The therapeutic relationship in in‐person, video call, and telephone psychotherapy sessions

Abstract

Objective

To examine whether and how patients' perceptions of working alliance, alliance ruptures, real relationship and affective reactions towards the therapist differ across in‐person, videoconferencing and telephone psychotherapy sessions.

Method

A total of 1175 adult patients undergoing various individual psychotherapies completed self‐report measures following sessions conducted in‐person (746 sessions), via video call (1037 sessions), or by telephone (149 sessions). Linear mixed models with marginal structural model (MSM) were employed. Interaction effects with age group, time‐in‐treatment and session frequency were also examined.

Results

Patients reported similar levels of alliance, real relationship, positive and negative affective reactions and occurrences of alliance ruptures and repairs in both in‐person and video call sessions. However, levels of alliance dimensions—agreement on therapy goals, therapy tasks and affective bond—were significantly lower in telephone sessions compared to in‐person (β = −.21 to −.25) and video call (β = −.34 to −.37) sessions. Positive affective reactions were also lower in telephone sessions compared to video call sessions (β = −.22). Age, time in treatment and session frequency significantly moderated some associations.

Conclusions

The therapeutic relationship is comparable between in‐person and video call sessions. Telephone sessions, however, are associated with lower working alliance and in‐session positive affective reactions.

Practitioner points.

• Psychotherapy delivered via videoconference produced levels of working alliance, real relationship, emotional reactions toward the therapist and rates of alliance ruptures and repairs comparable to those achieved in in‐person sessions.

• Telephone sessions were associated with modestly weaker working alliances and reduced in‐session positive affect, with some indications of stronger vulnerability among adults aged 30–49 years.

• When telephone sessions are necessary, clinicians may benefit from more explicit goal and task alignment practices (e.g., agenda‐setting, frequent summaries and check‐ins) and flexible format planning when feasible.

BACKGROUND

Before the onset of the COVID‐19 pandemic, clinical and counselling psychology was almost entirely conducted in person, with telehealth visits accounting for only approximately one in ten outpatient visits for mental health and substance use (Lo et al., 2022). However, due to pandemic‐related social restrictions (Stefana et al., 2020), most psychologists shifted to a hybrid or even exclusively remote model to treat patients during the pandemic (American Psychological Association, 2023). This shift led to a 26‐fold increase in the provision of psychological services using telecommunication technologies (Pierce et al., 2021). This modality of delivering psychological assessment, counselling and psychotherapy was found to be a valuable tool by most psychologists and psychotherapists (American Psychological Association, 2022; MacMullin & Cook, 2020) and was satisfactory for many patients (Mazziotti & Rutigliano, 2021) during the pandemic. Therefore, it is not surprising that although the number of clinicians offering exclusively teletherapy, that is, counselling and psychotherapy via telephone (TT) or videoconferencing (VT), decreased by two‐thirds from 2022 to 2023 (American Psychological Association, 2023), the percentage of those offering a hybrid option to their patients doubled in the 2020–2023 triennium, approaching 70% and showing an upward trend (American Psychological Association, 2023). However, despite such widespread adoption of teletherapy, relatively little is known about how this treatment modality impacts the therapeutic relationship.

Although current evidence suggests that teletherapy (mainly VT) and in‐person therapy (IT) appear to be equally effective in terms of treatment outcomes at post‐treatment or follow‐up and in attrition rates, as indicated by growing meta‐analytic evidence (Fernandez et al., 2021; Lin et al., 2022; Shaker et al., 2023), they represent two different types of therapeutic encounters (Aafjes‐van Doorn, 2022; Stefana, Békés, & Aafjes‐Van Doorn, 2025; Stefana, Damiani, et al., 2025). In fact, qualitative studies have found that these two modalities of providing psychotherapeutic services were experienced as ‘very different but equally rewarding’ (Shklarski et al., 2021), although altering the therapeutic relationship (McCoyd et al., 2022). For example, some patients who transitioned to teletherapy have reported an impaired bond with their therapist (Werbart et al., 2022). Similarly, some therapists reported issues in emotionally connecting with patients and building a strong therapeutic alliance with new ones (McCoyd et al., 2022). These difficulties in the therapeutic relationship appear to be related to the lack of physical presence and interpersonal contact inherent in the teletherapy setting (Lin et al., 2022; McCoyd et al., 2022; Werbart et al., 2022), which limits or totally excludes the bodily communications perceived visually or through the sense of smell by both patients and therapists.

Given that the type of therapeutic presence and the consequent range of perceivable nonverbal communication depend on the technology and equipment used by the dyad, the distinction between TT and VT becomes important in understanding possible alterations in the unfolding of the therapeutic relationship. Current evidence suggests that these two modalities of providing teletherapy treatment appear to be equally effective (Chen et al., 2024). Furthermore, beyond personal preference for one modality or the other—often linked to concerns regarding the absence of visual cues in phone call sessions versus the greater ease of use of phones (Békés et al., 2021)—current yet very limited evidence suggests that therapists' attitudes and perceptions of the therapeutic relationship are similar towards the two means (Békés et al., 2023). However, while therapists perceive the therapeutic relationship similarly in TT and VT sessions, patients' perceptions of the working alliance appear to be comparable for VT and IT care (Liebmann et al., 2019), suggesting that VT may replicate aspects of IT more effectively than TT.

ALLIANCE, ALLIANCE RUPTURES AND REPAIRS, REAL RELATIONSHIP AND OVERALL AFFECTIVE REACTIONS

The therapeutic alliance represents the cooperative relationship between a clinician and a patient, encompassing three fundamental components: (a) mutual agreement on treatment goals, (b) agreement on therapy tasks to achieve those goals and (c) development and maintenance of an affective bond (Wampold & Flückiger, 2023). Current evidence shows comparable levels of working alliance in patients receiving VT (Seuling et al., 2024) or TT (Irvine et al., 2020) compared to those receiving IT (see also Davis et al., 2024). Furthermore, there is evidence of a positive, modest but significant effect of therapeutic alliance on mental health outcomes in teletherapy (Aafjes‐van Doorn, Spina, et al., 2024), although it is smaller than that reported for IT (Flückiger et al., 2018). Notably, the task‐agreement component of the alliance has been identified as a key correlate of perceived session quality, particularly when patients are distressed (Stefana, 2026b).

An alliance rupture is a strain or deterioration in the therapeutic relationship between a patient and therapist (Eubanks‐Carter et al., 2010; Safran & Muran, 2000). It occurs when there are disagreements on the goals of treatment, challenges in collaborating on therapeutic tasks, or a weakened affective bond. These ruptures can manifest as subtle tensions or overt conflicts, leading patients to disengage, express dissatisfaction, or attempt to control the therapy process (Eubanks‐Carter et al., 2010; Safran & Muran, 2000). Repairing a rupture involves re‐establishing a strong emotional connection and resuming effective collaboration on the work of therapy and is moderately associated with positive patient outcomes (Eubanks et al., 2018). Although it has been pointed out that alliance ruptures may be exacerbated during teletherapy due to technical difficulties, dissatisfaction, or disappointment with the limited interaction (Dolev‐Amit et al., 2021), to the best of our knowledge, no studies have compared the occurrence of these ruptures and their repair in teletherapy versus IT.

The real relationship represents ‘the personal relationship existing between two or more people as reflected in the degree to which each is genuine with the other and perceives the other in ways that befit the other’ (Gelso, 2009, pp. 254–255). The strength of this non‐work connection between patient and therapist relies on both the magnitude and the valence of its two dimensions: realism and genuineness. Realism involves the extent to which one realistically perceives and experiences the other person, while genuineness pertains to the level of authenticity one displays toward the other and how one is truly oneself (Gelso et al., 2012). Contrary to sceptical expectations, during the pandemic, therapists tended to report a relatively higher quality of real relationship in their typical teletherapy sessions than in their typical IT sessions (Aafjes‐van Doorn et al., 2021; Békés et al., 2020, 2021). However, a post‐pandemic study found that therapists tend to perceive the real relationship as lower in teletherapy compared to IT (Aafjes‐van Doorn, Békés, et al., 2024), leading the authors to hypothesize that the relatively higher real relationship ratings during the pandemic likely reflected a shared challenge rather than changes in treatment format alone. To our knowledge, empirical literature on the levels of real relationship from the patients' perspective across different session formats (i.e., IT, VT and TT) is still missing.

Lastly, although the overall emotional reactions activated and experienced by patients in their relationships with therapists (Westen & Gabbard, 2002) have been demonstrated to be moderately to strongly associated with therapeutic alliance quality and session quality, no research to date has investigated the differences in the magnitude and valence of these reactions across different psychotherapy session formats (Stefana, Fusar‐Poli, Vieta, & Youngstrom, 2024).

Aims of the study

Our primary goal was to examine whether and how patients' perceptions of the levels of working alliance, alliance ruptures, real relationship and in‐session overall affective reactions towards the therapist differ across IT, VT and TT individual sessions. Based on the existing literature, we hypothesized that patients would report (a) equivalent levels of working alliance, (b) a higher occurrence of alliance ruptures, (c) a similar degree of alliance rupture repairs, (d) lower levels of real relationship, (e) similar levels of positive affective reactions and (f) higher levels of negative affective reactions in teletherapy compared to IT. We also expected greater differences between in‐person and telephone formats than between in‐person and video call formats.

Our second aim was to explore whether the association between the specific session format and each outcome was moderated by patients' age, time in treatment (months since treatment initiation) and frequency of sessions.

METHODS

Data set

This study represents a secondary analysis that utilized data from both a longitudinal investigation (Stefana, Fusar‐Poli, et al., 2023) and a randomized clinical trial (Stefana, Vieta, Fusar‐Poli, & Youngstrom, 2024). In the randomized controlled trial (RCT), only baseline data (collected before randomization) and follow‐up data from the control group were included in the analysis. The decision to combine the datasets was based on their close methodological alignment (shared recruitment sources, assessment procedures and identical measurement batteries). Eligibility criteria also overlapped, with the main difference being a minimum session‐frequency requirement in the RCT (RCT had to attend therapy at least twice a month, whereas no such requirement was set in the longitudinal study). Study design was included as a baseline covariate in the weighting and outcome models to account for any remaining differences between studies. Institutional Review Board (IRB) approval was obtained for both studies from the University of North Carolina at Chapel Hill (IRB number for the longitudinal study: 23‐0216, approved on 6 March 2023; IRB number for the clinical trial: 23‐1067, approved on 31 July 2023).

Participants

The study sample included 1175 adult participants, all of whom were undergoing various types of individual psychotherapeutic interventions for different mental health conditions. Most participants self‐identified as female (74%). Age‐wise, the largest group consisted of individuals aged 18–29 years, accounting for 39% of participants, followed by those aged 30–39 (28%) and 50 years or older (26%). White participants predominated, representing 80% of the sample. Most participants (85%) had at least one psychiatric diagnosis, with anxiety disorders (70%) and unipolar depressive disorders (59%) being the most common. Nearly half of the participants (47%) had been receiving therapy for more than 24 months and the majority attended sessions at a frequency of 2–3 per month (45%). Table S1 presents additional demographic, clinical and treatment characteristics. Clinical assessments revealed that participants had a mean Generalized Anxiety Disorder‐7 (GAD‐7) (Spitzer et al., 2006) score of 8.2 (SD = 5.3) and a mean Patient Health Questionnaire‐9 (PHQ‐9) (Kroenke et al., 2001) score of 9.9 (SD = 6.2).

Measures and covariates

A comprehensive set of self‐report measures was employed to assess patient characteristics and the dynamics of the therapeutic relationship.

Sociodemographic, clinical and treatment data

Participants completed an 11‐item questionnaire capturing sociodemographic, clinical and treatment variables (see Table S1).

Anxiety symptoms

GAD‐7 (Spitzer et al., 2006) was used to assess generalized anxiety symptoms. This 7‐item self‐report measure uses a 4‐point Likert scale ranging from 0 (not at all) to 3 (nearly every day). The scale has been shown to have strong internal consistency and convergent validity with measures of anxiety, stress, depression and worry (Kertz et al., 2013). The GAD‐7 demonstrated reliability in our sample with McDonald's ω = .92.

Depressive symptoms

The PHQ‐9 (Kroenke et al., 2001) was used to measure depressive symptom severity. This self‐report scale includes nine items, each rated on a 4‐point Likert scale ranging from 0 (not at all) to 3 (nearly every day). The PHQ‐9 has been shown to have superior criterion validity when compared to instruments like the Hospital Anxiety and Depression Scale and the WHO Well‐Being Index (Löwe et al., 2004). The internal consistency of this measure in our sample was McDonald's ω_t = .90.

Working alliance

The Working Alliance Inventory–Short Revised (WAI‐SR) (Hatcher & Gillaspy, 2006) is a 12‐item self‐report measure that evaluates the quality of the working alliance in a therapy session. It consists of three subscales, each comprising four items, which assess: (a) agreement on the tasks of therapy, (b) agreement on the goals of therapy and (c) the establishment of an affective bond between the patient and the therapist. Respondents rate each item on a 6‐point Likert scale, ranging from 0 (not at all) to 5 (completely) in the current study. The WAI‐SR has shown good convergent validity (r = .71) with the Helping Alliance Questionnaire for correlation between total scores (Munder et al., 2010). In our sample, the WAI‐SR total scale obtained a McDonald's ω_t of .97 for the whole scale, .93 for task, .93 for goal and .92 for bond subscales.

Alliance ruptures and resolution

The Post‐Session Questionnaire (PSQ) (Samstag et al., 1998) was utilized to evaluate occurrences of alliance ruptures and their resolution within therapy sessions. Section B of this questionnaire includes four items, starting with a screening question that asks whether the patient experienced any conflict, misunderstanding, or tension during the session (answered Yes or No). If the response is affirmative, three follow‐up questions are presented. These assess the intensity of the rupture (rated from 1 = low to 5 = high), the extent to which the issue was addressed by the end of the session (rated from 1 = not at all to 5 = very much), and the degree to which the problem was resolved (from 1 = not at all to 5 = very much) according to the patient's perspective. Higher rupture intensity and lower resolution have been associated with poorer ratings on the working alliance (measured using the 12‐item WAI) and session quality (using the Session Evaluation Questionnaire) (Muran et al., 2009).

Real relationship

The Real Relationship Inventory–Client‐Short Form (RRI‐C‐SF) (Stefana, Fusar‐Poli, Vieta, Gelso, & Youngstrom, 2024) was used to measure the real relationship between the patient and therapist, which consists of two subscales: Genuineness and Realism, each composed of four items. Realism refers to the extent to which one accurately experiences and perceives another person, while Genuineness pertains to the level of authenticity one displays towards others and how truly one expresses their own self (Gelso et al., 2012). The items are rated on a 5‐point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The RRI‐C‐SF has demonstrated convergence validity with the WAI‐SR (r = .59) (Stefana, Fusar‐Poli, Vieta, Gelso, & Youngstrom, 2024). Internal reliability McDonald's ω for the overall scale was .93, with subscale reliabilities of .88 for Genuineness and .87 for Realism.

Positive and negative emotional reactions

The In‐Session Patient Affective Reactions Questionnaire (SPARQ) (Stefana, Fusar‐Poli, Vieta, & Youngstrom, 2024; Stefana, Langfus, et al., 2023) was used to assess the patients' emotional reactions towards their therapist during therapy sessions. This self‐report questionnaire contains two subscales: Positive Affect (PA) and Negative Affect (NA), each composed of four items. Patients rate their responses on a 5‐point Likert scale ranging from 0 (not at all true) to 4 (very true). The PA subscale measures the patient's sense of security and ease in the therapeutic interaction, while the NA subscale captures more challenging emotions such as embarrassment, discomfort, fear of openness and concerns about the adequacy of the help being provided. The PA subscale was found to correlate strongly with the bond dimension of the WAI‐SR (r = .83) (Stefana, Fusar‐Poli, Vieta, & Youngstrom, 2024), as well as with the realism scale of the RRI‐C‐SF (r = .56) (Stefana, Fusar‐Poli, Vieta, Gelso, & Youngstrom, 2024). Similarly, the NA subscale demonstrated negative correlations with the WAI‐SR subscales for goals (r = −.42), tasks (r = −.44) and bond (r = −.46) (Stefana, Fusar‐Poli, Vieta, & Youngstrom, 2024), along with a comparable negative correlation with the genuineness scale of the RRI‐C‐SF (r = −.42) (Stefana, Fusar‐Poli, Vieta, Gelso, & Youngstrom, 2024). In this study, the reliability of the subscales was strong, with McDonald's ω = .87 for PA and ω = .78 for NA.

Comparability of assessment

All participants from both studies completed a sociodemographic, clinical and treatment questionnaire at intake. They also filled out the GAD‐7, PHQ‐9, WAI‐SR and SPARQ at each assessment point. The PSQ was administered exclusively to participants in the longitudinal study at T1. Lastly, the RRI‐C‐SF was administered at T1 in the longitudinal study and at T1, T2 and T3 in the RCT. Regarding the measurement scales used to assess elements of the therapeutic relationship, the WAI‐SR, RRI‐C‐SF and SPARQ demonstrated measurement invariance across in‐person and remote session formats (Stefana, 2026a; Stefana, Fusar‐Poli, Vieta, Gelso, & Youngstrom, 2024; Stefana, Fusar‐Poli, Vieta, & Youngstrom, 2024).

Procedures

Inclusion criteria for both studies required that participants be adults, aged 18 or older, proficient in English and actively participating in individual psychotherapy. An additional requirement for the RCT sample was a therapy frequency of at least two sessions per month.

Participants for the longitudinal study were recruited between March and April 2023, while participants for the RCT were recruited from September to November 2023. Recruitment occurred through two online patient registers, Research for Me and ResearchMatch (Harris et al., 2012). ResearchMatch, with over 158,000 registered volunteers, is a national resource developed by a consortium of academic institutions, supported by the National Institutes of Health's (NIH) Clinical and Translational Science Awards (CTSA) Program. Research has shown that volunteers from ResearchMatch generally provide accurate self‐report data, particularly when there are no monetary incentives involved (Chandler & Shapiro, 2016). Participants from ResearchMatch are also more likely to complete studies compared to those recruited via peer recruitment (Faro et al., 2021).

Informed consent was obtained electronically via Qualtrics, an online survey platform used to facilitate data collection for both studies.

Participants in the longitudinal study (N = 700) were assessed after a session within their ongoing psychotherapy (time 1 assessment; T1) and then reassessed after the very next session (time 2 assessment; T2; n = 512). The interval between these two sessions varied among individuals, with an average duration of 1 week and 5 days. In the RCT, participants (N = 475) were evaluated after a therapy session before randomization (T1) and then reassessed after 5 weeks (T2; N = 149) and 10 weeks (T3; N = 125), which did not necessarily mark the conclusion of their treatment. The number of therapy sessions attended between these assessment points varied for each participant, averaging 3.4 sessions.

In this study, anxiety (GAD‐7) and depressive (PHQ‐9) symptoms were measured over the previous 7 days instead of the typical 14‐day period.

Statistical analysis

We first estimated the association between session format (IT, VP and TT) and each outcome (WAI‐SR score and subscores, PSQ score, RRI‐C‐SF score and subscores, SPARQ scores) using linear mixed models. A random intercept was assigned to each subject to account for within‐subject correlation, and time (in weeks) and session format were included as fixed effects (β). Outcomes were standardized (i.e., mean = 0 and standard deviation = 1) across all assessments to facilitate better interpretation of the model coefficients.

Since the session format could change over time, an MSM was employed to adjust for confounding. Baseline confounders included study design (longitudinal or RCT), gender, age group, ethnicity, education level, therapist's gender, therapy location, time in treatment and frequency of sessions, PHQ‐9 and GAD‐7 scores and previous diagnoses of anxiety, bipolar disorder, depressive disorders, disruptive behaviour and dissocial disorders, eating disorders, neurodevelopmental disorders, post‐traumatic stress disorder, psychotic disorders, personality disorders (clusters A, B and C) and other disorders. The number of sessions between time points was considered a time‐varying confounder.

To implement the MSM, we calculated inverse probability of treatment weights (IPTWs) by creating a weighted pseudo‐population in which measured confounders are unassociated with the setting. These weights were calculated at each time point as the inverse probability of having a specific setting, given individuals' previous settings, the previous values of the time‐varying confounder (number of sessions between time points) and the baseline confounders. For each iii and each time t (i.e., study week), IPTWs are defined as

$${\text{IPTW}}_{\mathit{it}}={‚}_{t=0}^{\mathrm{T}}\frac{1}{\mathrm{P}\left({\text{SessionFormat}}_{\mathit{it}}=S_{\mathit{it}}\right|\stackrel{`}{{\text{SessionFormat}}_{\mathit{it}-1}},{\stackrel{`}{L}}_{\mathit{it}-1},V_{\mathrm{i}0})}$$

where the denominator denotes the probability of having a setting $S_{\mathit{it}}$ (In person, Telephone call, Video call), given the previous setting at time t – 1 ($({\stackrel{`}{\text{SessionFormat}}}_{\mathit{it}-1})$, the time‐varying confounder history up to time t – 1 (${\stackrel{`}{L}}_{\mathit{it}-1}$), and the time‐fixed (baseline) confounders $\left(V_{\mathrm{i}0}\right)$. To estimate the denominator, we fitted a multinomial logistic regression using the R package ipw (version 1.2.2; Wal & Geskus, 2011). The multinomial logistic regression had the setting as the dependent variable, while all previously listed confounders (both baseline and time‐varying) were independent variables. We truncated extreme weights below the .1st percentile and above the 99.9th percentile. To check for confounder balance, we calculated the standardized mean difference (SMD) before and after IPTW weighting using the R cobalt (version 4.6.1; Greifer, 2025). An appropriate balance was reached if the SMD for each confounder was below .1. Moreover, we inspected potential positivity issues by visually checking the overlap between IPTWs density curves within each setting (IPTWs can range from 0 to +∞). Linear mixed models were then weighted using the calculated IPTWs.

We further explored whether the association between session format and each outcome was modified by time in treatment, therapy frequency and patient age group. To this end, we added interaction terms (β_int) between session format and time in treatment, session format and frequency and session format and age group in separate models. To evaluate the statistical significance of these interaction terms, we performed an analysis of variance using the anova() function in R. All statistical analyses were two‐sided, and to account for multiple testing, we adjusted p‐values by fixing a false discovery rate (FDR) using the Benjamini–Hochberg procedure of .05, that is, it is expected that 5% of all statistically significant results are false positives. All analyses were conducted using RStudio 2025.05.1 on a Windows platform.

Statistical power

For each outcome, we conducted a priori statistical power calculations to assess the ability to detect small, medium and large effects. Effect sizes were defined using standardized regression coefficients of .2, .5 and .8, corresponding to Cohen's d values given the standardized outcomes. Power was estimated using simulation‐based methods implemented in the simr R package (version 1.0.8; Green & MacLeod, 2016) based on 1000 simulations of the specified linear mixed‐effects model and the available sample size.

Data transparency statement

This secondary data analysis was preregistered on the Open Science Framework (OSF; https://osf.io/ua3tb). The analysis code is publicly available on OSF, and the clinical data are available from the corresponding author upon reasonable request.

RESULTS

The sample comprised 1175 adult patients undergoing individual psychotherapy. In total, 1933 sessions were assessed. These sessions were attended as follows: VP = 1038 (54%; 631 from the longitudinal study, 407 from RCT study), IT = 746 (39%; 456 from longitudinal study, 290 from RCT study) and TT = 149 (8%; 97 from longitudinal study, 52 from RCT study). In Table S2, we reported the number of sessions by study, timepoint and session's setting. On average, there were 3.86 sessions (SD = 3.10) between assessment time points and an interval of 1.97 weeks (SD = 3.30) between these points.

Statistical power

In Table S3, we reported results from simulation‐based power analyses to evaluate the ability to detect small, medium and large effects. These analyses showed that for RRI, SPARQ and WAI‐SR items, the statistical power was high for IT versus VP, given any effect size. For comparison involving the TT setting, statistical power was high for medium and large effect sizes, while it was lower but still acceptable (approximately 60–70%) for small effects. In contrast, for individual PSQ items, adequate power was achieved only for medium and large effects.

Propensity score

To adjust for confounding, IPTWs were estimated for each individual. Extreme weights were truncated at the .1st and 99.9th percentiles; after truncation, IPTWs ranged from .12 to 4.42, with a mean of 1.00, indicating well‐behaved stabilized weights. Covariate balance before and after weighting was assessed using SMDs, which are reported in Table S4. All SMDs after weighting were below .1, indicating an adequate balance of measured confounders across settings. Figure S1 shows the distribution of IPTWs by setting; the substantial overlap of the weight distributions suggests adequate empirical support and no major violations of the positivity assumption.

Differences in Patient's experiences of therapeutic relationships in IT, VC and TT sessions

Patients attending TT reported significantly lower levels of positive affective reactions towards the therapist than those in VP (β = −.22, p = .037), while levels were comparable between telephone and in‐person sessions (β = −.19, p = .079).

Regarding the working alliance, patients in TT reported statistically significantly lower levels of agreement on therapy goals, therapy tasks and the affective bond with their clinician compared to those in IT (β = −.21, p = .030; β = −.25, p = .010; β = −.22, p = .017, respectively). Additionally, these working alliance dimensions were significantly lower in TT compared to VP (β = −.37, p < .001; β = −.34, p < .001; β = −.37, p < .001, respectively). These β values—by conventional interpretive guidelines (Campbell, 2021; Stefana, Békés, & Aafjes‐Van Doorn, 2025; Stefana, Damiani, et al., 2025)—represent a small (weak) effect and should be interpreted in context.

No significant differences were found between IT and VP sessions for these measures (adjusted p‐values ranging from .454 to .825). Similarly, levels of genuineness, realism, negative affective reactions towards the therapist and the occurrence of therapeutic alliance ruptures did not differ significantly across IT, VP and TT session formats (adjusted p‐values ranging from .390 to .950). Figure 1 presents the standardized β estimates (with 95% CIs) for each session‐format contrast across therapeutic‐relationship domains, while Table 1 lists the exact β values, confidence intervals and p‐values for those associations.

FIGURE 1.

Therapeutic relationship contrasts by session format.

TABLE 1.

Association between session format and therapeutic relationship measures.

Measure	n			Contrast	β [95% CI]	p
	T1	T2	T3
RRI‐C‐SF Genuineness	1175	149	125	TT vs. IT	.08 [−.07; .23]	.529
				TT vs. VP	.04 [−.1; .19]	.812
				VP vs. IT	.01 [−.06; .09]	.825
RRI‐C‐SF Realism	1175	149	125	TT vs. IT	−.03 [−.18; .13]	.827
				TT vs. VP	.01 [−.14; .16]	.950
				VP vs. IT	−.02 [−.1; .06]	.825
SPARQ PosIPive affect	1175	633	125	TT vs. IT	−.19 [−.36; −.02]	.079
				TT vs. VP	−.22 [−.39; −.05]	.037
				VP vs. IT	.02 [−.07; .12]	.825
SPARQ Negative affect	1175	633	125	TT vs. IT	.06 [−.11; .23]	.699
				TT vs. VP	.20 [.03; .37]	.060
				VP vs. IT	−.07 [−.16; .03]	.390
WAI‐SR Goal	1175	633	125	TT vs. IT	−.21 [−.36; −.05]	.030
				TT vs. VP	−.37 [−.52; −.21]	<.001
				VP vs. IT	.06 [−.04; .15]	.478
WAI‐SR Task	1175	633	125	TT vs. IT	−.25 [−.4; −.09]	.010
				TT vs. VP	−.34 [−.5; −.18]	<.001
				VP vs. IT	.05 [−.05; .14]	.529
WAI‐SR Bond	1175	633	125	TT vs. IT	−.22 [−.37; −.07]	.017
				TT vs. VP	−.37 [−.52; −.21]	<.001
				VP vs. IT	.06 [−.03; .15]	.454
PSQ Degree of tension a				TT vs. IT	.26 [−.26; .77]	.853
	112	57	0	TT vs. VP	.08 [−.43; .60]	.902
				VP vs. IT	.18 [−.17; .53]	.853
PSQ Extent issue addressed b				TT vs. IT	−.28 [−.8; .25]	.853
	112	57	0	TT vs. VP	−.05 [−.57; .47]	.902
				VP vs. IT	−.16 [−.51; .19]	.853
PSQ Degree of resolution c				TT vs. IT	−.13 [−.67; .42]	.902
	112	57	0	TT vs. VP	.06 [−.48; .59]	.902
				VP vs. IT	−.12 [−.47; .24]	.853

Note: The p‐values reported in the table are adjusted values.

Abbreviations: CI, confidence interval; IT, in‐person session; TT, telephone call session; VP, video call session.

^a Highest degree of tension felt during the session.

^b Extent to which the problem was addressed in the session.

^c The degree to which the problem was resolved by the end of the session.

Effect modification of time in treatment on the association between session format and therapeutic relationship

When considering interaction terms between session format and time in treatment (months since treatment initiation; 0–3 [reference category], 4–12, 13–24 and > 24 months), significant differences were observed between TT and the other formats for both negative affect and genuineness. No statistically significant differences were found between in‐person and video call settings on any measure across time‐in‐treatment categories (adjusted p‐values ranging from .360 to .920).

Negative affect

TT vs IT (adjusted p‐value for interaction = .040). Compared to IT, TT was associated with significantly higher levels of negative affect among participants 13–24 months into treatment (β = .87, p = .006, β + β_int = .89), while no significant differences were observed among those 0–3 months (p = .928), 4–12 months (p = .583), or ≥24 months (p = .493) into treatment.

TT vs VP (adjusted p‐value for interaction = .030). Similarly, TT did not differ significantly from VP among participants 0–3 months (p = .417), 4–12 months (p = .965) or 13–24 months (p = .321) into treatment. However, among those who had been in treatment for>24 months, TT was associated with significantly higher negative affect (β = .85, p = .007, β + β_int = 1.00).

Genuineness

TT vs VP (adjusted p‐value for interaction = .030). Among participants 0–3 months into treatment, TT was associated with significantly higher genuineness levels (β = .46, p = .008). However, this effect reversed among those 4–12 months (β_int = −.55, p = .023, β + β_int = −.09) and 13–24 months (β_int = −1.02, p < .001, β + β_int = −.56) into treatment, indicating lower genuineness for TT compared to IT. No significant difference was observed among those >24 months into treatment (p = .142).

TT vs VP (adjusted p‐value for interaction = .030). TT showed significantly higher levels of genuineness among participants 0–3 months into treatment (β = .37, p = .027), but significantly lower levels in those −12 months (β_int = −.48, p = .040, β + β_int = −.11) and 13–24 months (β_int = −.96, p < .001, β + β_int = −.59) into treatment. No significant difference was observed in those who had been in treatment for >24 months (p = .312).

Table 2 displays the beta coefficients for significant interaction effects between session format and therapeutic relationship measures influenced by time in treatment.

TABLE 2.

Effect modifications of session format with time in treatment, session frequency and patient age on therapeutic relationship measures.

Measure	Contrast	β [95% CI], p‐value	β [95% CI], p‐value	β [95% CI], p‐value	β [95% CI], p‐value	p‐int
Effect modification by time in psychotherapy treatment
		1–3 months (Ref)	4–12 months	13–24 months	≥25 months
Genuineness	TT vs. IT	.46 [.12; .80], .008	−.55 [−1.02; −.08], .023	−1.02 [−1.56; −.49], .000	−.30 [−.70; .10], .142	.030
Genuineness	TT vs. VP	.37 [.04; .70], .027	−.48 [−.94; −.02], .040	−.96 [−1.48; −.44], .000	−.20 [−.59; .19], .312	.030
Negative Affect	TT vs. IT	.02 [−.36; .40], .928	.15 [−.37; .67], .583	.87 [.25; 1.50], .006	−.16 [−.62; .30], .493	.040
Negative Affect	TT vs. VP	.15 [−.22; .53], .417	−.01 [−.52; .49], .965	.85 [.24; 1.46], .007	−.23 [−.67; .22], .321	.030
Effect modification by session frequency
		≤1 time per month (Ref)	2–3 times per month	1 time per week	≥2 times per week
Goal (Alliance)	TT vs. IT	.01 [−.42; .43], .978	−.04 [−.52; .44], .869	−.58 [−1.08; −.09], .021	.20 [−.75; 1.16], .679	.045
Effect modification by patient age
		18–29 years (Ref)	30–39 years	40–49 years	≥50 years
Positive Affect	TT vs. IT	.22 [−.12; .55], .200	−.52 [−.99; −.04], .033	−1.00 [−1.52; −.47], .000	−.31 [−.76; .14], .179	.030
Goal (Alliance)	TT vs. IT	.00 [−.30; .31], .976	−.63 [−1.06; −.20], .004	−.44 [−.91; .03], .064	.12 [−.29; .52], .566	.030
Goal (Alliance)	TT vs. VP	.05 [−.23; .33], .726	−.80 [−1.20; −.39], .000	−.35 [−.82; .11], .136	−.14 [−.52; .24], .473	<.001

Abbreviations: Genuineness, Genuineness dimension of the Real Relationship Inventory–Client‐Short Form; Goal, Goal dimension of the Working Alliance Inventory–Short Revised; IT, in‐person session; Negative Affect, Negative Affect scale of the in‐Session Patient Affective Reactions Questionnaire; p‐int = adjusted p‐value for interaction; Positive Affect, Positive Affect scale of the in‐Session Patient Affective Reactions Questionnaire; TT, telephone call session; VP, video call session.

Table 2 lists β estimates for significant session‐format × time‐in‐treatment interactions; Figure 2 then displays all FDR‐significant simple‐slope βs in a compact heatmap. All interaction models, including those without significant effects, are detailed in Table S5.

FIGURE 2.

Significant effect modifications of session format by time in treatment, session frequency and patient age. Numbers denote the contrasts between Therapy Formats within the group indicated on the x‐axis.

Effect modification of session frequency on the association between session format and therapeutic relationship

When considering interaction terms between session format and session frequency (categorized as ≤1 per month [reference category], 2–3 per month, 1 per week, ≥2 per week), significant differences were observed between IT and TT for alliance goal. No significant differences were observed between IT vs. VP and TT vs. VP (adjusted p‐values ranging from .070 to 1).

Alliance goal

TT vs IT (adjusted p‐value for interaction = .045). At the reference frequency of one or fewer sessions per month, the degree of agreement on goals did not differ significantly (β = .01, p = .978). At a frequency of one session per week, TT was associated with significantly lower goal agreement ratings compared to IT (β_int = −.58, p = .021, β + β_int − .57). No significant differences were observed at two or more sessions per month (p = .869 and p = .679).

Table 2 lists β estimates for significant session‐format × session‐frequency interactions; Figure 2 then displays all FDR‐significant simple‐slope βs in a compact heatmap. All interaction models, including those without significant effects, are detailed in Table S6.

Effect modification of patient's age on the association between session format and therapeutic relationship

Interaction terms between session format and patient's age group (categorized as young adults = 18–29 [reference category], early middle‐aged adults = 30–39, late middle‐aged adults = 40–49, older adults = ≥50) revealed significant differences between TT and the other formats for in‐session positive affect response towards the therapist and degree of agreement on therapy goals. No statistically significant differences between IT and VT across any therapeutic relationship measures were found (adjusted p‐values ranging from .141 to .974).

Positive affect

TT vs IT (adjusted p‐value for interaction = .030). TT did not significantly differ from IT among young adults (β = .22, p = .200). However, TT was associated with significantly lower positive affect among early (β_int = −.52, p = .033; β + β_int = −.30) and late (β_int = −1.00, p < .001; β + β_int = −.78) middle‐aged adults, but not among older adults (p = .179).

Alliance goal

TT vs IT (adjusted p‐value for interaction = .030). TT was linked to a significantly lower degree of agreement with the therapist on goals among early middle‐aged adults (β_int = −.63, p = .004; β + β_int = −.63), but no significant difference was observed among young (p = .976), late middle‐aged (p = .064) or older (p = .566) adults.

TT vs VP (adjusted p‐value for interaction < .001). Early middle‐aged adults in TT reported significantly lower degrees of agreement on therapy goals than peers in VP (β_int = −.80, p < .001; β + β_int = −.75), but no differences were observed among young (p = .726), late middle‐aged (p = .136), or older (p = .473) adult patients.

Table 2 lists β estimates for significant session‐format × patient‐age‐group interactions; Figure 2 then displays all FDR‐significant simple‐slope βs in a compact heatmap. All interaction models, including those without significant effects, are detailed in Table S7.

DISCUSSION

Teletherapy has become an increasingly common practice. While initial research suggests that synchronous teletherapy yields outcomes similar to those of in‐person therapy, there is a notable lack of studies examining its impact on therapeutic relationships beyond just the therapeutic alliance. Additionally, most of this limited research has concentrated on videoconferencing, neglecting comparisons with telephone‐based sessions, which could shed light on the importance of visual aspects within the therapy process. In this study, we used marginal structural models to compare patients' in‐session perceptions of working alliance, alliance ruptures and repairs, real relationships and overall affective reactions towards therapists across three types of individual psychotherapy formats: IT, VP and TT. We also examined whether these differences persisted when adding interaction terms between setting and patients' age group, time in treatment and session frequency in separate models.

A consistent finding across all therapeutic relationship domains was the comparability of IT and VP sessions: IT–VP contrasts were near zero and remained non‐significant after FDR correction, suggesting that the relational processes assessed here are largely preserved in videoconference sessions. In contrast, TT showed lower levels of working alliance and positive affect than VP, with differences of weak magnitude. However, TT subgroup findings should be interpreted cautiously, given the smaller TT subsample and the resulting lower precision.

Working alliance

Patients did not experience significant differences in the quality of their working alliance between IT and VP. This absence of differences in the perceived quality of the working alliance aligns with prior survey research on both patients' and therapists' experiences with IT and VP therapies (Aafjes‐van Doorn, Békés, et al., 2024; Schoenenberg & Martin, 2024; Seuling et al., 2024). These findings imply that the working relationship between patient and therapist in VC may closely resemble that of IT. This result holds significance for the effectiveness of teletherapy via VP, considering the proven association between working alliance and treatment outcomes (Aafjes‐van Doorn, Spina, et al., 2024; Cataldo et al., 2021).

However, we found that patients attending TT reported significantly—yet slightly—lower levels of working alliance compared to those in IT or VT, in terms of agreement on therapy goals, therapy tasks and the affective bond with the clinician. This finding contrasts with a recent meta‐analysis of comparative studies on differences in ratings of therapeutic alliance between TT and face‐to‐face psychotherapy sessions, which found no statistically significant differences based on the five studies included (Irvine et al., 2020). One possible reason for this discrepancy could lie in the type of psychotherapy delivered. On the one hand, the five studies included in the meta‐analysis were quite homogeneous, as four out of five were cognitive behavioural therapy and the fifth was solution‐focused therapy (although this study compared the ratings of the therapeutic alliance made by its participants in telephone‐delivered therapy with those from face‐to‐face counselling studies measuring the same attributes). On the other hand, our study included various types of individual psychotherapy. At this point, an important possibility is that the nature of the therapeutic alliance and its sensitivity to the session format may vary depending on the therapeutic orientation. Specifically, in goal‐ and task‐focused approaches such as cognitive‐behavioural therapy, visual contact and immediate feedback may be critical for clarifying goals and maintaining in‐session collaboration. Therefore, the observed decline in alliance scores during telephone sessions might be more pronounced in CBT‐based therapies. In contrast, approaches that emphasize transference, free association and internal processes—such as psychoanalytic or psychodynamic therapies, including those conducted without face‐to‐face visual contact (e.g., couch‐based psychoanalysis)—may involve different mechanisms for alliance formation, where session format may not influence the alliance in the same way. As our study did not distinguish between therapy types, this hypothesis could not be tested. Nevertheless, future research examining the impact of session format on alliance quality across different therapeutic orientations would provide an important contribution to the literature.

Our finding supports qualitative research's concerns about the detrimental effects of telephone‐delivered psychotherapy on the alliance. However, considering we found that (i) TT shows a noninferior real relationship when compared to patients attending VP or IT; (ii) the alliance is a working relationship between patient and therapist (Gelso & Kline, 2019) and (iii) patients rate therapists' validation of their experiences as a more important factor in the formation of the therapeutic alliance than nonverbal gestures and body language (Bedi & Hayes, 2020), we hypothesize that the detected difference may stem from TT fostering a qualitatively different type of therapeutic alliance than what is typically formed in face‐to‐face sessions. In this sense, the visual anonymity inherent in telephone therapy may foster an alliance as a construct that, while not identical, largely overlaps with the working alliance construct typically formed in sessions using a visual medium.

Furthermore, our findings indicate that the time in psychotherapy treatment did not influence the association between session format and the alliance dimensions of goals, tasks and bond. This suggests that the working alliance remains relatively stable across different stages of therapy, regardless of whether sessions are conducted in person, via VT, or by TT. This is consistent with existing meta‐analytic evidence showing that the working alliance is established early in therapy and remains relatively stable throughout (Flückiger et al., 2018, 2020; Lavik et al., 2018). While our results suggest that the working alliance remains relatively stable over the course of therapy, they do not preclude the possibility of meaningful fluctuations or ruptures that may occur at different stages. This finding challenges the common assumption that longer therapy inherently strengthens the alliance. Future research should adopt session‐by‐session tracking methods to examine the dynamic evolution of the alliance and investigate whether this pattern varies depending on therapeutic orientation or specific patient characteristics.

However, we observed that for patients attending therapy sessions once per week, those participating in TT reported significantly lower levels of agreement on therapy goals compared to those in IT. This suggests that at a weekly session frequency, the telephone format may be less effective in fostering agreement on therapy goals than in‐person sessions. More generally, our results suggest that session frequency does not significantly influence the strength of the alliance across different session formats. However, the decline observed specifically in weekly telephone sessions suggests a potential interaction between session frequency and format, which warrants further investigation in future studies. Future studies should test the common hypothesis that higher session frequency enhances the working alliance (Bruijniks et al., 2020) while infrequent sessions weaken its development and stability (Erekson et al., 2015). This can be achieved by examining potential differences in the trajectories of alliance development across different psychotherapy formats, considering not just the session level but the treatment level, and including studies in which patients are always treated within the same format—rather than adopting a mixed mode, as was the case for several patients in our sample.

Lastly, our findings reveal that adults aged 30–39 years attending TT reported significantly lower levels of agreement on therapy goals compared to those in IT or VT. To interpret these findings, it is helpful to consider two important aspects. First, adults in this age range are often starting to navigate complex life transitions, such as advancing careers, family responsibilities and personal development challenges (Lachman et al., 2015). These complexities may heighten the need for clear and collaborative goal‐setting in therapy. Second, these individuals are ‘digital adopters’ who witnessed and experienced the rapid evolution of communication technologies (Prensky, 2001). They are comfortable with face‐to‐face and digital platforms and may expect the richness of interaction that visual communication offers. According to Erikson's theory (Erikson, 1963), adults in this age range are navigating the psychosocial task of developing emotional intimacy. The lack of visual engagement in telephone sessions may interfere with this developmental need and potentially trigger feelings of isolation when mutual presence is not sufficiently established. Additionally, from a developmental perspective, these individuals may place a higher value on agency and mastery in the therapeutic process. The limited sensory input of telephone sessions may reduce their sense of control and mutual goal‐setting, potentially leading to disengagement. In contrast, older adults, who grew up using the telephone more extensively when internet‐based communications did not exist, may have different expectations. The absence of differences among the younger adults (aged 18–29 years) could be due to the small sample size, as only 28 patients in that age group reported on at least one TT. Therefore, TT, lacking visual engagement, might not meet the expectations of patients in this age group for interactive and collaborative dialogue, potentially leading to lower satisfaction and alignment with therapy goals.

Alliance ruptures and repairs

The occurrence and degree of alliance ruptures and repairs, from the patients' perspective, did not differ significantly across IT, VT and TT, regardless of session frequency or time in treatment. This finding does not support existing concerns that alliance ruptures may be more frequent in teletherapy sessions due to technical issues or dissatisfaction with the remote interaction format (Dolev‐Amit et al., 2021). This may suggest that alliance ruptures are more closely tied to relational dynamics than to technological modality per se. In other words, the quality of the therapeutic bond may be resilient across formats when core relational processes are maintained. Furthermore, when teletherapy is conceptualized not as a deficient substitute but as a qualitatively different format for alliance formation, it becomes evident that its vulnerability may have been overstated. Rather than being fragile due to the absence of physical presence, teletherapy may foster distinct yet stable forms of therapeutic engagement when the therapeutic relationship is adequately attuned.

Real relationship

The strength of both dimensions of the real relationship—genuineness and realism—was comparable across IT, VT and TT formats. This result is consistent with pandemic study evidence indicating that patients who transitioned to teletherapy perceived the quality of the real relationship as reasonably good (Békés & Aafjes‐van Doorn, 2022). Furthermore, it represents a sort of balancing point between the conflicting results of previous studies that found that the real relationship was perceived by therapists as lower (Aafjes‐van Doorn, Békés, et al., 2024) or higher (Aafjes‐van Doorn et al., 2021; Békés et al., 2020, 2021) in videoconferencing compared to in‐person but not equal.

Patients attending TT reported genuineness levels significantly higher compared to those who attended sessions IT or VT during the early stage of psychotherapy, but lower during the middle and late stages, and then comparable during the extended stage (i.e., when in therapy for at least 2 years). Realism levels were comparable throughout the various stages of psychotherapy. To interpret these results, we consider the following points: (i) the items composing the genuineness subscale of the RRI‐C‐SF relate to being oneself with the therapist, openness and honesty, the ability to communicate one's moment‐to‐moment inner experience, and the expression of feelings in therapy; (ii) evidence suggests that patients often feel more comfortable expressing their own thoughts and feelings during the first sessions, as the therapeutic relationship is still forming (McWilliams, 2004); (iii) patients and therapists co‐construct shared language and meanings during their psychotherapeutic journey (Kabrel et al., 2024; Nabzo et al., 2023); (iv) awareness of particular inner struggles increases and becomes clearer in the course of the first eight sessions of psychotherapy, and is often accompanied by feelings of sadness or pain (Kleiven et al., 2020); and (v) this openness in expressing one's own thoughts and feelings can decrease when deeper and more personal issues emerge, leading to greater resistance or hesitation (McWilliams, 2004). Based on the above, we hypothesize that the telephone format, by not involving visual contact, boosts the ease of expressing one's own thoughts and feelings during the early stage of therapy. As therapy progresses and inner conflicts and struggles emerge in the patients, the absence of reassuring visual feedback from the therapist becomes particularly inhibiting. Here, the patient's mechanism of projection—that is, attributing to others one's own unacceptable feelings, thoughts, or intentions (Cramer, 2015)—could play an important role. Notably, empirical evidence supports the importance (in terms of positive contribution to a positive relationship) of the therapist providing neutral facial expression feedback when the patient discusses a negative situation (Yokoyama et al., 2024). However, by continuing to work in therapy, the patient‐therapist couple reinforces and refines a dyad‐specific style of verbal and nonverbal communication that counterbalances the absence of visual feedback—for example, through a better understanding of the emotional significance of tone of voice or certain silences—which makes the patient feel safer and cared for, allowing them to express themselves freely. This hypothesis is valid both for therapy delivered entirely by telephone and for therapy characterized by a mixed format, such as a combination of in‐person and telephone sessions. These findings support the idea that patients' sense of the real relationship may be more resilient to format changes than previously assumed, perhaps because perceived authenticity and realism depend more on the therapist's interpersonal stance than on the medium itself. Taken together with our findings on working alliance, this suggests that while telephone sessions may modestly impact collaborative aspects of the therapeutic relationship, the personal authenticity and realism experienced by patients remain largely intact.

Overall affective reactions

The levels of in‐session positive and negative affective reactions toward therapists did not differ significantly across IT, VT and TT.

However, when considering the influence of time in treatment on these affective experiences, negative reactions were significantly higher in TT compared to IT and VT during the late stage of therapy (i.e., between 13 and 24 months from its beginning), but not before or after this period. Considering the content of the items on the negative affect scale of the SPARQ—which include feeling ashamed with the therapist about one's own fantasies, desires, mindset, behaviour, or symptoms; worrying that the therapist couldn't help; feeling shy as if one wanted to hide from the therapist; and feeling afraid to speak one's own mind for fear of being judged or criticized by the therapist—we interpret this finding as a confirmation of our hypothesis about the trajectory of the genuineness dimension of the real relationship. While the levels of genuineness remained significantly lower during both the middle and late stages of therapy, the levels of negative affective reactions towards the therapist were significantly high only during the late stage. However, it should be noted that the levels of genuineness experienced a substantial reduction from the middle to the late stages (the magnitude of β decreased by approximately half), possibly reaching a level low enough to avoid eliciting statistically higher levels of negative affect.

Furthermore, age moderated the association between session format and positive affect: TT was associated with lower positive affect than IT specifically among patients aged 30–49 years, whereas no overall TT–IT difference emerged in the full sample. One possible explanation is that adults in this age group may have greater expectations for in‐depth interpersonal interaction and rely more heavily on visual cues and body language, which are limited in TT. Middle‐aged adults, in particular, may value face‐to‐face interaction to navigate complex life challenges often encountered during this life stage, such as career transitions or family responsibilities (Lachman et al., 2015).

Clinical implications

Our findings suggest that VP can preserve key relational processes at levels similar to IT, as patients reported no significant differences across working alliance, alliance ruptures and repairs, real relationship and in‐session affective reactions. This supports continued integration of VP within hybrid models when IT is impractical, without assuming a systematic ‘relational cost’ in the domains assessed. Meanwhile, TT showed lower positive affect than VP overall, while TT–IT differences in positive affect were not observed in the full sample but emerged in specific age groups (30–49 years). Importantly, TT also demonstrated context‐dependent vulnerabilities that suggest it may be less optimal as a stand‐alone format when treatment relies heavily on sustained, explicit collaboration on goals and tasks or when therapy enters emotionally demanding phases. When TT is necessary for access reasons, clinicians might consider using broad strategies to offset reduced nonverbal cues—such as more explicit agenda‐setting, frequent summaries and routine checks of goal/task alignment and relational comfort—and, when possible, adopt flexible format planning (e.g., switching to VP or IT during phases when collaboration or affective safety appears more vulnerable).

Limitations

The findings of this study should be considered in light of some limitations. First, the study relies solely on patients' reports. While patients' perceptions are essential in psychotherapy, they represent only one facet of a complex dynamic system. Existing research on real relationships and working alliances has demonstrated the importance of dual‐informant measurement methods that include both patient and therapist perspectives (Hill et al., 2024). Therefore, relying exclusively on the patient perspective may introduce potential bias in the measurement of the constructs. Second, and related to the previous point, all variables were self‐reported responses. Although self‐report data are fundamental in psychotherapy research, standardized subjective self‐report measures carry inherent risks of bias due to factors such as limited self‐insight, response tendencies, and cognitive biases (Schwartz et al., 2023). However, self‐report measures provide fundamental information on patients' perceptions and internal experiences of the therapeutic relationship, which are critical for understanding the therapeutic process. Third, although we operationalized the late stage as the period between 13 and 24 months of treatment, it is important to note that temporal markers may not fully capture the qualitative depth of the therapeutic process. Fourth, the number of TT sessions was significantly lower than the number of IT and VP sessions, which limits the accuracy of TT estimates and decreases statistical power to detect small effects—especially in moderation analyses. Accordingly, subgroup findings involving TT should be interpreted cautiously and are best considered exploratory. Fifth, we did not collect information about the specific type of psychotherapy intervention received. As a result, we were unable to explore whether our findings are influenced by the specific type of psychotherapy, which could have a slightly differentiating effect on the quality of the therapeutic relationship (Heinonen et al., 2022; Prusiński, 2022). However, we did gather information on session frequency and time in treatment and tested their influence on the tested components of the therapeutic relationship. Sixth, although the current analysis pooled data from two parent studies, these studies were highly aligned methodologically (sharing recruitment sources, assessment procedures and identical measurement batteries), and study design was included as a baseline covariate in the weighting and outcome models. Nonetheless, pooling may still result in residual heterogeneity that is not fully explained by measured covariates. Seventh, as with all observational analyses, residual confounding due to unmeasured factors (e.g., clinical complexity, therapist‐level characteristics, or reasons for choosing a specific format) cannot be ruled out, even though we used IPTW to enhance covariate balance. Lastly, the cultural generalizability of these findings is limited. Our sample was relatively homogeneous (80% US‐based White people), and prior literature indicates that cultural values and intercultural dynamics (e.g., norms around self‐disclosure, expectations for collaboration vs. therapist directiveness and potential unintentional bias) can meaningfully shape therapeutic relationships. Additionally, we didn't collect information about therapists' ethnicity.

CONCLUSION

Our study found that patients reported similar levels of working alliance, real relationship, positive and negative affective reactions towards the therapist and occurrences of therapeutic alliance ruptures and repairs in both IT and VP. However, levels of working alliance were significantly lower in TT compared to IT or VP, regardless of session frequency or time in treatment. Additionally, the levels of genuineness in the real relationship were higher among patients attending TT during the early stage of psychotherapy, lower during the middle and late stages and became comparable after two years of treatment. These findings support the equivalence in the quality of the therapeutic relationship between IT and VP and suggest that clinicians may need to adjust their assessment and intervention approaches when delivering psychotherapy sessions by telephone.

AUTHOR CONTRIBUTIONS

Alberto Stefana: Conceptualization; methodology; investigation; funding acquisition; writing – original draft; supervision. Teresa Fazia: Conceptualization; formal analysis; writing – review and editing; visualization. Andrea Nova: Conceptualization; visualization; writing – review and editing; formal analysis.

FUNDING INFORMATION

The data analysed in this report were collected as part of a study funded by the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska‐Curie grant agreement No. 101030608.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Table S1.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.