How to Evaluate Sources for a Literature Review: A 5-Filter Method

The literature review is a filter, not a summary

How to evaluate sources for a literature review is a question best asked before you have read too many papers. The technique matters more when it runs before the reading. A literature review is a filter: from the field’s total output, the candidate selects the sources that will do work in their thesis. The five filters below run in order, and each removes sources that would otherwise reach the argument.

The point is not to read less. It is to read the right things thoroughly and to know why each source is in the reference list by the time the thesis reaches the panel. Trafford and Leshem (2008) describe examiners as testing not just whether a candidate knows their sources, but why each source is there. The filters below are the shape of that answer.

Filter 1: Relevance to your research question

Every source should have a specific reason for being in the literature review. If a paper answers a question adjacent to yours but not the specific one you asked, it may still be relevant, but the reason has to be nameable in one sentence. If you cannot say why the source is here, it is not here yet.

Write the reason next to the source in your notes. “Kiley 2009 is here because it names the threshold concepts that structure the doctoral experience I am studying.” “Mullins & Kiley 2002 is here because it documents examiner behaviour in the way I need to reference for chapter three.” If you cannot write the sentence, the source may be interesting but not yet relevant to this thesis.

Grant and Osanloo (2014) describe this as the “blueprint” step of a theoretical framework: every source in the literature review should serve a specific structural role in the argument you are building. Sources with no role are decorations, and decorations do not survive the viva.

Filter 2: Provenance and venue

Once relevance is settled, check where the source was published. A peer-reviewed journal, a university-press book, or a well-established scholarly volume carries different weight from a working paper, a preprint, or an unpublished report. None of the second category is automatically wrong; each has to be treated as what it is.

For journal sources, check the DOAJ if the journal is open access. Check the editorial board if the journal is unfamiliar. Check Crossref for a registered DOI. Predatory journals now number more than 20,000 titles (Cabells, 2026), and the ambient risk of citing one has grown accordingly. A single predatory citation in a doctoral bibliography changes the tone of a viva (Park, 2003).

For books, check the publisher. Academic presses (Oxford, Cambridge, Routledge, Sage, MIT, etc.) run editorial review. Self-published books, print-on-demand titles, and vanity-press works carry less evidentiary weight and need to be treated as opinion rather than as scholarship.

Filter 3: Verification (does the source exist as cited)

This filter matters more now than it did five years ago. AI-assisted drafting introduces fabricated citations into bibliographies at meaningful rates: Bhattacharyya and colleagues (2023) found 47% of ChatGPT-generated medical references completely fabricated; Walters and Wilder (2023) found 18 to 55% depending on the model. If any of your candidate sources were suggested by a language model, they need verification before they enter the reference list.

The check is fast: resolve the DOI at https://doi.org/[DOI]. Match the returned metadata (authors, title, journal, year, pages) against your citation. Check Retraction Watch for retraction status. If the DOI does not resolve, the citation is wrong until proven otherwise. Thesisroom’s citation guard runs the check in one pass, drawing on ten registries, which is the practical way to run verification at bibliography scale.

Filter 4: Currency and the last three years

Examiners check whether the literature review reflects the current state of the field. Park (2003) documents currency as one of the specific things panellists look for. A thesis submitted in 2026 whose reference list stops in 2022 signals that the writing was not kept live during examination preparation.

Currency does not mean “throw out older sources.” A foundational paper from 2002 or a canonical monograph from 1975 may be essential to the argument. It means “have the recent literature covered.” If a search of your topic in Google Scholar for the last three years turns up papers you do not know, you have work to do before submission.

The Source Finder is built for this: enter a claim, and it returns verified papers that support it, filtered by recency if that matters for the argument. It is a research aid, not a source picker; the writer still decides what to cite.

Filter 5: How the source will actually be used

The last filter is functional. For every source that has passed the first four, write a sentence saying what the source will do in your thesis. Options are limited: it establishes context, defines a concept, provides a methodological precedent, supports a specific empirical claim, offers a counterpoint you will address, or reports a finding your work builds on.

If a source does not fit any of these functions, it is not being used, and it should not be in the reference list. This filter catches the padded citations that survive the first four filters, because a source can be relevant, verified, current, and legitimate, and still not actually be doing anything in your argument.

Kamler and Thomson (2006) describe writing a literature review as text-and-identity work: the candidate is not just summarising sources but building a position in relation to them. Every source in the reference list should be part of the position, not part of the scenery.

Running the filters as a habit

Applying five filters to every source you consider sounds expensive. It is cheaper than the alternative, which is dragging weak citations through drafts for two years and having to prune them under submission pressure. The filters run fastest as a habit. Relevance and use are decided when the paper is first noted. Provenance and verification happen at the point of saving to a reference manager. Currency and use are checked again at the writing stage.

The Thesisroom method treats the whole degree as the seat where this habit lives. Sources filtered on day one are the same ones that will not surface a fabrication or a retraction in the viva.

What Thesisroom does with this

Thesisroom’s citation guard runs the automatic parts of the filter set (provenance, verification, retraction, currency) at bibliography scale in one pass. Source Finder helps with the “find a stronger source” step by returning verified papers that support a claim you specify. The Structural Critic reviews how sources are used in the argument, flagging citations that appear structurally isolated or unsupported by the surrounding text.

None of the filters remove the reading. What they remove is the accumulation of weak sources that would otherwise reach the bibliography.

Frequently asked questions

How many sources should I evaluate for a literature review?

There is no fixed number. Most doctoral literature reviews consider several hundred sources during the reading stage and cite between 150 and 400 in the final thesis. The exact number depends on the field and the thesis type (Paltridge, 2002). What matters is that every cited source has passed the five filters.

Should I use Google Scholar, OpenAlex, or Semantic Scholar?

Use more than one. Google Scholar has the broadest coverage but the lowest precision; OpenAlex is the largest open academic index and includes richer metadata; Semantic Scholar adds AI-derived citation-network features and is strong for computer science and biomedicine. Cross-checking against two increases coverage and reduces the risk of missing key work.

Can I cite a paper I have only read the abstract of?

Ideally no. Citing from abstracts alone is a common source of miscitation, because abstracts often omit the qualifications and boundary conditions the paper carefully sets out. If a citation is doing load-bearing work in your argument, read the full paper. If it is background context, an abstract-based citation is defensible but note the source.

How current does my literature review need to be?

Cover the last three years of relevant work at minimum. Examiners check this specifically (Park, 2003). A gap in the recent literature signals that the review was written and then not updated during the writing years. Foundational sources from further back are fine and often necessary.

What if a source I cited turns out to be wrong or retracted?

Update the citation. Retracted papers should be marked as retracted in the reference list, following COPE guidance. Miscited papers should be re-verified and either corrected or replaced. This is why running the verification during the literature review rather than the week of submission matters: you have time to respond.

Can Thesisroom find sources for me?

Find Sources returns verified papers that support a claim you specify. It never chooses a source for you. It returns database-native results (from Crossref, OpenAlex, Semantic Scholar), and the writer decides which ones to read and cite. The point is to make the search deterministic and traceable, not to substitute for the writer’s judgement.

One takeaway before you close this tab

The literature review that survives a viva is the one built by filter, not by accumulation. Five checks (relevance, provenance, verification, currency, use) run in order remove most weak sources before they anchor an argument. When you want the automatic parts run across your whole bibliography at once, Thesisroom’s citation guard does the verification pass in one go.

References

Bhattacharyya, M., Miller, V. M., Bhattacharyya, D., & Miller, L. E. (2023). High rates of fabricated and inaccurate references in ChatGPT-generated medical content. Cureus, 15(5).

Cabells. (2026). Predatory Reports database.

Grant, C., & Osanloo, A. (2014). Understanding, Selecting, and Integrating a Theoretical Framework in Dissertation Research: Creating the Blueprint for Your ‘House’. Administrative Issues Journal, 4(2), 12-26.

Kamler, B., & Thomson, P. (2006). Helping Doctoral Students Write: Pedagogies for Supervision. Routledge.

Paltridge, B. (2002). Thesis and dissertation writing: an examination of published advice and actual practice. English for Specific Purposes, 21, 125-143.

Park, C. (2003). Levelling the Playing Field: Towards Best Practice in the Doctoral Viva. Higher Education Review, 36(1), 47-67.

Trafford, V., & Leshem, S. (2008). Stepping Stones to Achieving your Doctorate. Open University Press.

Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13.