How to Use AI Paper Finders and Disclose Them in Research

AI paper finders can change your literature search

An AI paper finder can save you hours. It can also quietly reshape your evidence base.

That second part matters.

If you ask an AI tool to find papers, suggest related work, screen abstracts, rank studies, or summarize a field, you have made a methods choice. You should record it. In many papers, you should disclose it.

This guide shows you how to use AI paper finders without losing the trail. It does not rank tools. It gives you a workflow that works for dissertations, journal articles, systematic reviews, scoping reviews, conference papers, and early literature mapping.

If you already keep a research tool log, this guide fits well with the template in AI tools for research: a disclosure log template. If you need a broader tool evaluation checklist before you start, read How to evaluate AI research tools before you use them in a paper.

What counts as an AI paper finder?

An AI paper finder is any system that helps you find research papers through machine learning, language models, citation graphs, embeddings, semantic search, or automated screening.

That can include tools that:

• answer a research question with cited papers
• search by meaning rather than exact keywords
• recommend related papers from a seed article
• cluster papers into themes
• extract study details from abstracts or PDFs
• help screen records for a review

Some tools focus on discovery. Some focus on systematic review workflows. Some sit on top of large scholarly indexes. For example, Semantic Scholar offers search and recommendation functions through its Academic Graph API, while other tools such as Elicit and Consensus describe workflows for finding, screening, extracting, or answering questions from research papers. (semanticscholar.org)

Do not treat these tools like a smarter search bar. Treat them like part of your search procedure.

Decide the role before you search

Before you open an AI paper finder, write down the role you will allow it to play.

Use one of these roles.

First, you can use it for exploration. You ask broad questions, collect terms, find seed papers, and learn the vocabulary of a field. This belongs early in a project. It may not need a full methods paragraph, but you should still record it.

Second, you can use it for supplementary discovery. You search formal databases first, then use an AI tool to find papers you missed. This affects your corpus. You should report it in the methods section if the final paper depends on the set of studies you found.

Third, you can use it for screening or extraction. The tool helps you include, exclude, code, summarize, or extract details from records. This belongs in your methods section and your [AI disclosure](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/).

Fourth, you can use it for writing support after you already selected the literature. That is a different disclosure problem. See How to disclose ChatGPT usage in academic papers if your main issue concerns drafting or editing.

One mistake causes most disclosure trouble: researchers decide the role after the tool has already shaped the corpus.

Do the opposite. Set the role first.

Record the search before the tool makes it look obvious

AI paper finders often make literature search feel conversational. You ask, the tool answers, you revise, and soon the path disappears.

Your methods section needs the path.

Record five things every time you run a serious AI-assisted paper search:

1. tool name and version if shown
2. date of search
3. exact prompt or query
4. filters, databases, disciplines, date ranges, or document types
5. number of records found, screened, exported, kept, and removed

For systematic reviews, this overlaps with reporting duties that already exist. PRISMA-S extends PRISMA for literature search reporting, and the PRISMA 2020 flow diagram tracks records identified, included, excluded, and reasons for exclusion. (prisma-statement.org)

That does not mean every AI-assisted paper needs a PRISMA flow diagram. It means the habit is useful.

You cannot reconstruct a search from memory three months later. Future you will be tired, annoyed, and missing one tab.

A simple workflow for AI-assisted paper discovery

Start with your research question in plain language.

Then write the formal version. Include population, setting, method, period, outcome, or theory when those parts matter. Do not make the AI tool guess your boundaries.

For example:

How have undergraduate students used generative AI tools during academic writing, and what evidence exists about changes in writing process, authorship, or feedback practices?

Now turn that into two search tracks.

Use the first track in standard databases such as your library catalog, Web of Science, Scopus, PubMed, ACM Digital Library, IEEE Xplore, ERIC, PsycINFO, or subject indexes in your field.

Use the second track in an AI paper finder. Ask for papers that match your scope. Ask for missing synonyms. Ask for highly cited older work and recent studies. Ask for papers that disagree with the main claim.

Then compare the outputs.

Do not simply merge everything. Mark the source of each record. If a paper came only from the AI tool, label it. If the AI tool rediscovered papers from your database search, mark that too.

This small step protects your methods section. It also helps you spot search bias. Some AI tools favor papers that appear in their index, have clean metadata, receive many citations, or match common wording in abstracts.

Example search log

A plain spreadsheet works. So does a text file.

Use rows like this:

You can add columns for DOI, abstract, exclusion reason, and whether you verified the citation in the publisher record.

If your paper involves a systematic review or meta-analysis, use a more structured log. The guide on AI disclosure in systematic reviews and meta-analyses explains what reviewers should report when AI tools affect search, screening, extraction, or synthesis.

Verify every paper before you cite it

AI paper finders can surface useful work. They can also misread abstracts, miss retractions, merge similar titles, or produce citations that look plausible.

Check every record outside the AI tool.

Open the publisher page, DOI record, PubMed page, arXiv page, institutional repository, or library record. Confirm the title, authors, year, venue, and whether the paper says what the tool claims.

If the tool gives a summary, treat it as a lead. Read the abstract. For papers that carry weight in your argument, read the full text.

This advice sounds boring because it is boring. It is also where many literature reviews go wrong.

A tool can help you find a paper. It cannot take responsibility for your citation.

When should you disclose an AI paper finder?

Use a simple test.

If the tool shaped what you read, included, excluded, coded, summarized, or cited, disclose it.

If you used it once to learn field vocabulary and then ran normal database searches, a short note in your research log may be enough. If you used it to identify papers that entered your review, mention it in the methods section. If you used it to screen or extract records, describe that process with care.

If you feel unsure, read Do I need to disclose AI usage in my paper?. The safest habit is to record more than you publish. You can shorten the disclosure later.

An [AI Usage Card](/ai-disclosure-for-social-science-research/) helps because it separates messy working notes from a clear public statement. The card can list the tool, task, human checks, and final role in the manuscript. For an overview, see What are AI Usage Cards?.

Methods wording you can adapt

Use direct language. Name the tool category, the task, and the human check.

For an exploratory literature review:

We used an AI-assisted paper discovery tool on 26 October 2026 to identify candidate literature on generative AI use in undergraduate writing instruction. The tool supported exploratory discovery and synonym generation. We verified all included records through publisher pages, DOI records, or library databases before citation.

For a supplementary search:

After database searches in ERIC and Scopus, we used an AI-assisted paper finder to identify additional candidate studies. We searched with the prompt "empirical studies on generative AI use in undergraduate writing courses, 2022-2026." Two authors screened the returned records. We included only studies that met the same criteria as the database search.

For screening support:

We used an AI-assisted review tool to support title and abstract screening. The authors set the inclusion criteria, reviewed all AI-suggested exclusions, and resolved uncertain records through manual review. We did not exclude any record solely because of an AI-generated recommendation.

For a systematic review, add counts. Your readers need to know how many records came from each source and how many survived screening. PRISMA-S asks authors to report search sources and search strategies for each database, register, website, or other source used in the review process. (jmla.pitt.edu)

Acknowledgment wording you can adapt

Some journals prefer an acknowledgment rather than a methods statement when AI helped with discovery but did not define the study design.

You could write:

The authors used an AI-assisted paper discovery tool to identify candidate literature during the early search phase. The authors verified all cited works and take responsibility for the final selection and interpretation of sources.

If the AI tool also summarized papers, say so:

The authors used an AI-assisted research tool to generate preliminary summaries of candidate papers. The authors checked these summaries against the original texts before using any information in the manuscript.

Avoid vague wording such as "AI was used for research assistance." That tells the reader almost nothing.

If your journal asks for a formal AI statement, pair the acknowledgment with an AI Usage Card. You can generate one at ai-cards.org and paste the text into your cover letter, appendix, supplementary material, or disclosure field.

LaTeX snippet for an AI paper finder disclosure

If you write in LaTeX, add a short disclosure near the acknowledgments or methods section. Adjust the title to match your venue.

\section*{AI assistance disclosure}
 
The authors used an AI-assisted paper discovery tool on 26 October 2026
to identify candidate literature related to the study topic. The tool was
used for exploratory search and supplementary discovery after database
searches. The authors verified all cited records through publisher pages,
DOI records, library databases, or full-text sources. No record was included
or excluded solely on the basis of an AI-generated recommendation.

If you use an AI Usage Card, you can point readers to it:

\section*{AI usage card}
 
An AI Usage Card describing the tool, task, prompts, human checks, and role
of AI assistance in this manuscript is included in the supplementary
materials.

For more placement examples, see the LaTeX tutorial for AI Usage Cards and the Overleaf guide.

What to put in an AI Usage Card

For an AI paper finder, your card should answer four questions.

What did you use?

Name the tool and the date. If the tool shows a model, database version, workflow name, or export format, record that too.

What did you ask it to do?

State whether the tool helped with discovery, query expansion, related paper search, screening, extraction, summarization, or writing.

What did humans check?

Say who verified records, who screened papers, who read full texts, and who made final inclusion decisions.

What changed in the paper?

Explain whether the tool affected the bibliography, the review corpus, the methods section, a table, or only early exploration.

You can draft this by hand, but the generator saves time. Use the free AI Usage Card generator, then edit the text so it matches your project.

Common mistakes to avoid

Do not cite papers you have only seen through an AI summary.

Do not mix AI-found papers with database-found papers without marking the source.

Do not let the tool invent your inclusion criteria while you search.

Do not write "we used AI" without naming the task.

Do not assume your field treats discovery, screening, and writing support the same way. A humanities essay, an HCI paper, and a medical systematic review may need different levels of detail. If you work near journal policy requirements, check AI transparency requirements for journal submissions before submission.

The pattern is simple: the closer the AI tool gets to your evidence base, the more you should disclose.

A short checklist before submission

Before you submit, open your search log and ask:

• Did an AI tool help find any cited paper?
• Did it suggest search terms or related work?
• Did it screen, rank, summarize, or extract records?
• Did a human verify each included record?
• Does the methods section explain the tool's role?
• Does the acknowledgment or disclosure statement match the actual workflow?
• Have you generated an AI Usage Card?

If the answer feels unclear, fix the record before you fix the prose.

Readers do not need a diary of every click. They need enough detail to judge how you found the literature and who made the decisions.

Generate an AI Usage Card at ai-cards.org before you submit. Use it as your disclosure note, your appendix text, or your private record for editors and reviewers.