Online research is not opening twenty tabs. It is turning a fuzzy doubt into a question that evidence can settle, then keeping enough of the trail that another person can check you. My rule is strict: a search result is a lead, not a source, and an AI answer is an unverified tip from a stranger.
🎙️ Published & recorded: · Written and recorded by SwiftGrasp
01Write a question evidence can answer
“Research remote work” is an invitation to wander. Write the decision, population, comparison, measure, and time period. Ask “For software teams of fifty to two hundred people, did fully remote work change voluntary turnover during 2024?” Now you know which words matter, what data could answer, and what would not.
Topic
Is passwordless login better?
Verifiable question
For consumer web apps using passkeys in 2025:
• Does signup completion change versus email plus password?
• What recovery failures appear after users lose a device?
• Which browsers and operating-system versions were measured?
Decision: whether to run a passkey pilot, not whether passkeys are “good.”
Failure: the answer moved with the question
A team searched “best project management tool” and copied three ranking pages. They had never named team size, guest access, or data residency. Repair it step by step. Write the decision first. List three must-pass constraints. Define the comparison date. Reject pages that never test those constraints. A prettier list of tools would not have fixed the missing question.
02Climb a query ladder
Start broad enough to learn the vocabulary, then narrow toward documents. Do not cram every operator into the first query. Each search should teach you the next search: the agency name, report title, product code, statute number, or phrase used by experts.
1 Vocabulary: passkey account recovery failure
2 Exact phrase: "account recovery" passkeys
3 Source family: site:w3.org passkey recovery
4 Document type: filetype:pdf passkey deployment report
5 Exclude noise: passkey recovery -vendor -sponsored
6 Known title: intitle:"WebAuthn" implementation guidance
7 Time boundary: after:2024-01-01 before:2026-01-01
Useful operators vary by search engine. Test one at a time.
Real search failure: zero results
The exact query “Your search - "regional passkey abandonment benchmark 2025" - did not match any documents.” is not proof that no research exists. Remove the invented phrase “abandonment benchmark.” Search passkey conversion and WebAuthn registration separately. Find the vocabulary used in one credible paper, then search its title, author, and references. Quotation marks preserve bad wording as efficiently as good wording.
03Walk upstream to the primary source
News stories, summaries, and social posts are maps. Follow their links until you reach the study, court filing, company report, dataset, specification, transcript, or official release behind the claim. If every article cites another article, you have a citation carousel, not independent evidence.
Claim in blog: “The rule affects 40% of contractors.”
Trace:
blog → newspaper → trade association press release → survey PDF
At the PDF, check:
• Who designed and paid for the survey?
• How were contractors sampled?
• What was the exact question?
• Does 40% mean all contractors or 40% of 312 respondents?
• Is there a table or only a press-release sentence?
Prefer the closest source to the event, but do not worship “official.” A company is primary for what it announced and for the figures in its filing. It is not neutral evidence that its product is best. Primary describes proximity, not honesty.
04Judge the source and its incentives
Authority is claim-specific. A security engineer may explain encryption well and still be a poor source for the market size of her employer's product. Check authorship, method, evidence, corrections, funding, and what the publisher gains if you believe the claim.
SIFT THE SOURCE
Scope Does it address my exact question?
Identity Named author, role, contact, publication history?
Funding Sponsor, affiliate links, employer, political interest?
Method Sample, definitions, exclusions, raw data, limitations?
Trace Can I open the cited evidence?
Correction Date, version history, correction policy?
“Experts say” names nobody.A source that shows its limits is stronger, not weaker.
Search result that looked like evidence
A result read, “Top 10 VPNs for Privacy | We tested every leading provider.” The page's disclosure said it earned commission when readers bought through its links, and its “tests” did not name a protocol, server, date, or measurement. Repair: use the page only to collect product names, inspect independent protocol audits and privacy policies, and record the affiliate relationship beside any surviving claim.
05Match the date and version to the claim
A correct page can be wrong for your date. Software documentation may describe the current release while your server runs a version from two years ago. Laws have effective dates. Datasets have collection periods. Product pages quietly change while keeping the same URL.
Record four clocks when they matter:
PUBLISHED 2024-03-12
UPDATED 2025-09-08
DATA COLLECTED 2023-01 through 2023-12
VERSION PostgreSQL 16.4
Question to ask:
“Was this true on the date of the decision and for our version?”
Do not cite “updated yesterday” when the underlying table is from 2018.
Failure: right instructions, wrong release
A developer followed a current API page and got “400 Bad Request: Unrecognized field 'functions'.” The account was pinned to an older API version that expected a different field. Repair: capture the endpoint and version header, open the archived docs for that version, compare the request schema, change one field, and save the working request with its date. “The docs are wrong” was too broad to debug.
06Cross-check claims, not page counts
Five pages repeating one press release are one source wearing five shirts. For an important claim, seek evidence created through different routes: a regulator's filing, a company's report, a dataset, and an independent analysis. Then explain disagreements instead of averaging them away.
CLAIM: “Complaints doubled in 2025.”
Source A Company report 1,040 → 2,090 tickets
Source B Regulator database 210 → 260 complaints
Source C Support dashboard changed category rules in July
Reconciliation:
The company count doubled, partly after a classification change.
The regulator count rose 24%. These measures are not interchangeable.
Do not write: “Multiple sources confirm complaints doubled.”
Keep a claim table. Put each important sentence in one row and each genuinely independent source in a column. Empty cells make uncertainty visible. That is better than attaching six links to the end of a confident paragraph.
07Archive the page and preserve the quote
A citation should let someone recover the evidence you saw. Save the author, exact title, publisher, publication or update date, URL, access date, relevant quote, and page or section. For unstable pages, save a lawful snapshot or web-archive URL. A bare bookmark is not a research record.
CLAIM ID: C-07
Claim: Vendor will support version 3 through 2028-06-30.
Source: “Version 3 Support Policy,” Acme Docs
Published/updated: 2026-02-10
URL: https://example.com/support/v3
Accessed: 2026-07-25
Evidence: “Security fixes continue through June 30, 2028.”
Location: Long-term support, paragraph 2
Archived URL / local PDF hash:
Notes: Applies to Enterprise plan only.
Quote enough context to preserve the condition. “Security fixes continue through 2028” becomes misleading if the next sentence limits that promise to one paid plan. Screenshots help with visual evidence, but searchable text and a stable URL are easier to audit.
08Treat AI answers as unverified leads
A fluent answer can contain a fake title, a real author attached to the wrong paper, or a real paper that does not support the sentence. Never cite the chat. Open every reference, match its title and authors, read the relevant passage, and verify that the claim survives outside the answer.
AI answer, copied exactly:
“Smith and Rao (2021), Journal of Digital Methods,
doi:10.1234/jdm.2021.0042, found a 37% accuracy increase.”
DOI lookup:
DOI Not Found
10.1234/jdm.2021.0042
Repair:
1. Search the exact title without the DOI.
2. Search both authors plus the claimed phrase.
3. Search Crossref or the publisher by title and year.
4. If no record appears, mark the citation fabricated and delete the claim.
5. Find real studies from a database, then read methods and results.
The dangerous part was the plausible detail
The invented DOI looked structured, the journal name sounded ordinary, and 37 percent felt precise. None of that was evidence. Do not ask the same model, “Are you sure?” That produces another sentence. Leave the model and verify against the publisher, bibliographic index, or original document.
09Recover paywalled, deleted, and broken pages
A blocked URL is not permission to invent what it said or copy an unauthorized mirror. Search the title, author, distinctive quoted sentence, report number, and filename. Check the publisher's archive, an institutional repository, the author's page, a library, or a lawful web archive. Contact the author when the paper matters.
Original request:
HTTP/1.1 404 Not Found
The requested URL /reports/2024-risk-review.pdf was not found.
Recovery ladder:
1. Search: "2024 Risk Review" publisher name
2. Search: site:publisher.example filetype:pdf "Risk Review"
3. Remove /reports/ from the URL and inspect publications
4. Search the exact filename and report number
5. Check an archived capture of the original URL
6. Ask the publisher for the superseding document
7. Record that the original was unavailable and what replaced it
Paywall failure
A researcher quoted a search snippet because the article body was paywalled. The snippet ended before the word “not,” reversing the finding. Repair: do not cite the snippet. Search the title for an author manuscript or repository copy, use library access, read the abstract on the publisher page, and narrow the claim to what you actually opened. If the crucial sentence remains unavailable, leave it out.
10Keep a research record and stop on purpose
Research ends when the decision has enough verified evidence, not when your tabs become embarrassing. Keep queries that failed, sources you rejected, claim-level notes, and unresolved conflicts. This prevents repeated work and shows where confidence should stay low.
QUESTION
□ Decision, population, comparison, measure, time period
SEARCH
□ Broad vocabulary → exact phrase → source/domain → document
□ Record useful and failed queries
VERIFY
□ Open the primary source
□ Check author, method, funding, incentives, date, and version
□ Find independent evidence for the important claims
□ Open every AI-suggested citation; match claim to passage
PRESERVE
□ Title, author, publisher, URL, date, access date, quote, location
□ Archive unstable pages lawfully
□ Mark paywall, 404, replacement, and unresolved disagreement
STOP
□ Can another person reproduce the path?
□ Would missing evidence change the decision?
□ Is confidence stated honestly?
My stopping rule: if one more search is unlikely to change the decision, write the remaining uncertainty and stop. Endless browsing is not rigor. A short trail of inspectable evidence beats a long bibliography nobody opened.