---
title: "Research workflows: from question to verified brief"
description: "Research is a process, not a prompt Whatever tools you use (Perplexity, ChatGPT or Gemini deep research, Claude Research, Gemini Notebook), professional…"
url: https://optimizeall.com/learn/gemini-copilot-perplexity-and-more/research-workflows-and-verification
updated: 2026-10-05
---

Gemini, Microsoft Copilot, Perplexity & the AI Tool Landscape · Perplexity and research engines · lesson 11 of 19 · 16 min

# Research workflows: from question to verified brief

## Research is a process, not a prompt

Whatever tools you use (Perplexity, ChatGPT or Gemini deep research, Claude Research, Gemini Notebook), professional research follows five steps:

1. **Frame:** define the decision, the questions and what "good enough" evidence looks like.
2. **Gather:** use answer engines and deep research to find candidate sources fast.
3. **Verify:** open, date-check and assess the sources behind important claims; log them.
4. **Synthesise:** write conclusions only from verified notes.
5. **Review and store:** peer review for high stakes; store sources and notes for reuse.

## Step 1: Frame

```text
Decision: Should we launch our creator programme in KSA or UAE first?
Questions: disclosure rules; platform usage; creator fee norms; payment and tax
considerations for paying creators.
Evidence standard: official regulator or government sources for rules; reputable
research or platform data for usage; at least two independent sources for fee norms.
Deadline and depth: 2-page brief by Thursday.
```

## Step 2: Gather

Run the questions in an answer engine or deep research tool. Ask for **primary sources** and **dates**, and for "what could not be found". Use two tools on the most important questions: disagreement shows you where to dig.

## Step 3: Verify with a log

| Claim | Source | Type | Date | Verified? | Notes |
|---|---|---|---|---|---|
| Influencer ads must be labelled | Regulator guidance page | Primary | [date] | Yes | Section 2 |
| Typical fee per post for micro-creators | Agency blog | Secondary, commercial | [date] | Partly | Only one source; range varies |

**Source hierarchy (general rule):** regulator or official source > original research or data > reputable journalism > industry reports (check commercial interest) > blogs and forums.

## Step 4: Synthesise only from verified notes

Give your verified log to an assistant and ask it to write the brief **only from the log**, citing the original sources (never "according to Perplexity" or "according to ChatGPT"). Mark uncertain items clearly.

```text
Using only the verified claims in the attached log, write a 2-page brief.
Cite the original source for each claim. Put partly-verified items in a
separate "Uncertain" section. Don't add new facts.
```

## Step 5: Review and store

- For high-stakes briefs, a colleague checks three random claims against sources.
- Store the brief, the log and the sources in a research Space, Project or Notebook so the next update starts from here.
- Note a "re-check by" date for anything that changes (rules, prices, platform policies).

## When tools disagree

Two tools giving different answers is useful information. Go to the primary source. If the primary source is ambiguous, say so in the brief rather than choosing the more convenient answer.

## Academic and student use

Only cite sources you have located and read. AI tools can suggest papers that do not exist or misdescribe real ones. Check titles, authors, year and the claim in the paper itself, and follow your institution's rules on AI use.

## Worked example: KSA vs UAE brief

A marketing lead in London runs the five steps. Two tools disagree on whether a licence is required for influencers in one market; the regulator's page clarifies the current requirement and its date. Fee norms appear only in agency blogs, so the brief presents them as indicative ranges from commercial sources. A colleague spot-checks three claims. The brief goes to the CMO with a sources appendix and a re-check date.

## Hands-on

Run the five-step workflow on a real question. Produce a one-page brief with a verification log of at least five claims.

## Choosing tools for each step

| Step | Good tools |
|---|---|
| Frame | Any assistant ("help me turn this decision into research questions") |
| Gather | Perplexity, deep research modes in ChatGPT, Gemini or Claude |
| Verify | Your browser, the primary sources, a verification log (sheet) |
| Synthesise | An assistant working only from your log; Gemini Notebook for source-bound drafts |
| Store | A Perplexity Space, a Claude or ChatGPT Project, or a shared Gemini Notebook |

No single tool does every step well; the workflow is what makes the output reliable.

## Pitfalls

- Skipping framing and researching everything.
- Citing the AI tool instead of the source.
- Treating vendor or agency statistics as neutral.
- No re-check date for fast-changing facts.

## How to measure success

Every claim in your brief is traceable to a verified source, uncertainty is visible, and colleagues can update the brief from your stored log.

## Video lecture: Research workflows: from question to verified brief

Lecture coming soon · 16 chapters · about 8 minutes. Read the full transcript below.

1. Research workflows
2. Why a process
3. Step 1: Frame
4. Step 2: Gather
5. Step 3: Verify with a log
6. Simple example
7. Step 4: Synthesise from verified notes only
8. Step 5: Review and store
9. Business example: KSA vs UAE brief from London
10. For students and academics
11. Common mistakes
12. Tools for each step
13. Key idea
14. Watch me do it, part 1
15. Watch me do it, part 2
16. Recap and try this now

## Lecture transcript

### Research workflows

AI research tools can gather in minutes what used to take days. But gathering is not the hard part. The hard part is producing a brief that is accurate, traceable and still useful next quarter. In this lecture you will learn a five step research workflow that works with any tool, and a verification log that makes your conclusions defensible.

### Why a process

Why a process rather than a clever prompt? Because the tool only gathers. Deciding what to trust, what to leave out and how to present uncertainty is your job. Think of a courtroom. Evidence is gathered quickly, but it only counts once it has been examined. The five steps are your examination, and they turn a pile of AI generated material into a brief you can stand behind.

### Step 1: Frame

Step one, frame. Write the decision the research supports, for example, should we launch our creator programme in Saudi Arabia or the UAE first. List the questions, disclosure rules, platform usage, creator fee norms, payment and tax considerations. Set an evidence standard, official sources for rules, at least two independent sources for fee norms. And set the deadline and depth, a two page brief by Thursday. Framing stops you researching everything and finishing nothing.

### Step 2: Gather

Step two, gather. Use an answer engine or deep research to find candidate sources fast. Ask for primary sources, dates and what could not be found. For the most important questions, use two different tools. If they disagree, you have just found exactly where to dig.

### Step 3: Verify with a log

Step three, verify, and write it down. Keep a log with the claim, the source, its type, primary or secondary, the date, whether you verified it, and notes. A general source hierarchy helps. Regulator or official sources first, then original research, then reputable journalism, then industry reports, checking for commercial interest, and finally blogs and forums.

### Simple example

A simple example of the log in action. You need the current UK adult minimum wage for a budget. The answer engine cites the official government page. You open it, confirm the figure and its effective date, and log it as primary and verified. Because rates change each year, you add a re check reminder for the next change date. One row, thirty seconds, and nobody will ever quote last year's rate from your budget.

### Step 4: Synthesise from verified notes only

Step four, synthesise, but only from your verified notes. Give the log to an assistant and ask it to write the brief using only those claims, citing the original source for each one, with partly verified items in a separate uncertain section, and no new facts. Never cite the AI tool itself as the source. Your reader needs the regulator, the survey or the newspaper, not the tool that found it.

### Step 5: Review and store

Step five, review and store. For high stakes briefs, have a colleague check three random claims against the sources. Store the brief, the log and the sources in a research Space, Project or Notebook, so the next update starts from here rather than from zero. And set re check dates for anything that changes, like rules, prices and platform policies.

### Business example: KSA vs UAE brief from London

A realistic example. A marketing lead in London runs the five steps on the Saudi versus UAE question. Two tools disagree on whether influencers need a licence in one market. The regulator's page clarifies the current requirement and its date. Fee norms appear only in agency blogs, so the brief presents them as indicative ranges from commercial sources. A colleague spot checks three claims, and the brief goes to the chief marketing officer with a sources appendix and a re check date. Two months later the regulator updated its guidance. Because the brief had a re check date and a verification log, the marketing lead knew exactly which paragraph and which slide depended on that rule. She updated both in under an hour and sent the CMO a short note explaining what changed and why it mattered.

### For students and academics

A note for students and academics. Only cite sources you have actually located and read. AI tools can suggest papers that do not exist, or misdescribe real ones. Check the title, authors, year and the claim in the paper itself, and follow your institution's rules on AI use. If you cannot find a suggested source, it does not go in your bibliography.

### Common mistakes

Four common mistakes. Skipping the framing and researching everything. Citing the AI tool rather than the original source. Treating vendor or agency statistics as neutral facts. And leaving fast changing facts without a re check date, so last year's rule quietly survives in this year's deck.

### Tools for each step

Which tools fit which step? Any assistant can help you frame, turning a decision into research questions. Answer engines and deep research modes gather. Verification is you, your browser and the primary sources, recorded in a simple sheet. An assistant working only from your log synthesises, or Gemini Notebook if you want the draft bound strictly to your sources. And a Perplexity Space, a Project or a shared Notebook stores it all. No single tool does every step well. The workflow is what makes the output reliable.

### Key idea

Keep one image in mind. The AI tool is the librarian, fast at finding books. You are the judge, deciding what the evidence proves. And your verification log is the evidence file, the thing that lets anyone check your conclusion later. When those three roles stay separate, AI makes your research faster without making it weaker.

### Watch me do it, part 1

Let me run the KSA versus UAE brief. I write the framing card. Decision, which market first for the creator programme. Four questions. Evidence standard, official sources for rules, two independent sources for fee norms. Deadline, Thursday. Then I gather. I run the question about influencer licensing in two different research tools. One says influencers need a licence in one market. The other does not mention it at all. That disagreement is my signal. I mark it dig here, and I go to the regulator's own website.

### Watch me do it, part 2

On the regulator's site, I find the current requirement and its date, and log it as primary and verified. For fee norms, I only find agency blogs, so those rows are marked partly verified, commercial source. Then I synthesise. I give an assistant the log and ask for a two page brief using only the verified claims, citing the original sources, with partly verified items in an uncertain section and no new facts. The brief arrives with the fee ranges clearly labelled indicative. A colleague checks three random claims, and it goes to the CMO with the log attached.

### Recap and try this now

Recap. Frame the decision and evidence standard, gather fast with AI, verify and log, synthesise only from verified notes citing original sources, and review and store with re check dates. Try this now. Take one real question from your work, run the five steps, and produce a one page brief with a verification log of at least five claims. You will be surprised how much more confident you feel presenting it.

## Key takeaways

- Research is a five-step process: frame, gather, verify, synthesise, review and store.
- Aim for primary sources and log each claim's source, type, date and verification status.
- Synthesise only from verified notes, and never cite the AI tool as the source.
- Use research spaces or Projects to keep ongoing topics organised; compare tools on important questions.

## Try it

Run the five-step workflow on a real question. Produce a one-page brief with a verification log of at least five claims.

- [Previous: Perplexity: the cited answer engine](https://optimizeall.com/learn/gemini-copilot-perplexity-and-more/perplexity-cited-answer-engine)
- [Next: Perplexity for teams: Spaces, Comet and the Sonar API](https://optimizeall.com/learn/gemini-copilot-perplexity-and-more/perplexity-api-comet-and-spaces)
- [All lessons of Gemini, Microsoft Copilot, Perplexity & the AI Tool Landscape](https://optimizeall.com/learn/gemini-copilot-perplexity-and-more)