates and confidence, and the final estimate gave a range with clear assumptions instead of a single unsupported number.

Turn sub-questions into search terms: List key terms, synonyms, technical terms, acronyms and the words experts use. Choose source types: For each sub-question, which types of source are most likely to hold the answer? Search systematically: Start broad to learn the vocabulary, then narrow with operators and specialist sources. Search official and primary sources directly.

Skim, then read: Skim titles, summaries and headings; read in full only sources likely to answer a sub-question. Check credibility: Read laterally: what do other sources say about this source? See source-evaluation. Record findings: One line per fact: claim, number, source, date, link, page, confidence.

Follow citations: Trace statistics back to their original source; use reference lists to find better sources. Triangulate: Compare sources on key facts; note agreements, conflicts and gaps. Summarise: For each sub-question: what we know, how confident we are, what conflicts and what is still unknown.

Source types. Official statistics: examples National statistics offices, World Bank, OECD, central banks; good for Population, economy, sector data. Regulation and government: examples Laws, regulators' guidance, consultations; good for Rules, compliance, policy changes. Company sources: examples Annual reports, investor presentations, filings, pricing pages; good for Strategy, financials, products. Research and academia: examples Journals, working papers, university reports; good for Evidence, methods, effects. Industry reports: examples Analysts, trade associations; good for Market sizing, trends (check methods). News and trade press: examples Reputable outlets, specialist publications; good for Recent events, interviews. Communities and reviews: examples Forums, review sites, social media; good for Customer language, complaints (biased samples). Datasets: examples Open data portals, APIs, your own data; good for Analysis and verification.

Search operators (Google and many engines). "exact phrase": use Exact wording; example "net revenue retention". site:: use Search one site or domain type; example site:gov.uk apprenticeship levy. filetype:: use Find documents; example filetype:pdf annual report 2025 fintech. OR: use Alternatives; example (survey OR study) remote work productivity. intitle:: use Words in the title; example intitle:"market report" edtech Nigeria. before: / after:: use Date limits (where supported); example data privacy fines after:2025-01-01.

Research log template. 1: sub-question Market size; claim or data 4,200 secondary schools in Lagos; source State education ministry; date 2024; link …; page or quote p. 12; type Primary; confidence High. 2: sub-question Market size; claim or data 38% use paid digital tools; source Industry survey; date 2025; link …; page or quote Table 3; type Secondary, n=300; confidence Medium.

Finding primary sources. Search the publisher of a statistic, not the article quoting it. Look for "methodology", "technical notes" or "about this data" pages. Use regulators' and statistics offices' own search tools. For companies, use investor relations pages and official filings. For research, use Google Scholar, PubMed, SSRN or arXiv, and check whether a paper is peer-reviewed or a preprint.

Specialist databases worth knowing. Company filings (UK): where to look Companies House. Company filings (US): where to look SEC EDGAR. Academic research: where to look Google Scholar, PubMed, SSRN, arXiv, Semantic Scholar. Global statistics: where to look World Bank Open Data, OECD, UN data portals. Patents: where to look Google Patents, national patent offices. Web archives: where to look Internet Archive Wayback Machine for past versions of pages. Search trends: where to look Google Trends for relative interest over time.

Using AI search carefully. AI search and research tools can speed up finding sources, but they can also cite pages that do not say what is claimed. Open every cited source for key facts, check dates and record the original source rather than the AI summary. See ai-assisted-research. Save the log beside the final report.

Edge cases and judgement calls. Data is old: what to do Use it with the date stated; look for newer proxies. Sources conflict: what to do Compare methods and dates; report both with reasons. Non-English sources: what to do Search in local languages; translate carefully; note translation. Paywalled reports: what to do Look for executive summaries, press releases, author talks or library access. Only blog posts available: what to do Treat as low confidence; trace claims back or seek other methods. Too much information: what to do Return to sub-questions; stop collecting what does not answer them.

Frequently asked questions

Finding and using information that already exists, such as statistics, filings, studies and reports.

Search the publisher of a statistic, official statistics sites, regulators and company filings.

Exact phrases, site:, filetype:, minus and OR.

Changelog

  • 1.0.0, 10 October 2026: first release.

Use, adapt and share this skill under CC BY 4.0: credit Afeez Adebayo and link to this page.