00 — Before you start
The premise
Research once meant finding scarce information. Now the information is abundant and uneven, and the hard part is judging it: what is true, what is current, what is worth trusting. This guide is about that judgement, and it deliberately avoids naming today’s tools as if they were permanent.
Tools are examples
Method over feature
Verify everything
01 — Orientation
From finding to verifying
The old skill was locating a source. The new skill is deciding whether to believe the many you are handed. Abundance changes the whole job: you spend far less effort finding, and far more filtering, checking and combining. Treat research as an act of judgement, not retrieval.
Then
Scarcity
The work was access: getting to the one library, the one expert, the one document that held the answer.
Now
Abundance
The work is discrimination: choosing, among many confident answers, the ones actually worth trusting.
The shift
Where the effort moves
Speed of finding is no longer the bottleneck. Quality of judgement is. The rest of this guide is about spending your effort where it now matters.
02 — The question
The question comes first
Most bad research is a good tool aimed at a vague question. A sharp question does half the work: it tells you what would count as an answer, which sources could hold it, and when you are done. Spend time on the question before you spend any on the search.
2.1
Make it answerable
Turn “learn about X” into a question with a shape: who, when, how much, compared to what. Vague questions return vague results.
2.2
Name the answer
Decide in advance what a good answer looks like: a number, a date, a source, a comparison. You cannot recognise the end without it.
2.3
Split the big ones
A large question is usually several small ones. Break it down, answer the parts, and the whole tends to assemble itself.
2.4
Know when to stop
Research expands to fill the time you give it. Set the bar for “good enough” up front, and stop when you clear it.
Figure 01 — illustrating section 02
The research loop
Research is a loop, not a line. Ask a sharp question, gather, verify what you gathered, synthesise it, and let what you learn sharpen the next question. Most poor research skips verify and jumps from gather straight to done.
03 — Search craft
Searching well
A search box rewards precision. The difference between a novice and a fast researcher is rarely the tool; it is knowing how to narrow. A handful of operators, understood once, turn a vague scan into a targeted query, and they work across almost every search engine.
3.1
Narrow deliberately
Start broad to see the landscape, then add terms to cut it down. Each added word is a filter; use them on purpose.
3.2
Use the operators
Exact phrases, site limits, exclusions, file types and date ranges do most of the narrowing. They are near-universal and worth memorising once.
3.3
Speak the source’s language
Search with the words your target source would use, not your own. Technical results want technical terms; plain results want plain ones.
3.4
Read the results as data
The titles, snippets and domains that come back tell you whether you are close. If they are off, change the query, not your expectations.
Figure 02 — illustrating section 03
Operators that narrow
Operators that narrow a search
A small set does most of the work. A near-universal set of search operators does most of the work of narrowing a query. Learn these once and they carry across engines and across years, since they describe intent rather than any one product.
04 — Two engines
Search and answer engines
There are now two broad kinds of tool for a question. A search engine hands you sources to judge. An answer engine hands you a synthesised reply to verify. They are good at different jobs, and the skill is choosing the right one for the moment rather than defaulting to whichever is nearest.
Search engine
Answer engine
Figure 03 — illustrating section 04
Which engine, when
Different jobs. Most real research uses both, in turn.
The choice comes down to what you need. Sources, current facts and verification point to a search engine, where you judge. Synthesis, drafting and exploration point to an answer engine, where you verify. Real work uses both in turn.
05 — Assistants
Answer engines and AI assistants
AI assistants are a powerful class of answer engine, and worth understanding as a class rather than as this month’s product. They are genuinely useful for the right jobs and quietly unreliable for others. Knowing which is which is most of what separates good use from bad.
Good at
Synthesis and language
Explaining an unfamiliar area, restructuring information, drafting, brainstorming, and turning a mess of notes into a shape. Their strength is synthesis and language.
Weak at
Currency and citation
Being current, being certain, and citing accurately. They can produce fluent, confident text that is subtly or wholly wrong, including invented sources.
Staleness
The staleness limit
A model knows what it was trained on, up to a point in time. For anything recent, treat it as possibly out of date unless it is genuinely retrieving live sources.
Citations
The made-up citation
A named source from an assistant is a claim to check, not a reference to trust. If you cannot find the source yourself, assume it may not exist.
Figure 04 — illustrating section 05
Instrument strengths
| Instrument | Find | Verify | Current | Structured | Synthesise |
|---|---|---|---|---|---|
| Search engine | strong | strong | strong | weak | weak |
| Answer engine | strong | weak | partial | weak | strong |
| Video | partial | partial | partial | weak | partial |
| API / primary | partial | strong | strong | strong | weak |
| Community | partial | partial | partial | weak | partial |
No single instrument covers it all. Each class of instrument is strong at some jobs and weak at others. No single one covers finding, verifying, currency, structured data and synthesis. Reading across the row tells you what to reach for, and what to reach for next.
06 — Media
Video and other media
Some knowledge is easier to show than to write, and video holds a lot of it: demonstrations, walkthroughs, first-hand accounts, and expertise that never became an article. Treated as a source rather than entertainment, it is an under-used part of research.
6.1
Search it like text
Video platforms are search engines. The same query craft applies, and the results reach material that exists nowhere in writing.
6.2
Use transcripts
A transcript turns a video into scannable, searchable text. It is faster to read than to watch, and lets you jump to the part that matters.
6.3
Jump with timestamps
Chapters and timestamps let you treat a long video as a set of sections, taking the two minutes you need and leaving the rest.
6.4
Judge it like any source
A confident demonstration is still a claim. Check who made it and whether it holds up, exactly as you would a written one.
07 — The source
APIs and going direct
Sometimes the best research skips the reading and goes to the data. Many sources publish their information directly, in a structured form you can pull rather than paraphrase. Going to the source removes a layer of interpretation, and with it a layer of error.
7.1
Prefer the primary
An article about a dataset is a secondary source. The dataset itself is primary. When the raw data is reachable, it settles questions the commentary only argues about.
7.2
Structured beats prose
Data offered in a structured feed can be sorted, counted and checked yourself, rather than trusting someone else’s summary of it.
7.3
Know it exists
You do not need to be a programmer to know that official figures, catalogues and records are often published directly. Knowing to look is most of the advantage.
7.4
Cite the origin
When you use direct data, record where it came from and when you pulled it. Structured sources change, and a dated origin protects you later.
08 — Provenance
Sourcing and provenance
A claim is only as good as where it came from. Provenance is the chain from an assertion back to its origin, and following that chain is the core discipline of research. The nearer you get to the original, the more you can trust what you carry away.
8.1
Trace the chain
Ask where a claim originates. A report cites an article that cites a study: keep walking back until you reach the source, not the echo.
8.2
Primary over secondary
The original document, statement or dataset beats any account of it. Accounts drift, simplify and mistranslate as they pass along.
8.3
Watch for the loop
Several sources repeating one another is not confirmation. It often traces back to a single unchecked origin wearing many coats.
8.4
Keep the record
Note where each fact came from as you go. Provenance you did not record is provenance you have lost, and will not reconstruct later.
Figure 05 — illustrating section 08
The provenance ladder
Sources sit on a ladder of trust. From an unsourced assertion at the bottom to the primary origin at the top, research is the act of climbing it: for anything that matters, keep going until you reach the original.
09 — Triangulation
Cross-checking a claim
No single source, however good, should carry an important claim alone. Triangulation is confirming something from several independent sources that do not depend on one another. When they agree independently, confidence is earned rather than assumed.
9.1
Three, not one
For anything load-bearing, seek at least a few separate confirmations. One confident source is a lead, not a fact.
9.2
Independence is the point
Three sources that all copied the same origin are one source. Check that they arrived at the claim separately, or the agreement means nothing.
9.3
Weigh, don’t just count
A single primary source can outweigh many secondary echoes. Quality of origin matters more than number of repeats.
9.4
Mind the disagreement
When good sources conflict, that is information. Report the conflict honestly rather than picking the answer you preferred.
Figure 06 — illustrating section 09
Three independent sources
Three independent sources that agree, not three echoes of one.
Independent sources, not echoes. A claim is held up by independent sources that reached it separately. Three echoes of one origin are not three sources. Test for independence before you let agreement raise your confidence.
10 — Traps
The common traps
Modern research has a predictable set of failure modes, most of them quiet. Naming them is the defence, because each is easier to catch than to resist once you know its shape.
10.1
Confirmation
Searching, and reading, for what you already believe. The counter is to look actively for the strongest case against your view.
10.2
Fluency as truth
Confident, well-written text feels true. Polish is not evidence. Judge the claim, not the prose it arrives in.
10.3
Optimised noise
Much of what ranks well is built to rank, not to inform. A top result is a popular one, not necessarily a correct one.
10.4
The echo chamber
Tools and feeds tend to return more of what you already see. Deliberately seek sources outside the ones served to you.
10.5
Stopping too early
Taking the first plausible answer. The first answer is a starting point; the checked answer is the finding.
11 — First actions
The A–Z
A checklist that survives any change of tool, because none of it names one. Run it on your next real question and the method does the rest.
Write the question as a question, sharp enough to recognise the answer.
Decide up front what a good-enough answer looks like.
Split a big question into the smaller ones it hides.
Choose the engine for the job: sources to judge, or synthesis to verify.
Narrow with operators: exact phrase, site, exclude, filetype, date.
Use an assistant to get oriented, then verify every load-bearing claim.
Search video and media for expertise that never became an article.
Go to the primary source or raw data whenever it is reachable.
Trace each important claim back to its origin, not its echo.
Confirm load-bearing facts from independent sources, and test independence.
Record source, author and date as you use each fact.
Look actively for the strongest case against your current answer.