AI Disclosure in Systematic Reviews and Meta-Analyses

AI use in evidence synthesis needs a paper trail

Systematic reviews and meta-analyses depend on trust.

Readers need to know how you searched, screened, extracted, and analyzed evidence. If you used AI at any step, that fact belongs in your methods and disclosure record. Otherwise, readers cannot judge what the tool did, where errors may have entered, or how you checked its output.

That does not mean you should avoid AI.

It means you should document it with care.

If you want a fast way to do that, create an AI Usage Card and attach or adapt it for your review project.

Why systematic reviews need stricter AI disclosure

Evidence synthesis follows structured methods.

You predefine search strategies. You set inclusion criteria. You track screening decisions. You explain extraction rules. You justify statistical choices. That culture of traceability makes AI disclosure a natural fit.

AI can affect a review in ways that are hard to spot later.

A tool might rewrite search strings, rank abstracts, suggest exclusions, summarize trial outcomes, extract effect sizes, draft risk-of-bias notes, or help write the manuscript. Each use can shape the final review. Even small changes matter when the goal is reproducibility.

This is why generic statements like “we used AI for writing assistance” often fail.

In a systematic review, readers need to know where AI entered the workflow, what model or tool you used, what inputs you gave it, what outputs you accepted, and how humans checked those outputs.

For a broader foundation, see Why AI Transparency Matters in Research and AI Ethics and Documentation in Academic Research.

The key question is not whether you used AI

The key question is where you used it.

Many researchers think disclosure only applies when AI writes text. That view is too narrow for systematic reviews. In evidence synthesis, AI can influence decisions long before the manuscript stage.

A better approach is to map AI use across the full review pipeline.

Ask five direct questions:

Did AI help plan the review?

Some teams use chatbots to refine review questions, define PICO elements, or suggest keywords and databases.

That use deserves disclosure because it can shape the scope of the review from the start.

Did AI affect the search process?

This includes query expansion, synonym generation, translation of search terms, and ranking or filtering records before human screening.

Search transparency sits at the core of systematic review quality. If AI touched the search strategy, say so.

Did AI affect screening or eligibility decisions?

This is one of the most sensitive steps.

If a tool prioritized records, flagged likely exclusions, or produced screening recommendations, readers need to know whether humans reviewed every decision or only a subset.

Did AI assist with extraction, appraisal, or synthesis?

AI tools can pull sample sizes, interventions, outcomes, confidence intervals, and study limitations from PDFs. They can also draft evidence tables or summarize findings.

These outputs save time. They also introduce extraction errors, hallucinated values, and loss of nuance. Disclosure should explain how you verified every extracted item.

Did AI help write or edit the paper?

This includes drafting prose, rewriting text, summarizing sections, formatting references, and polishing language.

This use still needs disclosure, but it should not overshadow earlier uses that may have had a larger methodological impact.

What to disclose in a systematic review

Your disclosure should let another researcher understand the role of AI without guessing.

Focus on six facts.

First, name the tool and version if known. “ChatGPT” alone is weak. “ChatGPT, GPT-4.1, accessed via web interface in January 2026” tells readers more.

Second, state the task. Say whether the tool helped with search term generation, title and abstract screening, data extraction, coding study characteristics, narrative synthesis, or writing support.

Third, describe the inputs in plain terms. You do not need to paste every prompt into the manuscript, but you should summarize what you provided to the model. For example, “we supplied abstracts and asked the model to suggest likely inclusion decisions based on predefined eligibility criteria.”

Fourth, explain the level of human oversight. Did two reviewers check all AI-assisted screening suggestions? Did one reviewer verify every extracted number against the source PDF? This point matters more than marketing claims about the tool.

Fifth, state the limits you imposed. For example, “AI outputs did not determine final inclusion decisions” or “we did not upload full PDFs that contained licensed content to public tools.”

Sixth, record where the disclosure appears. In many reviews, the best places are the Methods section, an appendix, the acknowledgments if relevant, and a separate AI Usage Card.

If you are unsure whether your use crosses the line into reportable assistance, read Do I Need to Disclose AI Usage in My Paper?.

A simple disclosure structure that works

You do not need a long statement.

You need a clear one.

Here is a practical structure that fits most systematic reviews:

1. Tool
2. Task
3. Input
4. Human verification
5. Limits on use

That structure works in manuscripts, protocols, supplements, and internal lab records.

Here is a short example:

We used GPT-4.1 through the ChatGPT web interface to suggest search term variants for predefined PICO concepts and to draft initial summaries of included studies. Two authors reviewed all suggested search terms before database submission. One author checked all AI-generated study summaries against the source articles and corrected errors before synthesis. AI outputs did not determine inclusion, exclusion, risk-of-bias judgments, or final effect estimates.

That statement tells readers what they need to know.

For more sample wording, see AI Usage Cards Examples and Templates.

Where researchers make mistakes

The biggest mistake is vague disclosure.

Saying “AI was used during manuscript preparation” hides methodological uses that may matter far more than writing support. If AI touched screening or extraction, disclose that directly.

The second mistake is treating AI suggestions as if they were neutral.

They are not. Tools can inherit biases from training data, mishandle domain terms, miss negation, and confuse outcome measures. In medicine and public health, a single extraction error can change a pooled estimate or alter a conclusion about harm.

The third mistake is failing to keep records.

If you used AI during screening or extraction, save prompts, outputs, dates, and notes on corrections. You may need them during peer review, protocol updates, or team audits.

The fourth mistake is mixing human and AI decisions without a boundary.

Readers should know what the tool suggested and what the researchers decided.

A practical workflow for transparent AI use

Use AI only after you define your review protocol.

That rule protects the logic of the review. It also makes disclosure easier because you can compare AI-assisted steps against preplanned methods.

Next, log each use as it happens.

A simple table is enough. Record the date, tool, task, input type, output type, reviewer who checked it, and final action taken. This record becomes the raw material for your methods section and AI Usage Card.

Then verify every output against source material.

Do not trust extraction, summary, or screening suggestions without human review. In most review contexts, AI should assist triage and drafting, not replace judgment.

Finally, turn your notes into a standard disclosure.

That is where an AI Usage Card helps. It gives you one place to describe the tool, purpose, limitations, and oversight.

Example wording for a methods section

You can adapt this language to your own review:

\paragraph{Use of AI tools.}
The review team used GPT-4.1 via the ChatGPT web interface for two limited tasks:
(1) generating candidate search term variants based on predefined PICO concepts and
(2) drafting preliminary narrative summaries of included studies.
All search strings were reviewed and finalized by the authors before submission to databases.
AI outputs were not used to make final screening, eligibility, risk-of-bias, or meta-analytic decisions.
All AI-assisted summaries and extracted details were checked against the original articles by a human reviewer before inclusion in the manuscript.

If you want to include a shorter disclosure note in a supplement, try this:

\begin{quote}
AI assistance disclosure: We used GPT-4.1 for search term brainstorming and draft study summaries.
Human reviewers verified all outputs against the protocol and source articles.
The tool did not make final review decisions.
\end{quote}

If you write in LaTeX, our LaTeX tutorial for AI Usage Cards shows how to format disclosures cleanly.

How AI disclosure fits with PRISMA-style reporting

PRISMA asks researchers to report methods clearly enough for readers to assess the review process.

AI disclosure supports that goal.

It does not replace PRISMA items on search strategy, selection process, data collection, or synthesis methods. It adds detail about whether software with generative or predictive features influenced those steps.

In practice, AI disclosure often belongs next to the method it affected.

If AI helped build search strings, mention it in the search methods.

If AI ranked records for screening, mention it in the selection process.

If AI extracted data, mention it in the data collection section.

If AI drafted text, mention it in author notes, acknowledgments, or a dedicated disclosure statement.

That placement makes the paper easier to read and easier to audit.

Editors and reviewers will ask sharper questions

Journal policies still vary, but expectations are moving in one direction.

Editors want honest, specific disclosure. Reviewers want to know whether AI affected judgment-heavy steps. If your review includes health, policy, education, or legal evidence, scrutiny may be even higher because downstream decisions can affect real people.

A strong disclosure lowers friction.

It signals that your team tracked AI use, imposed limits, and kept human responsibility where it belongs. That helps editors assess the manuscript and helps readers trust the work.

For journal-specific context, see AI Disclosure Policies by Major Journals.

The best time to document AI use is now

Do not wait until submission day.

By then, you may forget where AI entered the workflow or how much you relied on it. Good disclosure starts during the review, not after it.

Systematic reviews already demand careful records. AI use should join that record.

If you used AI to brainstorm search terms, prioritize records, extract study details, summarize findings, or polish prose, document it while the trail is fresh. Then turn that record into a clear statement that readers and editors can trust.

Create your AI Usage Card now and give your systematic review the transparency that evidence synthesis demands.