12 Innovative Research Methods and What the Evidence Shows

Published August 10, 2024  |  Last updated October 31, 2026

What Are Innovative Research Methods?

Innovative research methods are techniques or designs that are newer than standard practice in a field. Some change how evidence is found, some change how causal questions are answered without a trial, and some change how data are collected. The table pairs each of the twelve methods below with a check to make first, drawn from the cited source.

Swipe sideways to see every column.

12 Innovative Research Methods and the Evidence for Each

Each entry says what the method is and what a published source reports about it. Under each entry, “When it fits” and “How to start” restate what the cited source reports or recommends; where a step is our own suggestion, it says “We suggest”. Most entries rest on the abstract of the cited paper, not the full text.

1. AI-Assisted Literature Screening

Large language models can be prompted to apply a review’s eligibility criteria to titles and abstracts. A GPT-4o prompt set built for a living systematic review reached 100% sensitivity for the studies that were finally included after full-text screening. Sensitivity for records that humans included at the title and abstract stage varied from 58% to 100% across updates, and the authors simulated workload reductions of 65% to 85%.² The worked example below shows what two other models did on a different review.¹

When it fits: Evidence-synthesis projects with many records and human screening decisions to test against; the study used the decisions from a living review as its reference standard.² How to start: Split your records into a development set and a test set, refine the prompts on the first, and report sensitivity against human decisions on the second, as the study did.²

2. Living Systematic Reviews

A living systematic review is continually updated, incorporating relevant new evidence as it becomes available. Its authors present it as a response to the problem that systematic reviews are difficult to keep up to date, and that failure to do so decays their currency, accuracy and utility.³

When it fits: Where research evidence is emerging rapidly, current evidence is uncertain, and new research may change policy or practice decisions.³ How to start: Define how new evidence will be identified and added as it becomes available, because continual updating is what the definition requires.³

3. Target Trial Emulation

Target trial emulation uses observational data to answer causal questions about the effects of interventions by explicitly emulating a hypothetical pragmatic randomised trial, when a real trial is unavailable or not feasible. The 2025 TARGET statement gives a 21-item checklist in six sections. It asks authors to specify the target trial protocol (the causal estimand, identifying assumptions and data analysis plan), to show how it was mapped to the observational data, and to report sensitivity analyses.⁴

When it fits: When the question is the comparative effect of an intervention and a randomised trial is unavailable or not feasible. TARGET covers emulations of a parallel-group, individually randomised target trial with adjustment for baseline confounders, so check that your design falls in that scope.⁴ How to start: Write the target trial protocol before you analyse the data, state how each element maps onto the data you have, and plan the sensitivity analyses the checklist asks you to report.⁴

4. Platform and Master Protocol Trials

A review of trial design in precision medicine describes basket and umbrella trials, master platform trials and N-of-1 studies as designs that are beginning to supplant standard phase I, II and III protocols, allowing faster evaluation of treatments.⁵

When it fits: Precision oncology, where treatments are matched to molecular features; the review is written from that setting and reports that some, but not all, novel trials show matched therapy correlating with better outcomes than non-matched therapy.⁵ How to start: Read the review’s account of each design’s advantages, limitations and challenges before you choose one, and confirm that the design suits your field and your regulator.⁵

5. Citizen Science

A 2025 review of habitat monitoring summarised a survey of 458 volunteer biodiversity recorders. Existing recorders were willing to consider habitat recording, but many lacked confidence and some lacked motivation. The authors built a six-question design framework addressing the interplay between volunteer motivation and confidence, and data accuracy. It covers clarity of purpose, cost to volunteers, protocol complexity and scale of sampling, and this affects the training and support volunteers need.⁶

When it fits: When you need observation or habitat data over a range and scale that professional surveys alone do not reach; the review frames citizen science as a complement to professional surveys and earth observation.⁶ How to start: Answer the six framework questions for your own project before you recruit, and look for accuracy evidence on your protocol, because the abstract does not report accuracy figures.⁶

6. Machine-Learning-Assisted Citizen Science

A review of palaeontology and archaeology projects proposes a four-phase workflow for combining machine learning with volunteer contributions: preparation, execution, implementation and reiteration. It notes that gathering data on objects such as fossils and artefacts has different requirements from gathering observations in biodiversity monitoring.⁷

When it fits: Large collections of fossils or artefacts, where the review says citizen science and machine learning can be mutually beneficial.⁷ How to start: Use the four phases as a project plan and allocate tasks to roles across citizen-science tasks, machine-learning development, research, stakeholder engagement and app or infrastructure development.⁷

7. Drones for Field Data

A scoping review identified 136 studies published since 2010 that used drones to study the ecophysiology of wild animals, with the focal species biased toward marine mammals, particularly cetaceans.⁸ A 2025 review of drone-based monitoring of harmful algal blooms lists validation, regulatory gaps and integration with health risk frameworks among the persistent challenges.⁹

When it fits: Non-invasive sampling of wild animals, where body condition and morphometric data from standard colour sensors predominated,⁸ and algal-bloom monitoring as a complement to satellite and in situ methods.⁹ How to start: Check the flight regulations that apply, which the algal-bloom review lists among its considerations, and plan how you will validate the sensor data, since validation is among the persistent challenges it names.⁹

8. Virtual Reality Experiments

A review of virtual reality in episodic memory research notes that the literature is spread across several disciplines and uses distinct apparatus, which makes it hard to tell whether differences between studies reflect genuine theory or methodological variation.¹⁰

When it fits: Studying learning and episodic memory in immersive environments, including how immersion affects memory and whether effects transfer from virtual reality to the real world.¹⁰ How to start: Record the hardware and software, because the review finds that apparatus varies between studies. We suggest adding a real-world comparison condition, since the review notes how little is known about how learning in VR compares with real life.¹⁰

9. Consumer Wearables

The INTERLIVE expert statement on consumer wearable and smartphone step counts says the validity of many consumer devices has not been rigorously and transparently evaluated, and it offers a checklist for validation studies.¹¹

When it fits: Measuring physical activity through step counts with consumer wearable or smartphone devices.¹¹ How to start: Use the statement’s six domains (target population, criterion measure, index measure, validation conditions, data processing and statistical analysis) to design and report a validation study; it offers a basic and an advanced protocol for step counters.¹¹ We suggest checking the validation evidence for your exact device and outcome before relying on it.

10. Text Mining of Large Text Sources

A systematic review of computational text analysis in intimate partner violence research found 22 studies across eight databases. Text mining can give researchers access to datasets, for example from social media or from organisations, that are too large to analyse manually.¹²

When it fits: Datasets from social media or from organisations that are too large to analyse manually; the review covers one field, so check that its evidence transfers to yours.¹² How to start: Choose a supervised or unsupervised approach (the included studies used rule-based classification, machine learning, deep learning and topic modelling) and evaluate it on a held-out labelled test set or with k-fold cross-validation, as most included studies did, reporting accuracy and F1.¹²

11. AI-Assisted Qualitative Analysis

A 2026 scoping review of 75 studies found large language models used across qualitative research, most often for coding assistance (43 studies) and theme identification (41), with thematic analysis the predominant method (38). Agreement between the models and human coders ranged from 36% to 99%.¹³ A 2026 rapid review of 42 health-research studies found agreement often above 80% for descriptive themes but about 30% for culturally nuanced themes, and reported occasional inaccuracies including fabricated quotes.¹⁴

When it fits: Early-stage work on large volumes of text, such as familiarisation, coding and theme development, where performance was strongest in inductive thematic and content analysis; it declined for reflexive and theory-driven analyses.¹⁴ How to start: Keep a human responsible for interpretation, and check each quote against the source text. Report the model, how it was deployed (API, web interface or local), the temperature, the context length and the prompts: in the scoping review only 13 of 75 studies reported temperature and 75% reported no parameter settings at all.¹³

12. Social Media Analysis

A 2026 scoping review of 136 studies of e-cigarette content on social media found that manual coding predominated, with smaller shares using computational or hybrid approaches. Computational approaches were generally applied to larger datasets and more consistently reported validation, and AI applications were concentrated in topic discovery and sentiment assessment.¹⁵ A 2026 scoping review of 50 studies of HPV vaccine content across eight platforms, with X most often studied, found content analysis, sentiment analysis, topic modelling and network analysis in use, and concluded that the discourse cannot be adequately understood through sentiment polarity alone.¹⁶

When it fits: Mapping public discussion of a topic across platforms; both reviews come from health communication, so check that the evidence transfers to your field.¹⁶ How to start: Define the platforms and the sampling first, decide whether manual, computational or hybrid coding suits the size of your dataset, and report how any computational output was validated, a gap the e-cigarette review highlights.¹⁵ Code themes or content as well as sentiment polarity.¹⁶

A Real Example: Testing AI Screening on 984 Abstracts

A Spanish-led team tested whether two language models could screen titles and abstracts for a biomedical systematic review. The paper is open access under CC BY 4.0, so we re-plotted its counts and recomputed the derived numbers in R.¹ The search of Medline, Embase and Scopus in October 2022 left 1,081 records after duplicates were removed. Human reviewers, working in pairs with a third to resolve disagreements, were the reference standard. Abstracts were missing for 97 records, so the models assessed 984, of which humans had included 80 and excluded 904.

Llama 3 70B and GPT-4o mini were each run in three configurations that differed only in temperature (0, 0.5 and 1), with the same prompts. The best Llama 3 configuration, LLA_2, had 77.5% sensitivity, 91.4% specificity and 90.2% accuracy. The best GPT-4o mini configuration, GPT_2 (called CHAT_2 in the paper’s text), had 56.2% sensitivity, 95.1% specificity and 92.0% accuracy.¹

Accuracy hides the problem. Because 904 of the 984 records were irrelevant, a configuration can reach about 92% accuracy while dropping 35 of the 80 relevant records, which is what every GPT-4o mini configuration did. The Llama 3 configurations missed 18 to 20.

Swipe sideways to see every column.

What this shows: on one review, two general-purpose models agreed with human screening on about nine in ten abstracts, found most relevant records in the Llama 3 runs, and left a human to read roughly 9% to 14% of the abstracts. What it does not show: results for other topics or newer models, how much time was saved, or that a pipeline without a human check is safe. The authors state that manual validation remains necessary to address false positives and negatives.¹

How to Choose an Innovative Research Method

Use the same sequence for any method in the list.

1. 1Start from the question. Write the question and the decision the answer will support. A method that does not serve it is a distraction.
2. 2Find validation evidence in your field. Look for a published test of the method on a question like yours, not on a different discipline.
3. 3Pilot against a trusted standard. Run the method on a small sample where you already know the answer, as the 984-abstract study did with human screening.
4. 4Pre-specify the protocol and reporting guideline. Fix the design, outcome and reporting guideline before you collect or analyse data.
5. 5Report enough to repeat it. Name the model, device or platform, its settings and the prompts or protocol. The screening study published its prompts and configurations so others can reuse them.¹

Need the Analysis Run or Checked? Consult Editverse

Editverse subject-matter experts work to COPE, ICMJE and GPP 2022 guidance, with institutional and country-specific ethics followed.

Editverse

Methodology Consultation

Not sure which method fits your question and your data?

You send: your research question, data and planned design.

You get: three 45-minute sessions with PhD methodologists, a design review with recommendations and a written methodology framework document.

When: You are at proposal stage or facing methodology concerns.

Methodology Consultation Package · $400

Get my method reviewed →

Editverse

Literature Review Support

Need a search strategy and gap analysis for your review?

You send: your topic and the databases you need covered.

You get: a search strategy document, a literature matrix in Excel, a gap analysis report and an annotated bibliography. You write the review from them.

When: You need a defensible search and synthesis foundation.

Literature Review Support Package · $500

Support my review →

Editverse

Statistical Analysis

Need the analysis run in the software your journal expects?

You send: your data, research questions and preferred software.

You get: annotated code or syntax, complete output files, an output interpretation guide and publication-ready tables.

When: Your data are collected and the analysis is pending.

SPSS/R/STATA Statistical Analysis · $400

Run my analysis →

Frequently Asked Questions

What are innovative research methods?

They are approaches that are newer than standard practice in your field, such as AI-assisted screening of the literature, AI-assisted qualitative analysis, living systematic reviews, target trial emulation, platform trials, citizen science, drones, virtual reality, consumer wearables, text mining and social media analysis. Newer does not mean validated: check the published evidence for the method in your own field before you commit.

Which innovative research method is best for my study?

The one that answers your question with evidence you can defend. Start from the question and the decision it supports, then look for validation studies of the method in your field. The at-a-glance table above pairs each method with a check to make first, drawn from the cited source.

Can AI replace human reviewers when screening abstracts?

Not on the evidence we reviewed. In one published test of 984 abstracts, the best Llama 3 configuration had 77.5% sensitivity and the best GPT-4o mini configuration 56.2%, against human reviewers. The authors concluded that manual validation remains necessary so that important studies are not overlooked.

What is target trial emulation?

It is the use of observational data to estimate causal effects by explicitly emulating a hypothetical pragmatic randomised trial, used when randomised trials are unavailable or not feasible. The 2025 TARGET statement gives a 21-item reporting checklist in six sections.

What is a living systematic review?

A systematic review that is continually updated, incorporating relevant new evidence as it becomes available. Its proponents suggest it may be particularly important where evidence is emerging rapidly, is uncertain, or may change policy or practice.

Are consumer wearables valid for research?

It depends on the device and the measure. The INTERLIVE expert statement says the validity of many consumer wearable and smartphone devices has not been rigorously and transparently evaluated for step count, and provides a checklist for validation studies. Check the validation evidence for the device and outcome you plan to use.

Can citizen science data be used in published research?

The evidence depends on the project design. A 2025 review of habitat recording builds a six-question design framework linking volunteer motivation, confidence and data accuracy, which affects the training and support volunteers need. Check the accuracy evidence for your own protocol before you rely on volunteer data.

How do I report a study that uses a new method?

Use the reporting guideline that matches your design and describe the method in enough detail to repeat it. For an emulated trial that is the TARGET checklist. For AI screening, report the model, its settings and the prompts, as the 984-abstract study did. A 2026 scoping review of 75 LLM-assisted qualitative studies found that 75% reported no model parameter settings at all.

References

1. López-Pineda A, Nouni-García R, Carbonell-Soliva Á, Gil-Guillén VF, Carratalá-Munuera C, Borrás F. Validation of large language models (Llama 3 and ChatGPT-4o mini) for title and abstract screening in biomedical systematic reviews. Res Synth Methods. 2025;16(4):620-630. Open access, CC BY 4.0; interval, accuracy and reading-load figures recomputed by Editverse in R. doi.org/10.1017/rsm.2025.15
2. Homiar A, Thomas J, Ostinelli EG, et al. Development and evaluation of prompts for a large language model to screen titles and abstracts in a living systematic review. BMJ Ment Health. 2025;28(1). doi:10.1136/bmjment-2025-301762. pubmed.ncbi.nlm.nih.gov/40701625/
3. Elliott JH, Synnot A, Turner T, et al. Living systematic review: 1. Introduction: the why, what, when, and how. J Clin Epidemiol. 2017;91:23-30. doi.org/10.1016/j.jclinepi.2017.08.010
4. Cashin AG, Hansford HJ, Hernán MA, et al. Transparent reporting of observational studies emulating a target trial: the TARGET statement. BMJ. 2025;390:e087179. doi:10.1136/bmj-2025-087179. pubmed.ncbi.nlm.nih.gov/40903028/
5. Fountzilas E, Tsimberidou AM, Vo HH, Kurzrock R. Clinical trial design in the era of precision medicine. Genome Med. 2022;14(1):101. doi.org/10.1186/s13073-022-01102-1
6. Barnes AE, Pocock MJO, Harris MM, Newton N, Robinson RA. Not just species recording: the potential of citizen science for habitat monitoring. Landsc Ecol. 2025;40(9):181. doi.org/10.1007/s10980-025-02155-4
7. Eijkelboom I, Schulp AS, Amkreutz L, et al. Making sense of fossils and artefacts: a review of best practices for the design of a successful workflow for machine learning-assisted citizen science projects. PeerJ. 2025;13:e18927. doi:10.7717/peerj.18927. pubmed.ncbi.nlm.nih.gov/39959835/
8. Yaney-Keller A, McIntosh RR, Clarke RH, Reina RD. Closing the air gap: the use of drones for studying wildlife ecophysiology. Biol Rev Camb Philos Soc. 2025;100(3):1206-1228. doi:10.1111/brv.13181. pubmed.ncbi.nlm.nih.gov/39822117/
9. Byrd K, Wu J, Lee J. Harmful algal bloom monitoring with unmanned aerial vehicles: tools, challenges, and public health implications. Toxins (Basel). 2025;17(10). doi:10.3390/toxins17100475. pubmed.ncbi.nlm.nih.gov/41150176/
10. Smith SA. Virtual reality in episodic memory research: a review. Psychon Bull Rev. 2019;26(4):1213-1237. doi.org/10.3758/s13423-019-01605-w
11. Johnston W, Judice PB, Molina García P, et al. Recommendations for determining the validity of consumer wearable and smartphone step count: expert statement and checklist of the INTERLIVE network. Br J Sports Med. 2021;55(14):780-793. doi:10.1136/bjsports-2020-103147. pubmed.ncbi.nlm.nih.gov/33361276/
12. Neubauer L, Straw I, Mariconti E, Tanczer LM. A systematic literature review of the use of computational text analysis methods in intimate partner violence research. J Fam Violence. 2023. doi:10.1007/s10896-023-00517-7. doi.org/10.1007/s10896-023-00517-7
13. Kempny C, Frings J, Rust P, Meister S, Fehring L. The use and methodological reporting of large language models in qualitative research: a scoping review. BMC Med Res Methodol. 2026;26(1). doi:10.1186/s12874-026-02913-1. pubmed.ncbi.nlm.nih.gov/42298426/
14. Gierbolini-Rivera RD, Franco Silva M, Augusto de Paula da Silva A, et al. Generative artificial intelligence for qualitative methods in health research: rapid review. J Med Internet Res. 2026;28:e98551. doi:10.2196/98551. pubmed.ncbi.nlm.nih.gov/42684312/
15. Tiong WN, A Rahim AI, Arifin Wan Mansor WN, Yaacob NA. Mapping e-cigarette content on social media: a scoping review of methods, AI applications, and promotional strategies. J Health Commun. 2026;31(5):205-221. doi:10.1080/10810730.2026.2684616. pubmed.ncbi.nlm.nih.gov/42277591/
16. Xu Y, Cong R, Liu M, et al. Study of human papillomavirus vaccine-related social media content across eight social media platforms: scoping review of communication content, information sources, and analytical approaches. J Med Internet Res. 2026;28:e92040. doi:10.2196/92040. pubmed.ncbi.nlm.nih.gov/42743585/

Editverse Research Team | Updated October 2026