This study applied the population, intervention, comparator, outcome, and study design framework (PICOS) (Supplementary Table S1) to establish the inclusion criteria for relevant studies. Eligible studies met the following criteria: (1) study design: Randomized controlled trials including parallel or multi-arm trials; (2) participants: Patients diagnosed with AD or cognitive impairment according to established diagnostic criteria, such as the Diagnostic and Statistical Manual of Mental Disorders (DSM), the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA), or the International Classification of Diseases (ICD); (3) Intervention and control groups: An experimental group receiving natural compounds or extracts, compared with a control group receiving a placebo, equivalent, or standard treatment; (4) Outcome measures: Cognitive outcomes assessed via the Mini-Mental State Examination (MMSE) and/or the Alzheimer Disease Cooperative Study-Activities of Daily Living Scale (ADAS-cog). Exclusion criteria included studies that: (1) were derived from the same trial; (2) Lacked analyzable data; (3) Were not available in full-text format; (4) Were not published in English; (5) Combined multiple natural compounds or extracts in the intervention.

The systematic review followed guidelines outlined in the Cochrane Handbook for Systematic Reviews of Interventions (version 6.3; Higgins et al., 2022) and the Centre for Reviews and Dissemination (University of York, 2009). The study adhered to the preferred reporting items for systematic reviews and meta-analysis framework (PRISMA) (Supplementary Table S2; Moher et al., 2009). The review protocol was registered in International prospective register of systematic reviews (PROSPERO) under registration number CRD42022369293. A systematic search was conducted across Cochrane, PubMed, PsycARTICLES, Scopus, and Web of Science databases from inception to April 10, 2024. The search strategy included the following terms: (Alzheimer’s disease OR Alzheimer dementia) AND (natural OR compound OR flower OR plant OR extract* OR powder OR oil) AND (cognition OR cognitive function OR cognit*).

Two independent reviewers screened articles for eligibility using EndNote X9 for reference management. First, titles and abstract were screened for relevance. Full texts of potentially eligible studies were then reviewed. Reference lists from included studies and relevant systematic reviews were also hand-searched to identify additional eligible articles.

The risk of bias was assessed using the Cochrane Risk of Bias tool (RoB2), which evaluates aspects such as randomization, deviations from intended interventions, missing outcome data, outcome measurement, and selective reporting (Sterne et al., 2019). Studies were categorized as having low risk, some concerns, or high risk of bias. When applicable, funnel plots and Egger’s test (Egger et al., 1997) were employed to evaluate potential publication bias.

A standardized form was used to extract the following data: Publication information (authors, title, year); Study characteristics (design and number of participants); Participant characteristics (drug type, dosage, and duration of intervention); Cognitive outcomes (mean values and standard deviations for MMSE and ADAS-cog scores). When data were presented as means with 95% confidence intervals (CIs) or as medians with interquartile ranges, these were converted into means and standard deviations (SDs) using methods from the Cochrane Handbook (Chapter 6.5.2; Higgins et al., 2022) or Wan et al.’s formulas (Wan et al., 2014).

Pooled data were analyzed using R (version 4.3.3) with the “meta” package. For long-term studies (≥ 6 weeks), endpoint and baseline data were used to calculate mean differences for intervention and control groups. Results were presented in forest plots as weighted mean differences or standardized mean differences (SMDs) with 95% CIs and two-sided p values. Subgroup analyses were conducted to explore variations in study designs and characteristics.

Global cognitive outcomes were evaluated using: (a) MMSE, scores range from 0 to 30, with higher scores indicating better cognition; and (b) ADAS-cog, scores range from 0 to 70, with higher scores indicating greater cognitive impairment. Effect sizes were calculated as Hedges’ g, classified as very small (< 0.2), small (0.2–0.5), moderate (0.5–0.8), and large (> 0.8; Hedges, 2009). The Hartung-Knapp-Sidik-Jonkman random-effects model was used to account for heterogeneity in treatment effects (Inthout et al., 2014).

Statistical heterogeneity was assessed using the chi-squared (χ²) test and I² statistic. I² value ≥50% indicated moderate heterogeneity, while values between 75 and 100% suggested substantial heterogeneity (Higgins et al., 2003). Sensitivity analyses were performed by systematically excluding studies to identify potential outliers influencing the overall effect size. Statistical significance was set at p < 0.05 for all analyses.

The study selection process is illustrated in Figure 1, adhering to PRISMA guidelines (Moher et al., 2009). A comprehensive search across five databases yielded 6,687 articles, supplemented by 55 additional articles identified through manual searches of reference lists from relevant studies. After the removal of duplicates, 5,491 articles remained. Of these, 5,393 were excluded for not being human intervention studies, randomized controlled trials or for only having abstract-level information available. Following the screening of titles and abstracts, 98 articles evaluating cognitive function were shortlisted for full-text review. Subsequently, 53 articles were excluded due to not meeting inclusion criteria. Ultimately, 45 trials were included in the qualitative review, of which 35 provided sufficient data for meta-analysis.

The systematic review incorporated 45 studies, with a detailed summary provided in Table 1. These studies addressed various dimensions, including study quality, sample size, participant characteristics (e.g., health status, diagnostic criteria), intervention types, dosages, durations, cognitive assessment measures, and key outcome metrics.

The total sample size across the included studies was 8,532 participants, with a mean age of 72 years. Among these studies, 31 focused on older adults clinically diagnosed with mild to moderate Alzheimer’s disease (AD) based on established diagnostic criteria such as NINCDS-ADRDA, DSM-III, or DSM-IV (e.g., Akhondzadeh et al., 2003a, 2003b; Akhondzadeh et al., 2010a; Akhondzadeh et al., 2010b; Fernando et al., 2023; Freund-Levi et al., 2008; Herrschaft et al., 2012; Hofferberth, 1994; Ihl et al., 2011; Le Bars et al., 1997; Le Bars et al., 2002; Maurer et al., 1997; Mazza et al., 2006; Muangpaisan et al., 2022; Noguchi-Shinohara et al., 2020; Quinn et al., 2010; Rafii et al., 2011; Rasi Marzabadi et al., 2022; Schneider et al., 2005; Thal et al., 1999; Turner et al., 2015; van et al., 2000; Wade et al., 2014; Wang et al., 2018; Xu et al., 2012). Two studies investigated vascular dementia (Erkinjuntti et al., 2002; Xu et al., 2012), and two other examined AD with comorbid neuropsychiatric symptoms (Herrschaft et al., 2012; Ihl et al., 2011). Additional studies addressed participants with multi-infarct dementia (Le Bars et al., 1997; Le Bars et al., 2002) or concurrent use of acetylcholinesterase inhibitors (AChEIs) (Freund-Levi et al., 2008). Twelve studies specifically targeted older adults with MCI (e.g., Boespflug et al., 2018; Calapai et al., 2017; Chatzikostopoulos et al., 2024; Kaddoumi et al., 2022; Lee et al., 2017; Lee et al., 2013; Lee et al., 2020; Noguchi-Shinohara et al., 2023; Park H. et al., 2019; Shin et al., 2009; Tsolaki et al., 2016; Tsolaki et al., 2020; You et al., 2021), and one study focused on amnestic and multi-domain MCI (Tsolaki et al., 2016). Two studies exclusively addressed participants with moderate-to-severe AD (Farlow et al., 2019; Farokhnia et al., 2014).

The interventions in the reviewed studies had an average duration of 27 weeks, ranging from 6 to 96 weeks, and predominantly utilized natural extracts. These extracts included Cocos nucifera (N = 1; Fernando et al., 2023), Cosmos caudatus (N = 1; You et al., 2021), Crocus sativus L. (N = 5; Akhondzadeh et al., 2010a, Akhondzadeh et al., 2010b; Farokhnia et al., 2014; Rasi Marzabadi et al., 2022; Tsolaki et al., 2016), Ganoderma lucidum (N = 1; Wang et al., 2018), Garcinia mangostana L. (N = 1; Muangpaisan et al., 2022), Ginkgo biloba (N = 8; Herrschaft et al., 2012; Hofferberth, 1994; Ihl et al., 2011; Le Bars et al., 1997; Le Bars et al., 2002; Maurer et al., 1997; Mazza et al., 2006; Schneider et al., 2005), Melissa officinalis (N = 3; Akhondzadeh et al., 2003a; Noguchi-Shinohara et al., 2023; Noguchi-Shinohara et al., 2020), Olea europaea L. (N = 2; Kaddoumi et al., 2022; Tsolaki et al., 2020), Panax ginseng (N = 1; Park H. et al., 2019), Polygala tenuifolia Willdenow (N = 1; Shin et al., 2009), Punica granatum (N = 1; Chatzikostopoulos et al., 2024), Salicornia europaea L. (N = 1; Lee et al., 2020), Salvia officinalis (N = 1; Akhondzadeh et al., 2003b), Vitis vinifera (N = 2; Calapai et al., 2017; Lee et al., 2017), Vaccinium ashei, and Vaccinium corymbosum L. (N = 1; Boespflug et al., 2018). These supplements were administered orally, primarily in the form of capsule powders, capsule liquids, or liquid solutions, as detailed in Table 1. The concentration of the natural extracts varied significantly across studies. Specifically, Cocus nucifera was administered at a dosage of 30 mL per day, Cosmos caudatus at 500 mg/day, and Crocus sativus L. at 30 mg/day. Ganoderma lucidum was provided at 1 g/day, while Garcinia mangostana L. dosage ranged from 220 to 560 mg/day depending on body weight. For Ginkgo biloba, the standardized Ginkgo biloba extract (EGb) 761 was administered in dosages ranging from 80 mg/day to 240 mg/day, typically at 120 or 240 mg/day. Punica granatum was given as 5 drops of seed oil daily. The administration of Melissa officinalis varied among studies: Noguchi-Shinohara et al. (2023), Noguchi-Shinohara et al. (2020) provided capsule powder containing at least 500 mg of rosmarinic acid per capsule, whereas Akhondzadeh et al. (2003a) offered a liquid solution with a concentration sufficient to provide at least 500 μg of citral per milliliter. The administration protocols for Olea europaea L. also differed between studies. Kaddoumi et al. (2022) supplemented participants with 30 mL/day of extra-virgin oil compared to refined olive oil at the same dosage. In contrast, Tsolaki et al. (2020) supplemented participants with high phenolic early extra-virgin olive oil or moderate phenolic oil at 50 mL/day, comparing these interventions to a Mediterranean diet. Panax ginseng was supplied in powder from at a dosage of 3 g/day, Polygala tenuifolia Willdenow (root extract powder designated as BT-11) at 300 mg/day, and Salicornia europaea at 600 mg/day. Salvia officinalis was administered as a liquid solution. Vitis vinifera was provided as powders at dosages of 250 mg/day (Calapai et al., 2017) and 72 g/day (Lee et al., 2017). Lastly, a combination of Vaccinium ashei Reade and Vaccinium corymbosum L. was administered in a 1:1 ratio at a daily dose of 25 g (Boespflug et al., 2018).

In addition to these natural extracts, fifteen studies supplemented participants with other natural compounds, including Bryostatin (N = 1; Farlow et al., 2019), Docosahexaenoic acid (DHA) (N = 3; Freund-Levi et al., 2008; Lee et al., 2013; Quinn et al., 2010), Huperzine A (N = 2; Rafii et al., 2011; Xu et al., 2012), Melatonin (N = 1; Wade et al., 2014), Physostigmine (N = 2; Thal et al., 1999; van et al., 2000), Galantamine (N = 5; Erkinjuntti et al., 2002; Raskind et al., 2000; Rockwood et al., 2001; Tariot et al., 2000; Wilcock et al., 2000), and Resveratrol (N = 1; Turner et al., 2015). Bryostatin was intravenously infused seven times over a 12-week period, with dosages of either 24 μg twice and 20 μg five times or 48 μg twice and 40 μg five times, and the mean infusion time was 45 ± 5 min. For DHA supplementation, Quinn et al. (2010) used algae-derived DHA without eicosapentaenoic acid (EPA) at 2 g/day, whereas Lee et al. (2013) and Freund-Levi et al. (2008) utilized fish-derived DHA containing EPA, supplementing participants with 1.3 g DHA plus 0.45 g EPA/day and 1.7 g DHA pluls 0.6 g EPA/day, respectively.

Huperzine A was administered at dosages of 0.1 mg, 0.2 mg, or 0.4 mg/day. Galantamine dosages ranged from 8 mg to 32 mg/day, with 24 mg/day being the most frequently used dosage. Resveratrol was administered with an escalating dosage ranging from 500 mg to 2,000 mg/day.

All studies included in the analysis employed a RCT design. Among these, one study implemented a single-blind procedure (Tsolaki et al., 2016), and two studies did not report a blinding procedure (Chatzikostopoulos et al., 2024; Wang et al., 2018). The remaining studies were conducted with double-blinding. Most studies adhered to well-designed case–control methodologies in accordance with predefined inclusion criteria and provided comprehensive descriptions of their objectives, definitions, and methodologies.

While all studies aimed to investigate the effects of natural compounds or extracts on cognitive health, there were variations in the selection of control groups. Specifically, 37 studies utilized isoenergetic placebos as the control intervention, whereas four studies employed commonly prescribed drugs. For instance, Rasi Marzabadi et al. (2022), Akhondzadeh et al. (2010a), and Mazza et al. (2006) used donepezil as the control intervention to compare its effects with those of Crocus sativus or Ginkgo biloba extract. Farokhnia et al. (2014) used memantine as the control intervention to compare its effects with Crocus sativus L. One study on Olea europaea L. (Kaddoumi et al., 2022) used refined olive oil as the control to compare with extra-virgin olive oil, while other studies on Olea europaea L. and Punica granatum employed the Mediterranean diet as the control intervention (Chatzikostopoulos et al., 2024; Tsolaki et al., 2020). Additionally, Fernando et al. (2023) utilized canola oil as the control intervention to compare its effects with those of virgin coconut oil. These methodological considerations, including blinding procedures and the selection of appropriate control groups, were meticulously implemented to ensure the robustness and validity of the findings regarding the impact of natural compounds and extracts on cognitive health.

The quality of the included RCTs was assessed using the Cochrane Risk of Bias tool. Of the 35 studies included in the meta-analysis, 25 exhibited a low risk of bias in randomization, 8 raised some concerns, and 1 was rated as high risk (Figure 2). Deviation from intended interventions showed low risk in 31 studies, while 4 raised concerns. Regarding missing outcome data, 27 studies were rated as low risk and 9 as having some concerns. Outcome measurement risk was low in 25 studies, with 10 raising concerns. Finally, selection bias for reported results was rated low in 24 studies, with 8 raising some concerns and 3 being high risk.

Overall, 18 studies were deemed to have low risk, 13 presented some concerns, and 4 demonstrated high risk of bias (Figure 2). These assessments provided a robust foundation for interpreting the results of the meta-analysis.

Table 1 displays all 45 studies included in the cognitive analysis. Of these, 34 studies reported trends toward cognitive improvement with supplement use, four studies found no significant differences between supplements and commonly prescribed drugs (donepezil, memantine), and seven studies observed no improvement in cognitive measures.

The cognitive assessments employed in the included studies utilized a variety of tools, such as the ADAS-cog, MMSE, CDR-SB, Disability Assessment for Dementia, Functional Rating Scale of Symptoms of Dementia, and Geriatric Depression Scale (Table 1). However, not all of these measures were incorporated into the subsequent meta-analyses.

To assess the overall effect size, the meta-analysis also conducted subgroup analyses to examine the impact of distinct categories, including natural extracts, natural compounds, and specific compound classes such as terpenoids, phenols, and alkaloids (Figures 3–6). However, due to limited data availability, more granular subgroup analyses focusing on specific plant structural extracts (Supplementary Figures S1, S2) and hormonal compounds (Figures 5, 6) could not sufficiently explore variations arising from study designs and participant characteristics. Consequently, these findings must be interpreted with caution and in consideration of the underlying limitations.

As summarized in Table 1, 25 studies provided sufficient data for meta-analysis using the ADAS-cog, a widely recognized neuropsychological tool for evaluating cognitive severity in dementia (Akhondzadeh et al., 2003a, 2003b; Akhondzadeh et al., 2010a; Akhondzadeh et al., 2010b; Chatzikostopoulos et al., 2024; Erkinjuntti et al., 2002; Le Bars et al., 1997; Le Bars et al., 2002; Lee et al., 2017; Lee et al., 2020; Maurer et al., 1997; Muangpaisan et al., 2022; Noguchi-Shinohara et al., 2023; Noguchi-Shinohara et al., 2020; Quinn et al., 2010; Rafii et al., 2011; Raskind et al., 2000; Rockwood et al., 2001; Schneider et al., 2005; Tariot et al., 2000; Thal et al., 1999; Tsolaki et al., 2020; van et al., 2000; Wade et al., 2014; Wang et al., 2018; Wilcock et al., 2000). Additionally, 19 studies provided sufficient data for meta-analysis using the MMSE, a commonly used tool to evaluate cognitive impairment in clinical and research contexts (Calapai et al., 2017; Chatzikostopoulos et al., 2024; Fernando et al., 2023; Kaddoumi et al., 2022; Lee et al., 2017; Lee et al., 2013; Mazza et al., 2006; Noguchi-Shinohara et al., 2023; Noguchi-Shinohara et al., 2020; Park K. C. et al., 2019; Quinn et al., 2010; Rafii et al., 2011; Shin et al., 2009; Tsolaki et al., 2016; Tsolaki et al., 2020; van et al., 2000; Wade et al., 2014; Xu et al., 2012; You et al., 2021).

Notably, the meta-analysis identified a significant improvement in ADAS-cog scores among participants receiving natural extracts or compounds compared to controls (SMD = −2.88, 95% CI −4.26 to −1.50, t₂₄ = −4.31, p < 0.01) (Figure 3). Similarly, MMSE scores showed significant improvement following interventions compared to controls (SMD = 0.76, 95% CI 0.06 to 1.46, t₁₈ = 2.27, p = 0.04) (Figure 4).

Despite the observed heterogeneity, these findings suggest that natural extracts and compounds exert a significant and substantial effect on global cognitive function. However, the variability in effect sizes across individual studies indicates that these impacts may differ depending on the cognitive domain or compound characteristics.

To address the heterogeneity, additional subgroup and moderator analyses were conducted to investigate the influence of factors such as the type of natural extract, compound class (e.g., terpenoids, phenols, and alkaloids), and outcome measures on the observed cognitive improvements following intervention.

Terpenoids derived from a variety of natural sources, including Ginkgo biloba, Crocus sativus L. (saffron), ginseng, Polygala tenuifolia Willdenow (Polygala), Ganoderma lucidum (Reishi mushroom), and Cosmos caudatus, exhibit diverse neuroprotective properties and hold therapeutic potential for AD. In Ginkgo biloba, terpenoid components such as ginkgolides and bilobalide are recognized for their antioxidant and neuroprotective effects. Although some studies report modest cognitive benefits in AD patients, others find no significant effects. These inconsistencies may stem from variability in formulations, dosages, study designs, and patient-specific factors, including the stage of the disease (DeKosky et al., 2008; Herrschaft et al., 2012; Hofferberth, 1994; Ihl et al., 2011; Kanowski and Hoerr, 2003; Le Bars et al., 1997; Le Bars et al., 2002; Maurer et al., 1997; Mazza et al., 2006; Schneider et al., 2005; Shi et al., 2010; Snitz et al., 2009).

Saffron contains active compounds such as crocin, crocetin, and safranal, which demonstrate antioxidant, anti-inflammatory, and neuroprotective effects. Specifically, crocetin has been shown to modulate Aβ pathology, while safranal enhances cognitive function in preclinical AD models (Ahmad et al., 2023; Akhondzadeh et al., 2010a; Farokhnia et al., 2014; Finley and Gao, 2017; Pandey et al., 2020; Rasi Marzabadi et al., 2022; Tsolaki et al., 2016). Similarly, ginsenosides―the triterpene saponins found in ginseng―exhibit efficacy in improving cognitive function and mitigating AD-related pathologies through their antioxidant and anti-inflammatory properties (Heo et al., 2012; Lee et al., 2008; Park H. et al., 2019; Wang et al., 2019). Tenuifolin, a triterpenoid saponin from Polygala tenuifolia, along with related compounds such as polygalasaponin F, has been associated with memory enhancement and neuroprotection by modulating neurotransmitter levels and reducing oxidative stress (Deng et al., 2020; Jia et al., 2004; Moratalla-López et al., 2019; Park H. et al., 2019). In Ganoderma lucidum, triterpenoids like ganoderic acids provide neuroprotective effects by attenuating neuroinflammation and oxidative stress, potentially leading to improved cognitive function in AD (Qi et al., 2021; Zheng et al., 2023). Overall, the diverse terpenoid compounds from these natural sources offer promising avenues for the development of therapeutic strategies targeting the multifaceted pathologies of AD.

The bioactive constituents of Cosmos caudatus include flavonoids (e.g., quercetin and kaempferol), phenolic acids (e.g., caffeic acid), and carotenoids (e.g., β-carotene). These compounds exhibit antioxidant and neuroprotective properties, suggesting their potential in mitigating AD-related neurodegeneration, despite the limited specific research available (Wang et al., 2023). Phenolic compounds found in Melissa officinalis (lemon balm), Olea europaea (olive), Garcinia mangostana (mangosteen), Salicornia europaea (samphire), Salvia officinalis (sage), and Vitis vinifera (grape) further highlight their therapeutic potential. Rosmarinic acid in lemon balm has demonstrated both anti-inflammatory and neuroprotective effects (Petrisor et al., 2022). Compounds derived from olives, such as oleuropein, hydroxytyrosol, and oleocanthal, have shown efficacy in reducing oxidative stress and Aβ pathology while providing cognitive benefits in AD models (Abdallah et al., 2022; Abuznait et al., 2013; Nardiello et al., 2018). Xanthones from mangosteen, including α-mangostin and γ-mangostin, possess strong antioxidant and anti-inflammatory properties, and catechin flavonoids provide additional neuroprotection (Do and Cho, 2020; Pratiwi et al., 2022; Yang et al., 2021). Salicornia europaea contains phenolics and carotenoids, such as lutein, which may help reduce oxidative stress (Fitzner et al., 2021). Sage’s phenolic compounds, including carnosic and ursolic acid, have been shown to improve cognitive function and memory while providing neuronal protection (Ghorbani and Esmaeilizadeh, 2017; Mirza et al., 2021; Yi-Bin et al., 2022). Compounds in grapes, particularly resveratrol, have been found to inhibit Aβ aggregation and neuroinflammation, along with providing additional antioxidant benefits through proanthocyanidins and flavonoids such as quercetin (Tabeshpour et al., 2018).

Several alkaloids, including bryostatin, huperzine A, physostigmine, and galantamine, exhibit significant therapeutic potential. Bryostatin, a macrolide derived from Bugula neritina, modulates protein kinase C and has shown promise in treating neurodegenerative and oncological conditions (Farlow et al., 2019; Nelson et al., 2017; Zonder et al., 2001). Huperzine A, extracted from Huperzia serrata, enhances cognitive function by inhibiting acetylcholinesterase, a key enzyme involved in AD pathology (Friedli and Inestrosa, 2021; Liu et al., 2020; Rafii et al., 2011; Xu et al., 2012). Galantamine, an FDA-approved acetylcholinesterase inhibitor, alleviates AD symptoms by enhancing cholinergic signaling, thereby providing cognitive and functional benefits (Santos et al., 2020).

Other bioactive compounds, such as omega-3 fatty acids (e.g., DHA) and melatonin, contribute significantly to neuroprotection in AD. DHA has been shown to reduce neuroinflammation, oxidative stress, and Aβ aggregation, while also supporting neuronal survival and synaptic plasticity (Freund-Levi et al., 2008; Khalid et al., 2022; Lee et al., 2013; Quinn et al., 2010; Thomas et al., 2015; Xiao et al., 2022; Yurko-Mauro et al., 2010). Melatonin, known for its role in regulating circadian rhythms and sleep, provides antioxidant and anti-inflammatory effects that mitigate AD pathology, including Aβ accumulation and tau hyperphosphorylation (Cardinali et al., 2014; Furio et al., 2007; Hardeland, 2018; Li et al., 2020; Lin et al., 2013; Wade et al., 2014; Xu et al., 2020).

These natural compounds present a multifaceted approach to combating AD by targeting oxidative stress, inflammation, and amyloid and tau pathologies. However, further research is required to optimize their clinical utility and establish standardized protocols for therapeutic application.

The meta-analysis presented in this study emphasizes the need for further research to validate the potential cognitive benefits of natural compounds and extracts, despite the promising results identified in individual interventions. While previous systematic reviews have addressed natural compounds in preclinical and clinical trials (Ahmad et al., 2023; Andrade et al., 2019; Li et al., 2023), this study distinguishes itself by focusing exclusively on RCTs that evaluated global cognitive domains using widely accepted measures such as the ADAS-cog and MMSE. By employing meta-analyses and providing statistical evidence, this study contributes to a more comprehensive understanding of the effects of natural compounds and extracts. Furthermore, the study’s selection criteria targeted RCTs that utilized a single species of natural extract or specific compound, excluding those involving multiple extracts, compounds, or formulations. This approach aligns closely with real-world practices typically employed by caregivers, making the findings highly relevant for daily clinical management of individuals with MCI or AD. It is essential for individuals considering any dietary supplement to consult healthcare professionals or AD specialists, who can provide personalized guidance based on individual circumstances and health status. Although short-term cognitive improvement was observed across all RCTs, the validation of long-term efficacy in individuals with cognitive impairment necessitates large-scale longitudinal trials. To the best of our knowledge, this study is the first systematic review and meta-analysis that compares the effects of various forms of natural compounds and extracts on patients with MCI or AD using recognized assessment measures such as the MMSE and ADAS-cog. Despite the positive findings, several limitations must be acknowledged. This include the diversity in intervention types, variations, in duration and participant characteristics, and the moderate to high risk of bias in the quality of studies using the same cognitive tasks. Additionally, the vast scope of this topic inevitably meant that not all types or species of natural compounds and extracts could be included. Some natural compounds are still in the preclinical or early clinical phases, and certain studies employing non-RCT designs, such as pilot or cross-over studies, were excluded from this review. This exclusion does not imply a lack of potential benefit to cognitive health. Despite a rigorous search methodology, including additional hand-searching, it is possible that some relevant articles were inadvertently missed. Furthermore, each natural compound and extract originates from distinct sources and possesses unique mechanisms of action and biological effects, necessitating a cautious interpretation of the findings. Future research should address these limitations and aim to provide a more comprehensive understanding of the therapeutic potential of natural compounds and extracts in improving cognitive function.