Define the information need before opening a database
Write a question narrower than the dissertation topic. “Digital education” names a field; “Which conditions influence secondary-school teachers’ use of digital learning materials?” can guide a search. Break it into concepts, synonyms, broader terms and terms used in other relevant languages. State date range, languages, document types and disciplinary boundaries before individual results reveal which limits would favour your argument.
This record does not prevent learning. If a newly discovered technical term improves retrieval, add it with a date and reason. If foundational studies fall outside the initial period, record why you widened it. Integrity is not rigid adherence to a poor plan. It is the ability to distinguish a reasoned revision from a retrospective selection designed around known results.
A search result is not yet reliable evidence
Open the complete record and check whether title, authors, year, journal or publisher, DOI and stable location agree. A DOI supports identification; it does not certify methodological quality. For important research, look for corrections, retractions and later versions. Crossref exposes bibliographic metadata and relations such as updates, although a complete status review may require the relevant disciplinary database too.
Read methods and limitations rather than relying on the abstract and discussion. Ask which population, data and instruments were actually studied. Work on a small, tightly defined group cannot support a universal claim about all students. Guidance documents support what an organisation recommends, not automatically the effectiveness of that recommendation. A press release may point you somewhere useful, but a substantial claim should return to the paper, statute, dataset or official statistics on which it rests.
Selection criteria that resist confirmation bias
Set relevant criteria before knowing individual findings. Inclusion and exclusion may refer to design, population, time period, language, availability of needed data or publication type. “Does not fit my argument” is not a defensible exclusion reason. Retain credible evidence that challenges the working thesis, then explain differences in design, context or uncertainty instead of burying that evidence in a passing footnote.
Reference management without losing provenance
Move bibliographic data into a reference manager early, then verify it against the source itself. Add the page, table or section while taking notes. Put quotation marks around verbatim text immediately and separate it visually from your analysis. Every note should retain its source, precise location and intended use. A folder of PDFs is not a search record, and an automatically imported citation is not a checked reference.
Store results and notes only in ways permitted by licences and access conditions. Copyright and database terms continue to apply after a download. Personal data do not belong unnecessarily in public appendices. Transparency requires a describable research route; it does not require redistributing protected articles or confidential information.
When a search is provisionally complete
No ordinary dissertation search is literally exhaustive. Stop for stated reasons: important query combinations have run in suitable services; recent results repeat concepts already represented; reference-list and citation searches reveal no material new perspective; and a final update search has a date. Report the boundary. “Searching databases A and B through date X identified…” is more accurate than “all research proves…”.
If the question changes substantially, open a new log segment rather than silently rewriting the original one. This makes conceptual development visible and protects against presenting exploratory work as though it were pre-planned. Your reader can then judge both the evidence and the path by which it entered the argument.