Research and source verification

You design research that keeps claims traceable and treats search results as unreliable input until they have been verified.

After this chapterYou can define a research question, carry out a source plan, and produce a reproducible claim-source register.

Your progress0 of 40 lessons

4.1

Formulate a precise research question

Define population, period, concepts, comparison, and desired decision.

A broad question quickly produces a long but fuzzy text. Formulate exactly what you want to know, for whom, within which period, and what the answer will be used for. Define terms that sources may use differently.

Split the main question into sub-questions for facts, explanations, alternatives, and risks. Determine in advance what evidence is sufficient and which topics remain out of scope. Let ChatGPT signal missing definitions or unfeasible requirements before it searches.

Population and context
Period
Definitions
Decision goal
Evidence threshold

Terms in plain language

Scope: The boundaries of what you do and do not investigate.

How you can use this

Not 'What is the best home battery?', but a comparison for a specific consumption, grid type, budget, region, and decision date.

Try this prompt

Rewrite [broad question] as a research question. Define target group or population, period, region, key concepts, comparison, decision objective, sub-questions, exclusions, and evidence threshold. Do not start research yet.

Knowledge check

You are a beginner who wants to improve at presenting within six months while continuing to work. Which research question makes a useful course choice easiest to assess?

Which course fits my starting level, learning goal, available evenings and budget within these six months?Which course has the highest participant rating and offers the widest range of presentation techniques?Which popular course covers the most techniques and offers a well-known certificate for my CV?

Your practical exercise

Formulate one question in two versions and demonstrate why the sharper version is more testable.

What must be settled before testing a prompt to compare two versions fairly?

The same test cases, relevant context and assessment criteria; record prompt and source versions.
Expected behaviour is defined in advance; actual output and assessment are stored separately.

4.2

Design a strategy for finding sources

Link each type of claim to the most appropriate type of source.

Not every source answers the same question. Use laws and regulations from official publications, product behavior from current official documentation, statistics from the responsible authority, and practical experience as context, not as universal evidence.

Plan search terms, source categories, date limits, and exclusions. Web results are unreliable input: open the source, check the author, method, date, and interests. Multiple websites may repeat the same original source and then do not count as independent confirmation.

Claim type to source type
Primary source where applicable
Date and scope of application
Check independence

How you can use this

A price claim requires current supplier information; a yield claim requires method and comparable measurement data.

Try this prompt

Create a source plan for [research question]. For each sub-question, provide the claim type, desired primary source, acceptable secondary source, date limit, search terms, and exclusion criteria.

Your practical exercise

Create a source plan and find one example of source duplication.

Source for this lesson

OpenAI – Web search
Search and source links support verification; you still need to check whether the source supports the claim.
Checked: 2026-09-07

4.3

Extract and compare sources separately

First summarize sources individually before synthesizing similarities and conflicts.

Label each source with its title, owner, date, type and scope. Then extract its key claims, data used, method, limitations and exact location of the relevant passage. Only then compare the sources. This prevents differences from being smoothed over too early.

A comparison must show whether sources truly refer to the same population, period, and definition. When figures differ, do not calculate or explain the difference without sufficient information. Mark it as an open question and return to the source.

Source metadata
Claim and location
Method and limitation
Comparable definitions
Keep conflicts

How you can use this

Two labor market studies can report different percentages because the age limit, region, or measurement period differs.

Try this prompt

Analyze each source separately in a table with metadata, core claim, evidence, method, limitation, and exact location. Then compare definition, population, period, and outcome. Do not fill in differences yourself.

Knowledge check

Report A gives 30% participation among students; B gives 45% among association members. Both figures cover the same year. What comes before drawing a combined conclusion?

Calculate the difference of 15 percentage points and treat it as evidence that membership increases participation.Check who was surveyed, what participation means and how it was measured before comparing the percentages substantively.Calculate an average of 37.5% and add a footnote explaining that the reports have different target groups.

Your practical exercise

Compare three sources and check five extractions in the original documents.

4.4

Link claims to appropriate evidence

A reference is only useful when the source actually supports the specific claim.

Create a claim source table. Record the exact statement, the type of claim, the source passage, date, scope of application, and strength of evidence. Split compound sentences when one source only supports a part.

Check derived conclusions separately. A correct source can still be applied incorrectly. Ask for uncertainty language that matches the evidence: established, likely, possible, or unknown. Leave missing evidence visible instead of making a neat but unfounded sentence.

Exact claim
Appropriate passage
Scope
Derivation separate
Certainty proportional

Terms in plain language

Direct support: The source confirms the claim itself.
Inferred support: The conclusion follows from an additional calculation or interpretation that you must also check.

How you can use this

A study on large American companies does not automatically support a general Belgian SME claim.

Try this prompt

Create a claim-source table from [draft]. Split compound claims. For each claim, give the source passage, date, scope, direct or inferred support, level of certainty and required check. Do not silently remove unproven claims.

Knowledge check

A study measures time saved by 12 participants in one workshop. It does not measure costs. Your draft says: “This method saves everyone time and money.” Which revision is supported by the source?

“This method can save time and money; the time saved by twelve participants supports both benefits.”“Participants in workshops like these save time and money; the study of twelve participants provides an example.”“Time savings were measured for these twelve participants; money savings and applicability to other groups have not been demonstrated.”

Your practical exercise

Check ten claims from an existing text and adjust the wording of certainty where needed.

4.5

Building a reproducible research log

Record question, search path, selected sources, extractions, decisions, and changes.

Reproducibility means that someone else can understand how you went from question to conclusion. Keep the search date, used search terms, inclusion and exclusion reasons, source versions, claim source table, and open uncertainties. The log does not need to contain every click, but it does need every decision that materially influenced the outcome.

Also manage the risk of information becoming outdated. Give claims that may change a check date and a trigger for rechecking, such as a change in the law, a new financial year or a product release. Finalise a report only after the source list and the main checks are complete.

Search date and terms
Selection decisions
Source versions
Open uncertainties
Recheck trigger

How you can use this

A software comparison gets a recheck trigger upon price change or major release; an annual figure upon the new reporting year.

Try this prompt

Create a reproducible log for [research] with question version, search terms, source selection, extractions, claim status, changes, control date, and recheck trigger. Indicate what has not been investigated.

Knowledge check

You exclude a recent article because it studies adults, while your question concerns only young people. What do you record so a reviewer can reconstruct this selection?

The source identity, the version used and its exclusion because the population falls outside the predefined scope.The title, publication date and a note that the article was not included in the final comparison.The search term, search date and article location, with the final source list as the decision record.

Your practical exercise

Carry out a small investigation using public sources and, if possible, have someone reconstruct a key claim through your log. Working alone: close your summary, reopen the source through your log and reconstruct the claim in a separate round. If a topic of your own does not work out, repeat the real source-checking exercise from Beginner. Done: the source, date, supporting passage and scope can be found; state that a solo rereading is not an independent review.

Worked example

Keep a source conflict visible until it is resolved

Fictional practice material; incorrect answers have been created deliberately for this exercise.

An entirely fictional research case with no real websites. You are looking for the registration deadline for an evening course. Practice source A, a course page dated 3 September: “Register by 20 September; adults; starts 1 October.” Practice source B, a registration form dated 5 September: “Evening course for adults; register by 18 September; starts 1 October.” Neither source explains the difference.

Input

Extract the information from both sources separately. Compare the target audience, course, start date and registration deadline. Then write a brief practical conclusion. Do not automatically treat source B as authoritative because of its later date.

First practice answer

Registration is open until 20 September; the course starts on 1 October. This deliberately flawed answer chooses the later deadline and conceals source B.

Check

Source register: A says 20 September; B says 18 September. The target audience and start date appear to match. The claim “starts 1 October” has direct support from both practice sources. The claim “register by 20 September” has conflicting evidence. Ask the organiser which deadline applies to this course and keep the confirmation. In actual research, open the real sources and record the exact location and check date.

Improved version

Both fictional sources give 1 October as the start date. The registration deadline is not yet clearly confirmed: A says 20 September, B says 18 September. Contact the organiser before 18 September to confirm the correct deadline. This is a cautious practical next step, not a finding that 18 September has legal or administrative precedence.

Try it yourself

A third fictional source turns out to be a copy of A. Does it provide independent confirmation of 20 September?

View the model answer

No. These are two mentions of the same original information. The conflict with B remains. Record the relationship between the original and derivative source and wait for appropriate confirmation from the organiser.

Chapter assignment

Bring everything together

Submit a research file with a defined question, source plan, three source extracts, comparison, ten claim-source links, uncertainties, and recheck triggers.

My notes

Maximum 10,000 characters per note.

Progress and notes are stored only in this browser on this device. Do not enter sensitive data. Download your notes regularly. This course sets no automatic expiry date. You can delete the data through your browser’s site-data settings; export anything you wish to keep first. Browser settings or cleanup may erase it earlier. These local notes are not sent to Finaudax.