An AI-Written Background Section Sounds Convincing but Has No Evidence Map

A background section can read smoothly and still leave you unable to answer the one question a supervisor or examiner will ask: which claim rests on which study, and how do you know? That is the gap this article addresses. If a draft paragraph flows well but you cannot trace each sentence back to a source you have actually read, you do not have a background section yet — you have prose waiting for an evidence map.

The fix is not to rewrite more elegantly. It is to rebuild the section around a small set of decisions: what the section must establish, which studies are eligible to support each claim, how you will document the search, and what "ready" looks like. Below, we walk through a concrete case — a student cleaning up a thesis background over winter break before a January project start — and show the artifacts that turn convincing prose into a defensible section.

Fluency is not evidence. A paragraph that reads confidently can hide three separate problems: claims with no source, sources that do not actually support the claim, and a search process you cannot describe if asked. Each of these is a rigor problem, not a writing-style problem.

NIH guidance frames rigorous research practice around careful assessment of prior knowledge — its strengths, weaknesses, and knowledge gaps — alongside transparent methods and reproducibility. A background section is exactly where that assessment becomes visible. If your section asserts a gap without showing how prior work was weighed, the rigor is not on the page.

Cochrane's guidance on searching and selecting studies points in the same direction: plan searches from explicit eligibility criteria, use multiple sources, combine free-text terms with subject headings, document the process, deduplicate records, and screen in stages. Those are process commitments, and a background section built on an undocumented search cannot demonstrate them.

PRISMA 2020 reinforces the reporting side: transparent reporting of search, selection, and synthesis methods, with checklists and flow-diagram templates. You may not be writing a systematic review, but the underlying expectation — that a reader can see how evidence was found and chosen — applies to any serious background section.

Before touching the prose, find out what you actually have. Read your draft with a pen and mark every sentence that makes a factual or gap claim. Then, for each mark, ask: can I point to a specific paper I have read that supports this?

You will usually land in one of three states:

• Fully mapped. Most claims trace to sources you have read and can describe. You need light cleanup.
• Partially mapped. Some claims trace to sources; others rest on memory, a secondary summary, or a sentence that sounds like it should be cited.
• Unmapped. The section reads well but you cannot reliably connect claims to studies. This is the case the title describes.

Be honest about which state you are in. The repair work differs sharply, and treating an unmapped section as "just needs polish" is how the problem survives into the next draft.

A background section is not a literature dump. It has a job: to establish the prior knowledge your project builds on, identify the gap, and justify why your question matters. Decide those three things explicitly before you search, because they determine what counts as relevant evidence.

Write a short scope statement for yourself:

1. The claim you must support. What does the reader need to accept as established?
2. The gap you are claiming. What is not yet known or not yet done well?
3. The boundary. Which populations, methods, or time frames are in and out of scope?

This scope statement becomes your eligibility criteria in miniature. Cochrane's advice to plan searches from eligibility criteria applies here directly: without criteria, you cannot tell a relevant study from a merely related one, and your search will drift.

The evidence map is the deliverable that makes the section defensible. It is a simple table, and it sits between your search and your prose. At minimum, give it these columns:

Two rules keep the map honest:

• One row per claim-source pair. If a claim needs two sources, it gets two rows.
• "What the source actually shows" is written from the source, not from your draft. This is where unmapped sections collapse: the draft says one thing, the paper says something narrower.

When a row's "how it supports the claim" cell is hard to fill, that is a signal the claim is too strong or the source is wrong. Fix it in the map, not in the prose.

With criteria and a map structure in place, search deliberately. Cochrane's recommendations give you a checklist: multiple sources, free-text terms combined with subject headings, documented process, deduplication, staged screening, and linking multiple reports of the same study.

In practice, for a thesis background section:

1. Draft your search strings from your scope statement, combining free-text terms with subject headings where the database supports them.
2. Search more than one source so you are not dependent on a single index.
3. Record what you searched, where, and when — a simple log is enough, and it is what lets you describe your method later.
4. Deduplicate before screening, so the same study does not appear twice under different records.
5. Screen in stages — title/abstract first, then full text — applying your criteria consistently.
6. Link multiple reports of one study so a single study is not counted as several independent findings.
7. Rerun searches before you finalize, so recent work is not missed.

If you want a scholarly search interface to run these queries, WisPaper search is built for finding academic papers, and WisPaper Scholar Search offers a public entry point for locating papers. Use them as search tools within your documented process — the log and the criteria are still yours to keep.

Screening is where most unmapped sections get their false confidence. It is tempting to keep a paper because it is about your topic. The question is narrower: does this paper support this specific claim?

Apply a three-way test to each candidate:

• Direct support. The paper reports the finding your claim states.
• Partial support. The paper is consistent with your claim but narrower, or in a different population or method.
• Background only. The paper motivates the topic but does not establish the claim.

Record the verdict in your evidence map. Claims that only have background support need to be softened or re-sourced. This is also where you catch the classic failure: a source cited for a strong claim that, on reading, only gestures at it.

For longer stretches of reading and synthesis, Scholar Agent supports literature review and related research tasks. Treat its output as material to check against the sources themselves, and log what you keep in your map.

An evidence map is only as good as the sources behind it. If your PDFs, notes, and citations are scattered, the map rots. Save the papers you actually screened into one place, tagged by the claim they support.

My Library lets you save or upload papers and organize them, which is the natural home for the sources in your map. Once papers are in your library, Library QA lets you ask questions based on papers you have uploaded or added from search results — useful for checking what a specific paper actually says before you write its row in the map.

If your project spans several chapters or a whole manuscript, Research Projects gives you a place to organize the work, so the background section's evidence does not live in a separate folder from everything else.

A background section fails quietly when a citation is wrong. A reference that looks plausible but does not exist, or that points to a different paper than you think, undermines every claim attached to it.

Before a citation goes into the draft, confirm it against a real record. TrueCite screens BibTeX citations against real academic databases to flag hallucinated references — a concrete check to run on your bibliography before submission. Pair it with your own read of the source: TrueCite checks the reference, while your evidence map checks the claim.

Now the prose follows the evidence, not the reverse. Work through your map row by row and write the sentence each row supports. Where a claim has direct support, state it. Where support is partial, narrow the claim to match. Where a claim has no row, either find a source or cut the claim.

A few practical rules:

• One claim, one clear source. Avoid stacking citations on a sentence unless each genuinely supports it.
• Let the gap claim carry its weight. Your gap statement should be traceable to what the sources do not cover, not asserted on top of them.
• Keep the map beside the draft. When a reviewer questions a sentence, you go straight to the row.

This is also where the section starts to read like rigorous work rather than polished summary: the prior knowledge, its strengths and weaknesses, and the gap are all visible and traceable.

The section is ready when you can pass a simple audit:

• Every factual sentence has a row. No orphan claims.
• Every row has a source you have read. Not a summary of a summary.
• Every "how it supports" cell is filled. You can explain the link in one sentence.
• Your search is describable. You can state your sources, terms, and screening steps, consistent with Cochrane's documentation expectations.
• Your citations check out. References resolve to real records.
• Your reporting is transparent. A reader can see how evidence was found, selected, and synthesized, in the spirit of PRISMA 2020.

If any of these fail, you know exactly which rows to fix. That is the difference between a section that merely sounds convincing and one that holds up.

In the winter-break case, the student ends January with three things: a background section whose every claim traces to a mapped source, an evidence map that doubles as a defense document, and a search log that describes how the literature was found and screened. The prose may read similarly to before — but now it can survive a question.

The transferable lessons are the ones worth carrying into your next chapter or manuscript:

• Scope before search. Criteria make relevance decidable.
• Map before prose. The table is the thinking; the paragraph is the output.
• Document as you go. A search log is cheap to keep and expensive to reconstruct.
• Verify references and claims separately. One check protects the citation; the other protects the argument.

Your concrete next action: open your current background draft, mark every factual and gap claim, and build the first version of the evidence map — even if half the rows are empty. The empty rows are your to-do list, and they are far more useful than a paragraph that only sounds convincing.