Source Evaluation SKILL.md

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

---
name: source-evaluation
description: "Evaluate sources before relying on them: lateral reading and the SIFT method, deeper checks of authority, method, evidence, currency, purpose and transparency, red flags for statistics, surveys, studies, reports, news, websites, social media and AI outputs, and confidence labels with reasons. Use whenever someone needs to check whether a source is credible, evaluate a statistic, study or report, verify AI-cited sources, judge a website or decide how much to trust information before using it."
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/source-evaluation/
  category: productivity
  subcategory: research-learning
  version: "1.0.0"
  created: "2026-10-10"
  updated: "2026-10-14"
---

# Source Evaluation

Not all sources deserve the same trust. A statistic in a slick report may come from a tiny self-selected survey; a news story may misread a study; a website that looks official may be a lobby group; an AI tool may cite a page that does not say what it claims. Research by the Stanford History Education Group found that professional fact-checkers judged online sources more accurately than other experienced readers because they read laterally: they left the page early to see what other sources said about it. [Research, Wineburg and McGrew, Lateral Reading, 2017] Mike Caulfield's SIFT method turns that into four quick moves: Stop, Investigate the source, Find better coverage, Trace claims to the original. [Reference, Caulfield, SIFT]

This skill evaluates sources before relying on them: quick SIFT checks, deeper assessment of authority, method, evidence, date, purpose and bias, special checks for statistics, studies, reports and AI-generated content, and a clear confidence label for each source.

## Operating principles
1. **Read laterally.** Check what independent sources say about the source before trusting it.
2. **Trace to the original.** Claims about studies and statistics must be checked at their origin.
3. **Check method, not just reputation.** Even reputable sources publish weak analyses.
4. **Consider purpose and incentives.** Who funded it, and what do they want readers to believe?
5. **Date matters.** Old data may no longer be true.
6. **Label confidence.** Every source used gets high, medium or low confidence with a reason.

## SIFT in practice
| Move | What to do | Time |
|---|---|---|
| Stop | Notice your reaction; ask whether you know this source | Seconds |
| Investigate the source | Search the publisher or author in a new tab; look for independent descriptions | 1 to 2 minutes |
| Find better coverage | Look for reputable coverage of the same claim | 2 to 5 minutes |
| Trace claims | Follow quotes, statistics and studies to the original | 5 to 15 minutes |

## Deeper assessment
| Criterion | Questions |
|---|---|
| Authority | Who wrote or published it? What expertise and track record? |
| Method | How was the data collected? Sample size, selection, definitions, time period? |
| Evidence | Are claims supported by data or citations? Can they be checked? |
| Currency | When was it published and when was the data collected? Has anything changed since? |
| Purpose | Inform, sell, persuade, lobby or entertain? Who funded it? |
| Consistency | Does it agree with other reliable sources? If not, why? |
| Transparency | Are methods, data and conflicts of interest disclosed? |

## Checks for specific source types
| Source | Red flags | Checks |
|---|---|---|
| Statistics | No method, round numbers, no date | Find the original; check definition and sample |
| Surveys | Small or self-selected sample, leading questions | Sample size, who was asked, exact wording |
| Academic studies | Not peer-reviewed, small samples, one study only | Journal quality, replication, systematic reviews |
| Industry reports | Vendor-sponsored conclusions | Sponsor, method appendix, independent confirmation |
| News | Single anonymous source, headline beyond the story | Read the full piece; find the primary source |
| Websites | Unclear ownership, no contact details | Look up the organisation laterally |
| AI outputs | Confident claims, citations that do not match | Open every cited source; verify key facts |
| Social media | Screenshots without links, viral claims | Find the original post and context |

## When to use, and when not to
Use it whenever research supports a decision, report, article or public claim, especially for statistics, health, finance, legal or political topics and AI-generated summaries.

Don't spend deep evaluation time on trivial, low-stakes facts; a quick SIFT is enough.

## Inputs
- The source or claim.
- What it will be used for and how important it is.
- Other sources already found.

## Workflow

### Step 1. Decide how much checking is needed
High-stakes claims (money, health, law, public statements) get deeper checks.

### Step 2. Run SIFT
The four moves above, in a few minutes.

### Step 3. Assess method and evidence
For data and studies, read the methodology: sample, definitions, period, analysis.

### Step 4. Look for conflicts of interest
Funding, authorship, sponsorship, affiliations.

### Step 5. Compare with other sources
Agreement among independent, reliable sources raises confidence; conflicts need explanation.

### Step 6. Label confidence
High, medium or low with a one-line reason.

### Step 7. Record
In the research log: source, assessment, confidence, how it will be used.

## Confidence labels
| Label | Typical profile |
|---|---|
| High | Primary or highly reputable source, transparent method, recent, confirmed by others |
| Medium | Reputable but secondary, some method gaps, or older data |
| Low | Unclear method, interested party, unconfirmed, or old |

## Reading a study quickly
1. What question did it ask, and who was studied (sample size, population, place)?
2. What did it measure, and how?
3. Was it an experiment, an observational study or a survey?
4. How big is the effect, not just whether it is significant?
5. What limitations do the authors state?
6. Has it been replicated or included in a systematic review?
A single observational study rarely proves cause and effect; systematic reviews and meta-analyses usually give a more reliable picture.

## Evidence hierarchy (rough guide)
| Stronger | Weaker |
|---|---|
| Systematic reviews and meta-analyses | Single studies |
| Randomised experiments | Observational studies |
| Large, representative samples | Small or self-selected samples |
| Transparent methods and data | Undisclosed methods |
| Independent funding | Interested sponsors |
The hierarchy guides judgement but does not replace it: a well-designed observational study can be more useful than a poor experiment.

## Using AI to help evaluate
An AI assistant can summarise a source's methodology, look up the publisher and list potential conflicts of interest, but the final judgement needs a person to open the original and read the key sections, especially for high-stakes claims.

## Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Only one source exists | Use with a low or medium label; say it is unconfirmed |
| Expert disagreement | Report main positions and the evidence for each |
| Source reputable but study weak | Judge the study, not the brand |
| Sponsored research | Use cautiously; look for independent confirmation |
| Retracted or corrected articles | Check for retractions and corrections before citing |
| Preprints | Label as not yet peer-reviewed |

## Handoffs to related skills
- Finding sources: `desk-research`
- Checking specific claims: `fact-checking`
- AI research outputs: `ai-assisted-research`
- Academic studies: `literature-review`
- Synthesis: `research-synthesis`

## Worked example
A team planned to quote "70% of buyers research online before contacting sales" in a pitch. A quick SIFT showed the figure appeared on dozens of marketing blogs; tracing it led to a vendor report several years old with no published method. Better coverage found more recent independent surveys with disclosed samples showing a related but different finding. The team replaced the claim with the newer source, stated the sample and year, and labelled confidence medium.

## Common mistakes
- Judging a page by how professional it looks.
- Citing a statistic from an article instead of the original.
- Ignoring dates and methods.
- Trusting AI citations without opening them.
- Treating one study as settled evidence.

## Output format
```
1. Source and claim
2. SIFT findings
3. Method and evidence assessment
4. Conflicts of interest
5. Comparison with other sources
6. Confidence label and reason
```

## Self-check before you answer
- Was the source checked laterally?
- Were key claims traced to their original?
- Were method, date and purpose considered?
- Does each source have a confidence label with a reason?

---
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/source-evaluation/

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.