The use of resin-based composites has become a part of modern dentistry, which involves preparing conservatively, having good handling properties, and being able to re-create the optical properties of natural teeth.[1] Patients have higher expectations of restorations that are undetectable by others to the adjacent tooth structure, and the demand for highly esthetic direct restorations has grown.[2] But the composition and finishing of the restorative material is not the only criterion for achieving optimum esthetic results; accurate shade selection is crucial.[3] Traditional shade matching largely relies on the visual evaluation of shades with shade guides and is subject to the influence of ambient lighting conditions, tooth dehydration, experience of the observer, color perception, metamerism, and inter-commercial differences between composite systems.[4] A systematic review published in 2021 that comprised 249 papers related to the topic found that in most cases, digital shade-selection methods were more accurate and precise than traditional visual methods, but clinical standardization was needed for reliable results.[5]

The introduction of artificial intelligence (AI) has brought along a promising new strategy for dental shade selection, allowing digital images to be objectively analyzed and differences in hue, chroma, value, and translucency to be identified.[6] AI systems can handle vast quantities of color data and deliver consistent shade recommendations, minimizing reliance on the operator's perception as opposed to traditional visual methods. This is quickly gaining traction.[7] According to a 2026 clinical study with 60 patients, the accuracy of the AI-driven shade matching was 86.7%, higher than the accuracy of conventional visual assessment methods (58.3%), and the AI-driven shade matching was more reproducible than the conventional visual assessment method (κ=0.78 versus κ=0.42).[8] However, not all evidence is positive: a 2026 study of the AI chatbots versus a dental specialist found that the AI models achieved less accuracy when selecting shades, with Gemini Advanced at 47.9% and ChatGPT-4.0 at 42.9%, versus the 57.1% achieved by the dental specialist, and this accuracy can vary based on the platform, imaging conditions, and algorithm used. The results of this study underscore the importance of clinical trials and do not equate technological advancement with better restorative results.[9]

This is even more critical when using modern resin material. Modern single-shade and multi-shade composites incorporate advanced optical phenomena such as structural color, light interaction, translucency, and potential for color adjustment, all of which help integrate with the surrounding tooth structure.[10] However, first impressions of shade matching do not necessarily signify esthetic stability over the long-term. A four-trial randomized clinical trial systematic review and meta-analysis of 263 restorations in 2024 reported that there were no differences in color matching and color stability between single-shade and multi-shade resin composite over 12 months, with low certainty of evidence due to the small number of participants in the studies.[11] Therefore, the success of a resin-based esthetic restoration should be interpreted as a combination of the proper selection of the initial shade and the long-term stability of the color, surface characteristics, marginal integrity, and long-term survival.

Despite the advancement of AI-driven shade selection technologies, one significant clinical issue of patients getting better, esthetically acceptable restorations in the long term from more accurate digital shade selection remains to be solved. Many of the investigations have focused on the accuracy or repeatability of shade selection techniques, while few clinical studies have directly correlated AI-supported shade determination with esthetic outcomes and longevity of the newer resin-based restorations. The distinction is clinically relevant because a technically correct shade match might not yield the best results once the restoration is placed into function and is challenged by changes in tooth structure, dietary stains, and oral aging. Thus, a direct comparison of the AI-assisted and conventional shade matching is warranted to clarify if shade matching is clinically better using the technological approach. Therefore, the present study aimed to comparatively assess the AI-assisted shade matching and conventional shade matching in terms of esthetic outcome and long-term success of modern resin-based restorative materials.

A prospective, randomized, comparative study was performed to evaluate AI-assisted and conventional shade matching methods of contemporary resin-based restorative materials. This study was conducted in the Department of Dental Materials, Army Medical College, Rawalpindi. This study was to be carried out for 12 months from 1st July 2025 to 30th June, 2026. Ethical approval was secured from the institutional research ethics committee prior to the study, and informed written consent was granted after explaining the objectives, procedures, possible benefits, and follow-up needs to all study participants. The sample size was calculated using OpenEpi version 3.01 for comparison of two independent proportions. The calculation was based on the findings of a recent clinical study that reported shade-matching accuracy of 86.7% with an AI-based method compared with 58.3% with conventional visual assessment, with a 95% confidence level, an 80% power, and a ratio of 1:1.[8] The sample size per group was calculated at about 60 restorations, for a total of 120 restorations. The sample size was increased to 134 restorations (67 per group) to allow for an expected 10% loss due to follow-up or restoration-related exclusions during the follow-up period. Direct resin-based restorations were necessary in the restorative dentistry outpatient clinic, and consecutive sampling was used to recruit the participants who met the inclusion criteria. Patients who presented for treatment with a need for an esthetic direct resin composite restoration in the anterior or premolar area of the dentition with sufficient remaining tooth structure for adhesive restoration and who consented to return for scheduled examinations were eligible for the study. Only teeth having clinically acceptable surrounding tooth structure and a definite natural shade were included. Patients were excluded if they had significant periodontal disease, significant tooth discoloration, intrinsic staining or developmental enamel defects which would affect shade determination, extensive existing restorations in the shade assessment area, active untreated caries in adjacent teeth or active untreated caries in the shade assessment area within the previous three months. Those who had parafunctional habits, severe bruxism, could not maintain adequate isolation, were known to have an allergy to restorative materials, or were known to have a behavioral or systemic condition that would interfere with follow-up were also excluded. Following enrollment, eligible patients were randomized 1:1 between the AI-assisted shade matching group and the conventional shade matching group with a computer-generated random sequence. Standardized intraoral photographs were captured by the AI-assisted group under controlled illumination with a calibrated digital imaging system or a smartphone-based imaging protocol. The pictures featured the tooth to be restored, adjacent reference teeth, and were subjected to the chosen AI shade-matching software. The recommended resin composite shade was recorded and was provided by the AI system before the restorative procedure. An experienced restorative dentist selected shadings under standardized clinical illumination for the conventional group with the manufacturer's shade guide. Shade selection was done prior to rubber dam placement, whenever possible, as it was possible that the colour of the teeth could change due to dehydration. The shade selected was recorded, and the same modern resin-based restorative material was used in both groups following the manufacturer's instructions to reduce material variation. The standard adhesive protocol was followed for the restorative procedure. The teeth were isolated adequately, cleaned with a non-fluoridated prophylaxis paste, and prepared according to clinical demands of the lesion in a conservative manner. The selected shade of resin composite was incrementally filled, and each increment was light cured for an appropriate period of time as recommended by the manufacturer. Standardized instruments and protocols were used to finish and polish. Clinical photographs were taken as a baseline after completion of the restoration. The esthetic results were assessed by a standardized clinical photograph and direct clinical assessment. The color matching, marginal discoloration, surface staining, surface texture, anatomic form, and esthetic acceptability were measured with modified United States Public Health Service/FDI criteria.[12] When objective color measurement was possible, color differences were also measured by a digital spectrophotometric or calibrated photographic method and reported as ΔE00. Participants were called back at baseline, 3 months, 6 months, and 12 months following the placement of the restoration(s). On each follow-up visit, the restorations were evaluated for color match, color stability, marginal discoloration, surface integrity, marginal adaptation, secondary caries, fracture, postoperative sensitivity, and restoration repair/replacement. The esthetic success was defined as a maintenance of an acceptable color match and a clinically acceptable surface and marginal characteristics without replacement. The criterion for longevity was restoration survival as determined by fracture, loss of restoration, unacceptable discoloration, recurrent caries, major marginal deterioration, and the need for replacement. Follow-up evaluations were conducted by the same examiner, who was blind to the shade selection procedure (where possible) to minimize observer bias. The clinical follow-up schedule was chosen due to the previous randomized clinical studies in which stoneability and color match of resin-composites have shown to be measurable over time with significant differences over 12 months. The data collected were fed into IBM SPSS and then analyzed using SPSS Statistics version 26.0. Continuous variables (age of the patient, color-difference values, clinical scores) were examined for normality using the Shapiro–Wilk test and reported as mean ± SD. The categorical variables were sex, tooth type, shade category, acceptable esthetic outcome, marginal discoloration, restoration failure, repair, and replacement, and were analyzed as frequencies and percentages. Data distribution was used to decide whether the independent t-test would be employed to compare continuous outcomes between the AI-assisted group and the conventional group. Chi-square and Fisher's exact test were used to compare categorical variables. Repeated-measures analysis of variance (parametric) or the Friedman test (non-parametric) were used to assess changes in esthetic parameters over follow-up visits with post-hoc comparisons. The restoration survival between the two groups was assessed by Kaplan-Meier survival analysis, and compared by the log-rank test. To determine factors associated with restoration failure, Cox proportional-hazards regression was performed. The two-sided p-value was used as the significance criterion (p<0.05) and the effect size was presented with 95% confidence intervals.

A total of 134 restorations were evaluated, with 67 in each group. There were no statistically significant differences between the AI-assisted and conventional groups in terms of any baseline characteristic (all p>0.05), which led to satisfactory comparability between the groups. (Table 1)

AI-guided shade matching showed the best initial shade-selection performance. There was significantly greater shade matching, clinically acceptable color matching, and higher esthetic success in the AI-assisted group and significantly less objective color difference (ΔE00). In surface texture and anatomic form, there was no difference between the groups. (Table 2)

AI shade matching was maintained through follow-up. At 3, 6, and 12 months, acceptable colour matching was significantly higher in the AI-assisted group. During follow-up, marginal discoloration dropped significantly in the AI group, and the difference in surface integrity was not statistically significant. (Table 3)

Color difference increased with the significant rise in both groups after 12 months, which meant that there were progressive changes in color after restoration placement. The AI-assisted group, however, consistently had lower ΔE00, and the between-group difference was statistically significant at 12 months. (Table 4)

There was significantly less marginal discoloration and surface staining with AI-assisted restorations at 12 months. Fracture, secondary caries, postsurgical sensitivity, repair, replacement, and overall failure were observed with less frequency in the AI group, but with no statistical significance. (Table 5)

There was a numerically higher restoration survival at 12 months between the group treated with AI assistance and the control group, but the difference was not statistically significant using the log-rank test. (Table 6)

Multivariable Cox regression showed that the higher the baseline ΔE00, the greater the restoration-failure hazards at 6 months, while marginal discoloration at 6 months was also independently associated with increased restoration-failure hazards. There were no significant independent predictors of failure other than conventional shade matching, older age, or location of the premolars. (Table 7)

Table 1. Baseline demographic and clinical characteristics of the study participants

Table 2. Comparison of shade-matching and baseline esthetic outcomes

Table 3. Comparison of esthetic outcomes during follow-up

Table 4. Changes in objective color difference (ΔE00) over 12 months

Table 5. Comparison of restoration-related complications and clinical failures

Table 6. Kaplan–Meier restoration survival and Cox regression analysis

Table 7: Cox proportional-hazards regression for restoration failure

The current prospective randomized comparative clinical trial showed that initial shade selection and esthetic results with AI-assisted shade matching were significantly better than conventional visual shade matching. There was a statistically significantly higher number of correct shade matching restorations in the AI-assisted group (86.6%) than in the conventional group (64.2%) (p=0.003), and there was statistically significant higher number of clinically acceptable shade matching restorations in the AI-assisted group (91.0%) than in the conventional group (74.6%) (p=0.009). Likewise, the mean baseline ΔE00 was significantly lower when using shade selection with AI than when using conventional matching (p<0.001). The results suggest that the decrease in operator-dependent color perception can help the restorative shade selection process become more accurate. They are similar to the 2021 review by Tabatabaian et al. that assessed 249 articles, which found that the digital shade-selection methods generally had higher accuracy and precision than visual methods, but that standardization of the environment and the method were also important.[5] Additionally, a 2022 systematic review and meta-analysis of 13 studies found that the color difference values were more significant with conventional visual shade matching than with computer-assisted shade matching (p=0.007), indicating the lower ΔE00 results in the current AI-based group.[13] The improvement was especially similar to the recent clinical evidence that specifically investigated AI. A 2026 clinical study of 60 patients found that the shade matching accuracy of the AI method was 86.7%, significantly higher than the 58.3% achieved by the traditional visual method, and the reproducibility of the AI method was also significantly higher (κ=0.78 vs. κ=0.42).[8] The very comparable accuracy for AI of 86.6% in the present study supports the likelihood of the finding and indicates that AI-supported systems could decrease some of the variability inherent in visual shade determination. The evidence on the use of AI is, however, not entirely positive; Ünal and Polatoğlu, in a 2026 clinical study, found only weak agreement among traditional, digital, and experimental AI-based methods, with nonexistent agreement between traditional visual selection and ChatGPT-4 (κ=0.064, p=0.287).[14] The difference could be attributed to variations in the AI platform, the imaging protocol, the reference standard, the tooth location, and the fact that the AI system was specially developed for dental color analysis. Given this, the results of the present experiments should be seen as supportive of the particular standardized protocol in which the AI was employed, not as proof that any AI platform is superior. The results are also in line with the general literature, which shows better consistency with digital approaches than with visual-only approaches. A clinical study of 31 participants by Abu-Hossin et al. (2023) showed that there was only a slight agreement between visual shade selection and intra-oral scanners, but there was moderate agreement between the scanners, and better repeatability for Trios 3 than for Cerec Omnicam.[15] Similarly, a study made by researchers at the University of Queensland in 2024 comparing direct shade matching with digital images revealed that digital images generated under standard conditions produced good shade matching results and that the digital image could be a good alternative to direct shade matching when the conditions were appropriate.[16] The observations are pertinent to the present study because the present AI system requires standard intraoral photography, not a randomly taken clinical image. Therefore, the improvement in advantage seen may have been due, not only to the algorithmic process, but also to increased standardization in image acquisition. The lower ΔE00 values obtained in the cases of the shades made with artificial intelligence shade selection were an important objective reason for the higher clinical esthetic success in this study. The overall esthetic success at baseline was 92.5% in the AI group and 77.6% in the conventional group (p=0.014). This correlation between digital color measurement and objective color difference measurement was confirmed by the in vivo study of Clinical Tooth Color Matching investigators in 2023, which used standardized digital photocolorimetric analysis and spectrophotometry in 60 maxillary central incisors. That study proved that the standardized photographic color analysis (SPCA) could be an objective method to assess the color of a tooth.[17] In addition, the 2025 systematic review on digital shade matching showed that non-proximity digital instruments in general yielded more consistent clinical outcomes than visual instruments, with varying levels of methodological quality.[18] The present findings provide further support to this evidence by demonstrating that the enhancement of shade determination may be evaluated as a clinical restoration, rather than just a shade-selection stage. The evidence from the present study is consistent with evidence about contemporary resin-based materials, in which a progressive colour change was observed. In a systematic review and meta-analysis of four RCTs with 263 restorations, Leal et al. (2024) found no significant difference in color matching and color stability between single-shade and multi-shade resin composite after 12 months, with low certainty of evidence due to small sample sizes and a low number of events.[19] In a randomized clinical trial with single-shade and multi-shade composites for 12 months, Anwar et al. (2024) also observed that the color-match differences were not significant in the short-term but became significant at 9 and 12 months, with better color-match scores at the latter times in the multi-shade group. These studies show that these optical changes due to the material can become more evident over time.[20] The present study evaluated shade selection strategies with the same restorative system, so that the shade stability difference observed can be attributed to the quality of the initial shade match and not to the differences between the two restorative systems (composites). The low incidence of marginal discoloration and surface staining in the AI group further confirms this overall esthetic benefit. There was also a significant difference between the groups in the rate of marginal discoloration, with 11.9% of AI-assisted restorations experiencing discoloration at 12 months compared to 26.9% of the conventional restorations (p=0.036), and in the rate of surface staining, with 10.4% of AI-assisted restorations showing some surface staining at 12 months, compared to 23.9% of the conventional restorations (p=0.044). It is noteworthy that while the color mismatch is not directly responsible for failure to adapt to margins or stain, restorations with a smaller color mismatch can be viewed for the longer term even when they experience the same aging process. Another study by Anwar et al. also shows that it is possible to meaningfully assess marginal discoloration and optical performance over 12 months, based on the modified USPHS criteria.[20] Generally satisfactory marginal discoloration was found by them in both groups of materials, which is contrary to the present study, where a significant difference between the two groups was obtained after 12 months. The difference could be due to the type of composite used, cavity location and sample size, operator technique, and importantly, comparing shade-selection methods, which may differ, rather than the actual composite formulation. The present results should also be considered in view of the recent research findings of Žagar et al. (2024) that reported differences in tooth color derived from photographs taken with various smartphones, which emphasizes the role of imaging characteristics for photographic techniques in color assessment of teeth.[21] Recently, a study in 2026 on visual, digital-photographic, and spectrophotometric shade selection revealed that spectrophotometry had the highest clinical correlation, while the non-calibrated digital photography had only a statistical correlation but limited clinical correlation.[22] The observations here confirm the value of the standard illumination and calibrated imaging procedure employed in the present study. AI's superior performance does not mean that the need for image calibration is being removed, but rather it looks like it may be helpful when given high-quality, consistent visual information. The overall results of this study indicate the potential of AI in assisting shade matching in today's restorative dentistry, not replacing it. The major benefit seems to be given by the objective and reproducible procedure used in the shade determination, as evidenced by the higher number of correct shade matches, lower ΔE00 values, and better stability of acceptable colour match over a course of 12 months in the present study. However, the discrepancies among recent investigations using AI suggest that implementation in clinical practice should rely on validated algorithms, standardized acquisition of image data, suitable illumination, and objective verification (if available). The lack of significance in the 12-month survival period also underscores that selection of better shade will not ensure restoration success, since there are multiple determinants of restoration longevity. Large multicenter randomized clinical trials with standardized spectrophotometric reference measurements and longer follow-up time are therefore warranted to confirm the esthetic benefit shown here and whether this will eventually result in reduced restoration replacement and repair. Limitations There were several limitations to the current project. The first is the limited length of follow-up time (12 months), which confined long-term survival and color stability of the restorations. The sample size was relatively small and was obtained mainly from a single tertiary-care dental teaching hospital, which may limit the generalizability of the results to other clinical settings and populations; and third, the study did not consider the overall rates of medication prescribing, nor did it report on the specific timing of the procedures. Third, while the imaging and illumination protocols were standardized, the performance of the AI devices was still heavily dependent on image quality, device characteristics, and the type of AI application. Fourth, the operator could not be blinded when selecting the shade because the intervention was, by its nature, different between the two groups. Finally, shade selection was not the only factor that affected the survival of the restorations, as occlusal loading, tooth location, oral hygiene, and patient-related factors could have contributed to a lack of statistically significant difference found in the survival of the restorations.

The AI shade matching showed significantly lower colour difference, an accuracy rate, and maintenance of acceptable esthetic outcomes compared to the conventional visual shade matching system over 12 months. AI-assisted restorations also had minimal surface staining and marginal discoloration. While there was a higher numerical survival rate for restorations with AI-assisted shade matching, the rate was not statistically significant. These findings indicated that AI could be a valuable and standardized tool to select shades in modern resin-based restorative dentistry, especially in conjunction with controlled imaging and clinical evaluation. Further multi-center, long-term studies are needed to assess whether these esthetic advantages result in better long-term survival of the restorations.