Week 12: Source Criticism with SIFT

Welcome to Week 12 of the AI Literacy Course.

In Lesson 11, we learned how to identify checkable claims, recognise different AI errors, prioritise risk, and decide which claims require verification.

This week, we develop that work through SIFT:

• S — Stop

• I — Investigate the source

• F — Find better coverage

• T — Trace the claim to its original context

SIFT helps us move from a confident impression to a traceable judgement.

The central principle is:

Credibility comes from a reproducible source trail, not a confident impression.

Learning goals

By the end of this lesson, learners should be able to:

• Explain the four SIFT moves

• Stop before reacting, sharing, or acting

• Investigate whether a source is suitable for a particular claim

• Use lateral reading to learn about a source

• Find more relevant and responsible coverage

• Trace a claim, quotation, statistic, or image to its original context

• Match evidence and source type to the claim

• Evaluate date, place, method, and applicability

• Document a source trail and communicate remaining uncertainty

1. SIFT is a flexible investigation method

SIFT is not a machine that automatically produces truth. It is a set of practical moves for investigating online and AI-generated claims.

Move	Main question
Stop	What is the claim, and how much checking does it need?
Investigate the source	Who is behind this source, and is it suitable for this claim?
Find better coverage	Where can I find more relevant, responsible, or transparent information?
Trace the claim	Where did the claim or evidence originate, and what was its original context?

The moves do not always happen in a fixed order.

You may:

• move between them;

• repeat one move;

• trace a quotation before investigating the page that shared it;

• go directly to an official source for a current legal rule;

• stop when sufficient evidence has been found;

• stop because the claim does not justify more time;

• consult a qualified person when the consequences are serious.

SIFT is a flexible method, not a rigid checklist.

2. Stop: Identify the claim, stakes, and emotional pull

The first move is Stop.

Do not immediately believe, reject, share, or act on the information.

First, identify the exact claim.

Vague:

“AI is destroying education.”

More checkable:

“The article claims that students who use generative AI complete less independent writing.”

A bounded claim is easier to investigate.

Ask:

• What exactly is being claimed?

• Is it a fact, interpretation, prediction, quotation, statistic, image, or recommendation?

• Why does this claim matter?

• Who might be affected?

• What could happen if it is wrong?

• What evidence would be appropriate?

• How much checking does this claim deserve?

Notice emotional and urgent framing

Content may create:

• anger;

• fear;

• hope;

• excitement;

• outrage;

• a feeling of identity or belonging;

• satisfaction because it confirms what we already believe.

A strong emotional reaction does not prove that a claim is true or false. It is a signal to slow down.

Ask:

• Why do I want this claim to be true?

• Why do I want it to be false?

• Am I being pressured to react immediately?

• What evidence would change my mind?

Stopping interrupts emotional momentum and gives judgement time to begin.

3. Investigate: Understand the source laterally

The second move is Investigate the source.

The goal is not to label the entire source “good” or “bad.” The goal is to decide whether it is suitable for this particular claim.

Ask:

• Who created the content?

• Who published or distributed it?

• What is its purpose?

• What relevant expertise does it have?

• What evidence does it provide?

• Who funds it?

• What interests or incentives may matter?

• Does it have editorial standards?

• Does it publish corrections?

• Is it accountable for errors?

Do not rely only on the About page

A source’s About page may provide useful information, but it is self-description.

Use lateral reading:

1. Leave the page.

2. Open new tabs.

3. Search for information about the publisher or creator.

4. See what responsible independent sources say.

5. Look for relevant expertise, corrections, controversies, or conflicts of interest.

Reading one page more closely cannot reveal everything that the page leaves out.

Suitability depends on the claim

A source may be useful for one question but weak for another.

For example:

• A company may be the best source for its official product specifications.

• The same company is not independent evidence that its product is better than every competitor.

• A public authority may be the best source for the wording of a rule.

• Independent reporting may be necessary to evaluate how that rule is applied.

Ask:

What can this source reasonably establish?

4. Find better coverage

The third move is Find better coverage.

If the original page is weak, incomplete, or unsuitable, do not remain trapped inside it. Look for coverage that is better suited to the claim.

“Better” does not mean “agrees with me.”

Better coverage may offer:

• relevant expertise;

• direct access to evidence;

• current information;

• transparent methods;

• clear corrections;

• local applicability;

• responsible context;

• accountable authorship.

Depending on the question, better coverage may come from:

• official guidance;

• primary records;

• established journalism;

• peer-reviewed research;

• specialist organisations;

• transparent datasets;

• qualified local authorities;

• documented testimony from affected people.

Search strategically

A practical search process is:

1. Extract the central claim or key terms.

2. Search outside the original page.

3. Add relevant details such as country, date, organisation, or subject.

4. Look for sources responsible for the information.

5. Compare whether different pages are genuinely independent.

For example, a Swedish legal claim should normally lead to current Swedish legal or official sources—not simply pages that discuss a similar rule in another country.

Repetition is not always corroboration

Several pages may repeat the same original claim.

If five websites copy one incorrect report, they are not five independent confirmations.

Ask:

• Where did each page get the information?

• Are they citing one another?

• Do they all lead back to the same origin?

• Did any source check the claim independently?

5. Trace the claim to its original context

The fourth move is Trace.

Tracing means following a claim, quotation, statistic, image, study, or story back to an earlier or original source.

Ask:

• Where did this first appear?

• What was the original wording?

• What is the exact passage?

• What does the surrounding context say?

• Was the quotation shortened?

• Did a headline exaggerate the evidence?

• Did a repost change the meaning?

• Was the material used in a different time or place?

• Does the original method support the later conclusion?

Leave the page, then return to the evidence

SIFT involves two important movements:

Leave the page for orientation, then return to the evidence for close reading.

Lateral reading helps you understand the source and find better coverage. Tracing returns you to the relevant passage, data, record, image, or study.

The original source is not automatically true

Tracing shows where a claim began and how later versions changed it.

The original source may still contain:

• weak methods;

• incomplete evidence;

• commercial interests;

• selective reporting;

• misleading framing;

• an honest mistake.

Finding the origin answers:

“Where did this claim come from?”

It does not automatically answer:

“Is this claim true?”

6. Match the evidence to the claim

Different source types can establish different things.

Source type	What it may establish	Important limitation
Personal testimony	A person’s reported experience	Does not alone establish how common the experience is
Survey	Patterns among respondents	Quality depends on sampling and questions
Official record	The wording of a rule, decision, or recorded event	May not show how the rule works in practice
Journalism	Reporting and contextualisation	Quality and evidence vary
Research study	Findings produced through a stated method	One study may not settle the question
Commentary	An argument or interpretation	Should not be confused with factual reporting
Advertising	Claims made to promote something	Has a persuasive purpose
Satire	Humour or criticism through imitation	May be mistaken for literal news

Do not reduce the analysis to:

“This source is biased.”

Instead ask:

• What evidence does it provide?

• What can that evidence support?

• What interests or limitations matter?

• Does it answer the actual question?

A simple method check

For research, surveys, and data, ask:

1. Who or what was studied?

2. How was the information collected?

3. What do the results actually show?

4. Does the conclusion go beyond the results?

When relevant, also examine:

• sample size;

• measurement choices;

• uncertainty;

• missing groups;

• conflicts of interest.

Technical language, graphs, and statistics can look impressive while supporting only a limited conclusion.

7. Check date, place, and applicability

Correct information can still mislead when applied in the wrong context.

Ask:

• Is the information current?

• Does it apply in Sweden or another country?

• Is the population relevant?

• Is the law or policy context the same?

• Has the situation changed?

• Does the evidence fit this specific case?

A recently published page may repeat old information. A real rule from another country may not apply in Sweden. A study of one population may not justify conclusions about everyone.

Applicability means whether evidence fits the time, place, population, jurisdiction, purpose, or situation being discussed.

8. Apply SIFT to AI output and media

AI output should usually be treated as a set of claims and leads—not as final evidence for those claims.

Use SIFT with AI:

Stop

Extract the exact claim and consider the consequences if it is wrong.

Investigate

Identify who is responsible for the information. If AI names a report, organisation, author, or law, investigate that source outside the chatbot.

Find better coverage

Locate sources with appropriate expertise, responsibility, evidence, and local relevance.

Trace

Open the original document or passage and compare it with the AI claim.

AI may assist by:

• extracting checkable claims;

• suggesting search terms;

• naming possible organisations responsible for the information;

• helping organise a source trail.

But AI cannot independently verify its own original response. Every important source and passage still needs to be opened and checked.

Images and video

For an image or video, tracing may involve:

• locating an earlier version;

• checking the original date and caption;

• identifying where it was first published;

• seeing whether it was cropped or edited;

• checking whether it has been reused in a new context.

AI-generated images, deepfakes, and media verification will be examined more fully in Lessons 13 and 14.

9. Document and communicate the judgement

A source-critical judgement should be reproducible.

This means another learner can follow your steps and understand why you reached your conclusion.

Use a source trail:

Stage	What I found	Evidence or link	What it means
Exact claim
Stop
Source investigation
Better coverage
Original context
Applicability
Final judgement

A list of links is not enough. Explain why each source was chosen and what it contributed.

Communicate uncertainty

The final judgement may classify the claim as:

• Supported: Appropriate evidence supports the claim.

• Partly supported: Some elements are supported, but the claim is exaggerated, incomplete, or too broad.

• Unsupported: Adequate evidence was not found.

• False: Reliable evidence directly contradicts the claim.

• Still uncertain: Available evidence is incomplete, conflicting, or insufficient.

“Unsupported” does not always mean “proven false.” It means the required support was not found.

Choose a final action

Depending on the result, decide whether to:

• share;

• share with qualifications;

• correct;

• reject;

• escalate to a qualified person;

• leave the claim unshared.

Classroom activity: SIFT source audit

Choose a shared or AI-generated claim with manageable stakes.

Avoid acting on unverified high-stakes medical, legal, financial, or safety information.

Step 1: Write the exact claim

Rewrite it as a bounded, checkable statement.

Step 2: Stop

Record:

• why the claim matters;

• what emotional or urgent framing it uses;

• what could happen if it is wrong;

• how much checking it deserves.

Step 3: Investigate the source

Use lateral reading to examine:

• the creator or publisher;

• purpose;

• relevant expertise;

• funding or interests;

• corrections or accountability;

• suitability for this claim.

Step 4: Find better coverage

Locate one or two sources that are more relevant, responsible, transparent, or authoritative for the question.

Check whether the sources are genuinely independent.

Step 5: Trace the evidence

Trace one central quotation, statistic, study, image, or factual claim to its original or earlier context.

Read the surrounding passage.

Step 6: Check applicability

Consider:

• date;

• place;

• population;

• jurisdiction;

• method;

• scope.

Step 7: Make a judgement

Classify the claim as:

• supported;

• partly supported;

• unsupported;

• false;

• still uncertain.

Step 8: Choose an action

Decide whether to:

• share;

• share with qualifications;

• correct;

• reject;

• escalate;

• leave unshared.

Step 9: Submit the source trail

Complete the source-trail table and write approximately 300–500 words explaining:

• how your first impression changed or remained;

• what lateral reading revealed;

• which better coverage mattered;

• what tracing showed;

• whether date, place, or method changed the judgement;

• what remains uncertain;

• why you chose the final action.

Reflection questions

1. What makes a claim bounded and checkable?

2. Why is emotion a signal to pause rather than proof that a claim is false?

3. What does it mean to investigate whether a source is suitable?

4. Why is an About page not enough?

5. What is lateral reading?

6. What makes coverage “better” for a specific question?

7. Why do several repeated pages not always provide independent confirmation?

8. What can tracing reveal?

9. Why is an original source not automatically true?

10. What makes a source trail reproducible?

Key vocabulary

SIFT:
A flexible source-investigation method: Stop, Investigate the source, Find better coverage, and Trace claims to their original context.

Lateral reading:
Leaving a page to examine what other responsible sources say about the source and claim.

Better coverage:
Information that is more suitable for the question because it offers stronger relevance, evidence, expertise, transparency, currency, or accountability.

Trace:
Following a claim, quotation, statistic, image, study, or story back to an earlier or original context.

Original context:
The initial wording, surrounding material, date, source, and situation in which information appeared.

Source type:
A category such as testimony, official record, journalism, research, commentary, advertising, or satire.

Applicability:
Whether evidence fits the time, place, population, jurisdiction, purpose, or situation.

Source trail:
A record of searches, sources, passages, comparisons, reasoning, judgement, and remaining uncertainty.

Supported:
Backed by appropriate evidence that fits the claim.

Partly supported:
Supported in some respects but exaggerated, incomplete, or too broad.

Unsupported:
Lacking adequate evidence after reasonable investigation.

Summary

In Week 12, we learned how to use SIFT to investigate web and AI-generated claims:

• Stop and identify the exact claim.

• Investigate whether the source is suitable.

• Find better and more relevant coverage.

• Trace the evidence to its original context.

SIFT is flexible. Learners may move between the steps, repeat them, or stop when they have enough evidence for the task and consequence.

The process involves two important movements:

Leave the page for orientation, then return to the evidence for close reading.

An original source is not automatically true, repeated pages are not automatically independent, and emotional reactions do not prove or disprove a claim.

A responsible judgement shows its source trail and communicates whether the claim is supported, partly supported, unsupported, false, or still uncertain.

Lesson 12 Interactive Quiz: Source Criticism with SIFT

Choose one answer for each question. Then select Check my answers. You will receive feedback for every question. This practice quiz does not collect names or scores.

1. Which statement best describes how SIFT should be used?

A. Every claim must receive the same amount of investigation. B. Every move must be completed once in a fixed order. C. The moves are flexible and can be repeated or reordered when needed. D. SIFT is needed only when a source looks unprofessional.

2. A post creates immediate anger and urges readers to share it before it is removed. What should the learner do first?

A. Pause, identify the exact claim, and decide what evidence is needed. B. Reject it because emotional content is always false. C. Share it quickly and investigate afterwards. D. Accept it if it confirms something already suspected.

3. A website’s About page says it is “the world’s most trusted health authority.” What is the best next step?

A. Accept the statement because the website knows its own qualifications. B. Read the About page repeatedly until its standards become clear. C. Judge credibility mainly from the site’s design and professional language. D. Leave the site and investigate its expertise and reputation through independent sources.

4. Which is likely to provide better coverage for a claim about a current Swedish employment rule?

A. A popular discussion page describing a similar rule in another country. B. Current Swedish legislation or guidance from a responsible Swedish authority. C. An undated personal blog that repeats the claim without sources. D. A chatbot response that uses formal legal vocabulary.

5. A learner traces a viral statistic to the report where it first appeared. What can the learner conclude?

A. The statistic is automatically true because the first report was found. B. Later pages no longer need to be compared with the original wording. C. The origin and context are now clearer, but the evidence and method still require evaluation. D. The statistic is reliable if the original report contains a chart.

6. What can one person’s detailed testimony reasonably establish?

A. What that person reports experiencing. B. How frequently the same experience occurs in the whole population. C. That one identified cause explains every similar experience. D. That the same conclusion applies in every country and context.

7. Five articles repeat the same claim, but all five cite one original report. What does this show?

A. Five independent investigations confirmed the claim. B. The claim is true because it appears on several websites. C. No further tracing is necessary because the report is popular. D. The pages may be repeating one origin rather than providing independent confirmation.

8. A small voluntary survey is used to make a claim about all adults in Sweden. What is the most important critical question?

A. Does the report use enough technical vocabulary? B. Does the sample and collection method support such a broad conclusion? C. Is the report’s chart visually attractive? D. Has the conclusion been repeated on social media?

9. Which use of AI is appropriate during a SIFT investigation?

A. Ask AI to certify that its original response is correct. B. Treat a second generated answer as independent evidence. C. Use AI to suggest search terms, then independently open and evaluate the sources found. D. Let AI decide whether its own citation supports the claim without opening it.

10. Which source trail is the most reproducible?

A. It records the exact claim, searches, sources, relevant passages, comparison, judgement, and remaining uncertainty. B. It lists several links without explaining why they were selected. C. It gives a confident conclusion without showing the evidence path. D. It includes one screenshot and asks the reader to trust the investigator.

Reflection: Choose one shared or AI-generated claim. Which SIFT move would you begin with, what evidence would you seek, and what would make you stop the investigation?