Skip to content

afeezadebayo/research-learning-skills / ai-assisted-research

ai-assisted-research

Verified

Use AI research tools well: choosing between deep research modes, AI search, assistants and browsing agents, writing research briefs with preferred sources and evidence standards, verifying citations, filling gaps, resolving conflicts, protecting confidential data and writing a human-checked synthesis. Use whenever someone uses deep research, AI search or an AI agent to research a topic, wants better research prompts, needs to verify AI-cited sources or turn an AI research report into a reliable summary.

AI-Assisted Research

AI research tools such as deep research features in assistants, AI search engines and agents that browse can read dozens of sources in minutes and produce long, polished reports. They are useful for getting oriented quickly, finding sources and drafting syntheses. They also make specific mistakes: citing sources that do not support the claim, mixing outdated and current information, presenting weak sources confidently, missing paywalled or non-English material, and inventing details when evidence is thin. The skill is not in pressing the button; it is in briefing the tool well and verifying what comes back.

This skill uses AI for research effectively and safely: choosing the right tool, writing a strong research brief, steering source quality, asking for structured outputs with citations, verifying claims, filling gaps with targeted searches and combining AI findings with human judgement.

Operating principles

  1. Brief like a research lead. Question, scope, sources to prefer and avoid, output format and evidence standards.
  2. Citations are leads, not proof. Open sources for every key claim.
  3. Ask for uncertainty. Request confidence levels, conflicts and gaps, not just answers.
  4. Prefer primary sources. Tell the tool which source types count.
  5. Iterate. Follow up on gaps and weak sections with targeted questions.
  6. Protect confidential data. Do not paste sensitive information into tools that are not approved for it.

Tool types

ToolStrengthsWatch for
Deep research modes in assistantsLong multi-source reports, planning stepsOver-confident synthesis; check citations
AI search enginesFast answers with linksShallow coverage; SEO-heavy sources
Assistant with web searchConversational follow-upVariable source quality
Agents with browsing (for example Claude Code or computer-use tools)Systematic collection into tablesNeed clear instructions and limits
Academic AI toolsFinding and summarising papersCheck papers exist and say what is claimed

When to use, and when not to

Use it to get oriented on a new topic, build a source list, draft literature or market overviews, compare options, collect structured data from many pages and prepare for interviews.

Don't rely on it alone for high-stakes facts, legal, medical or financial advice, or claims you will publish without verification.

Inputs

  • The research plan or question.
  • Preferred and excluded source types.
  • Regions, dates and languages.
  • Output format and evidence standards.
  • Any internal documents the tool may use (if approved).

Workflow

Step 1. Write the brief

Use the template below.

Step 2. Choose the tool

Quick orientation: AI search. Comprehensive overview: deep research mode. Structured collection: agent with browsing.

Step 3. Run and skim

Read the summary and structure; note claims that matter most for the decision.

Step 4. Verify key claims

Open the cited sources; confirm wording, numbers and dates; check source quality.

Step 5. Fill gaps

Ask targeted follow-ups or run your own searches for weak or missing areas.

Step 6. Cross-check conflicts

Where sources disagree, find the original data and explain the difference.

Step 7. Write your own synthesis

Use the AI draft as raw material; your final output states what is verified, with confidence labels.

Research brief template

Question: <specific question and the decision it informs>
Scope: <regions, time period, segments>
Prefer sources: <official statistics, filings, peer-reviewed studies, reputable press>
Avoid: <vendor blogs, undated pages, forums unless for customer language>
Output: <structure, e.g. summary, findings by sub-question, table, sources>
For each finding: cite the source with date, give a confidence level and note conflicts.
List what you could not find or verify.

Verification checklist

  • Each key number traced to a source that states it.
  • Source dates recorded; outdated data flagged.
  • Source types match the brief (primary where needed).
  • Quotes match originals.
  • Conflicts explained.
  • Gaps listed rather than filled with guesses.

Prompts for better AI research

  • "Before answering, list the sub-questions you will research."
  • "Use official statistics and peer-reviewed studies first; mark any vendor or blog sources."
  • "For each claim, give the exact sentence from the source it comes from."
  • "What evidence would contradict this conclusion? Search for it."
  • "List the three weakest parts of this report."

Structured collection with agents

For tasks like comparing 40 competitors' pricing or collecting regulations across 10 countries, give an agent a table schema, the list of targets and rules (official pages only, record URL and date, leave blank if not found). Review a sample of rows against the sources before trusting the full table. Structured collection is where AI saves the most time, because the output is easy to check row by row.

Common AI research errors

ErrorHow to spot it
Fabricated or mismatched citationsThe linked page does not contain the claim
Outdated factsSource dates are old or missing
Blended figuresNumbers from different years or definitions combined
Over-generalisationOne study presented as consensus
Missing counter-evidenceNo section on limitations or opposing views

Documenting AI use

Note in the research log which tool was used, the brief, the date and which claims were verified by a person. This makes the work reproducible and shows readers how much to trust each part. Where your organisation or publisher requires it, disclose AI assistance in the final output.

Combining AI with human sources

Use AI findings to prepare sharper questions for experts and customers, then let interviews confirm or challenge them. The combination is faster than either alone and catches errors that both might miss.

Re-verify AI findings before reuse.

Keep briefs reusable.

Edge cases and judgement calls

SituationWhat to do
Citation does not support the claimRemove or correct; check other claims from the same output
Topic changes fastRestrict dates; verify with the newest primary sources
Non-English marketsAsk for local-language sources; verify translations
Paywalled researchUse abstracts, press releases or library access; mark as partially verified
Confidential contextUse approved enterprise tools or remove sensitive details
Output too longAsk for a one-page summary with the evidence table

Handoffs to related skills

  • Planning: research-planning
  • Source credibility: source-evaluation
  • Claim checking: fact-checking
  • Manual search: desk-research
  • Academic reviews: literature-review
  • Synthesis: research-synthesis

Worked example

A strategy lead used a deep research tool to map the African payments market. The first report was 20 pages with 60 citations. Verification found that six key numbers came from vendor blogs, two citations did not mention the claimed figures and one market estimate was five years old. Follow-up prompts restricted sources to central banks, regulators and company filings; the lead checked the ten most important claims personally and wrote a three-page synthesis with confidence labels. Leadership used it to choose two markets for deeper study.

Common mistakes

  • Copying AI reports into decks without checking.
  • Vague briefs that let the tool choose weak sources.
  • Not asking for gaps and uncertainty.
  • Trusting the number of citations as a sign of quality.
  • Pasting confidential data into unapproved tools.

Output format

1. Research brief
2. Tool used and settings
3. Key claims with verification results
4. Gaps and follow-up findings
5. Final synthesis with confidence labels

Self-check before you answer

  • Was the tool briefed with sources, scope and output format?
  • Were key claims verified in the original sources?
  • Are gaps and conflicts stated?
  • Is the final synthesis written and checked by a person?

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/ai-assisted-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.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall