Skip to content

Latest commit

 

History

History
227 lines (154 loc) · 6.35 KB

research-methods.md

File metadata and controls

227 lines (154 loc) · 6.35 KB

Research Methods That Make OSINT Better

Good OSINT is less about knowing one clever search trick and more about using repeatable methods. The methods below are useful across investigations because they reduce guesswork and make conclusions easier to defend.

1. Start with a question, not a person or a tool

Weak start:

Find everything about X.

Better start:

Was this public claim published before 12 March?
Is this image consistent with the claimed location?
Is this domain officially connected to the organization?

A narrow question tells you what evidence matters and when to stop.

2. Build an evidence ladder

Not every source deserves the same weight. A useful default order is:

Primary record / direct source
        ↓
Independent first-hand source
        ↓
Reliable secondary reporting
        ↓
Aggregator / repost
        ↓
Unverified social claim

The ladder is not absolute. A primary source can lie and a secondary source can correct it. The point is to ask how close each item is to the event or fact being examined.

3. Search sideways, not only deeper

When one search result looks promising, researchers often keep digging into the same source. That creates tunnel vision.

Instead, search sideways:

  • another search engine;
  • another language or spelling;
  • a date-restricted search;
  • an archive;
  • a primary-source database;
  • a source that disagrees with your current hypothesis.

This is especially useful when many articles repeat the same original report.

4. Use negative-space research carefully

Sometimes what is missing is useful, but absence is weak evidence.

Examples:

  • no archived page does not prove the page never existed;
  • no search result does not prove something was never published;
  • no WHOIS identity does not prove a domain has no identifiable owner;
  • no metadata does not prove an image was edited.

Phrase absence correctly:

I found no public evidence in the sources checked.

not:

There is no evidence.

5. Track provenance from the first minute

For every useful item, record:

  • URL;
  • title or description;
  • publisher/account;
  • publication time if available;
  • access time;
  • archive link if appropriate;
  • what claim the source supports;
  • whether it is primary, secondary, or derivative.

Doing this later is slower and increases the chance of mixing sources.

6. Separate discovery from verification

Discovery asks:

What might be relevant?

Verification asks:

What can I actually support?

A username hit, reverse-image match, subdomain, social post, or AI-generated lead belongs in discovery until it survives verification.

7. Use confidence as a reasoned judgment

Avoid confidence labels with no explanation.

Instead of:

Confidence: High

write:

Confidence: High - the claim is supported by two independent primary records and a contemporaneous archive capture.

Confidence should reflect source quality, independence, consistency, and unresolved contradictions.

8. Keep an alternative-hypothesis column

A simple investigation table can prevent confirmation bias:

Observation Current explanation Alternative explanation What would distinguish them?
Same username Same operator Coincidence or recycled handle Direct cross-link or unique shared identifier
Same image Same source Repost or copied media Earliest upload and provenance
Same infrastructure Same organization Shared hosting/provider Registration, certificates, official links, historical records

Do not try to make alternatives sound equally likely. Just record plausible alternatives that could change the conclusion.

9. Investigate time as carefully as content

Common timeline mistakes include:

  • confusing publication and update times;
  • ignoring time zones;
  • treating an archive capture time as the page's publication time;
  • using a repost date as an original upload date;
  • assuming a screenshot was captured when it was posted.

Normalize important timestamps to one timezone in your notes and preserve the original value too.

10. Pivot on stable identifiers

Names and usernames change. Stable or semi-stable public identifiers can be better pivots, depending on the investigation:

  • official domains;
  • public organization IDs;
  • repository URLs;
  • certificate names;
  • document identifiers;
  • public company numbers;
  • published email domains;
  • image hashes for duplicate detection.

A pivot creates a lead. It does not automatically establish identity or ownership.

11. Use multilingual and transliteration searches

For international research, search variants of:

  • local script;
  • Latin transliteration;
  • alternative transliterations;
  • former names;
  • abbreviations;
  • translated job titles or organization names.

Record which spelling produced each result so another researcher can reproduce the search.

12. Challenge your strongest finding

Before finalizing a report, take the conclusion you trust most and try to break it.

Ask:

  1. Could these sources all depend on one original source?
  2. Is there a plausible alternative explanation?
  3. Did I confuse correlation with attribution?
  4. Is the timestamp actually what I think it is?
  5. What evidence would make me lower my confidence?

This step catches more errors than adding another dozen tools.

AI-assisted OSINT

AI can be useful for:

  • generating search variants;
  • translating search terms;
  • summarizing notes you already collected;
  • extracting entities from public documents;
  • suggesting alternative hypotheses;
  • turning raw notes into a draft timeline;
  • helping write small scripts for data cleanup.

But AI output is not evidence.

Use this rule:

AI may suggest the next question.
A verifiable source must support the answer.

Do not cite a model response as proof of a factual claim. Re-check names, dates, quotations, URLs, and relationships against the underlying source.

The stop rule

An investigation needs a stopping condition. Stop when:

  • the question is answered to the required confidence;
  • additional searching is producing repeated information;
  • remaining uncertainty cannot be resolved with lawful public sources;
  • continuing would exceed the original scope;
  • the next step would require private access, deception, or intrusive collection.

Knowing when to stop is part of good OSINT.