Reimagining Informed Consent: AI Avatars in the Era of Complex Clinical Trials

How AI Can Help Improve Informed Consent

In recent years, efforts to improve the efficiency and effectiveness of informed consent in clinical trials have been sought, propelled by the needs created by the COVID-19 pandemic. Although technological platforms have facilitated efficient ways to collect consent, participant comprehension remains a challenge. As clinical trial designs become more complex by incorporating adaptive protocols or decentralized elements, studies demonstrate that fundamental information is often not well understood by participants.1 For example, a trial participant and attorney shared how the lengthy and complex language of current consent forms led her to overlook important information, despite her legal expertise.2

This anecdote underscores a proposed mandate of the US Food and Drug Administration (FDA) that Informed Consent Forms (ICFs) should start with key information—the essential details that help potential participants decide to join a study.3 Clearly, there is a need to simplify consent language, but there is also potential for modern technologies to support equitable education of participants. In particular, synthetic avatars powered by artifficial intelligence (AI) may support participants’ true understanding of trial aspects. This article describes how an AI-based synthetic avatar can assist site operators and clinical research organizations (CROs) during the recruitment and informed consent process in providing a multimodal experience via video and audio.

Leveraging ISPE GAMP® 5: A Risk-Based Approach to Compliant GxP Computerized Systems (Second Edition) and its Appendix D11,4 the ISPE GAMP® Good Practice Guide: Validation and Compliance of Computerized GCP Systems and Data—Good eClinical Practice (Second Edition) [5], and the FDA’s recent draft guidance6 on enhancing informed consent in clinical research, we aim for safe and effective use of an AI-based solution. We also discuss recurring challenges, ethical considerations, and the benefits to various stakeholders—including patients, Contract Research Organizations (CRO)s, sites, and sponsors.

Informed Consent Purpose and Elements

An ICF serves two main purposes: First, it explains aspects of clinical trials to the patient, and second, allows the patient the option to agree to its conditions. Hence, it should provide the relevant information to make this decision in a fair, unbiased, and informative manner, suitable for the typical knowledge level of a patient.7

Typical elements of an ICF include:

• Purpose of the trial: Explains the goal of the research, including details about the investigational product.
• Details of the trial: Covers aspects such as the trial’s duration, the role of alternative therapies or placebos, and any procedures involved.
• Potential benefits: Describes possible benefits for the participant, while fairly acknowledging associated uncertainties.
• Rights: Outlines legal considerations, including data privacy, regulatory inspections, and the participant’s right to withdraw at any time
• Risk and discomforts: Lists potential side effects or adverse events that may occur, presented clearly and in an informative manner based on current evidence.

With fair treatment and effective communication as key objectives, patient-centered and culturally sensitive approaches are of high relevance. Consent forms should be tailored to the diverse needs of study participants, fostering an inclusive environment in which all participants are truly informed about trial procedures.

These considerations are even more important when aiming for a high degree of diversity in trials in the realms of rare diseases and personalized medicine, where trials tend to be smaller and more specialized.8 Although decentralized clinical trials offer a partial solution by reducing the need for travel and potentially increasing trust through more personalized engagement, establishing trust through effective patient information also needs to be considered, especially when decentralized layouts lack direct human interaction.

Technology Advancements

Supporting ideas presented in this article, two important technological advancements are considered with both potential in Good Clinical Practice (GCP) and other guidelines and regulations (GxP and non-GxP) of the pharmaceutical, biotech, and life sciences industries. These include electronic informed consent (eConsent) platforms and AI-based synthetic avatars.

eConsent Platforms

eConsent platforms9 (see also FDA guidance) allow for more dynamic and interactive consent processes.10 Through these platforms, informed consent processes may adapt according to the user’s comprehension and feedback, addressing individual needs and enhancing understanding. In this context, further regulation and guidance such as the European Union’s General Data Protection Regulation (GDPR)11 or changes to the US Common Rule12 should be considered regarding data protection and privacy, as well as ethical aspects.

AI-Based Synthetic Avatars

AI-based synthetic avatars are digital constructs that represent humans, characters, or entities in virtual environments. These are typically created using a combination of 3D modeling, animation, and AI technologies. After designing the avatar’s visual human-like appearance, dynamic behavior and movement are enabled by modeling a skeletal system close to real-world behavior. Programmers and animators ensure that the avatar can display a range of motions and expressions.

Synthetic avatars can be further augmented to enable interactive behavior, responding to input in real time. For instance, AI components process inputs from users or the environment and respond in a way that mimics human reactions. In more advanced applications, such as customer service bots or virtual assistants, avatars engage in conversations with humans. Synthetic avatars are currently evolving, enhancing their realism and responsiveness. Avatar synthesis is available as a software-as-a-service, for utilization in various contexts of use. Such use requires careful consideration of GxP requirements in areas like clinical trials, as demonstrated throughout the article.

Managing Informed Consent: Key Challenges

Challenges in managing informed consent in clinical trials affect participant understanding and ethical considerations. Primary challenges in informed consent include poor understanding, growing complexity, and lack of respect for autonomy. Addressing these challenges through a combination of educational enhancements, technological integration, and clear communication strategies holds great potential for fairer and more inclusive clinical trials.

Biesecker et al. have demonstrated this potential in a study,13 evaluating the effectiveness of an interactive computer-based informed consent process compared to traditional paper-based methods. The researchers found that participants who used the interactive system showed significantly better comprehension of study information, including details about randomization and potential side effects.

Limited Understanding of Basic Trial Concepts

Many studies indicate that some participants only have a poor understanding of even the basic aspects of the clinical trials they join,1 sometimes participating out of altruism rather than informed decision-making. An example of this challenge is the “therapeutic misconception,” which means that participants mistakenly believe that the clinical trial is designed to address their personal medical needs. This misconception leads them to overestimate the benefits and underestimate the risks involved.

Growing Complexity

Modern clinical trials may include decentralized elements, intricate treatment approaches, and adaptive study protocols. At the same time, participants expect to receive clear, truthful information about the trial’s purpose, procedures, alternatives, and potential risks and benefits. Often, the technical jargon and volume of information can be overwhelming.

Lack of Respect for Autonomy

Informed consent is fundamental for ethical clinical trials because it respects participants’ rights to make their own decisions. However, true autonomy is achieved only when consent is fully informed. Thus, facilitating consent is provided entirely voluntarily, free from undue pressure or deception.

Key Findings

Key findings from this study include:

• Participants using the interactive system spent more time reviewing the consent information.
• The interactive group showed better understanding of the study, with higher quiz scores on consent-related questions.
• Participants reported higher satisfaction with the interactive consent process.

To overcome these challenges, this article focuses on patient-centered forms of providing information, leveraging innovative technology like synthetic avatars when explaining key trial elements as a preparatory step during recruitment or when educating the patient before an interview with clinical staff and obtaining formal consent.

Taking a Life Cycle Approach

As described in ISPE GAMP® 5: A Risk-Based Approach to Compliant GxP Computerized Systems (Second Edition),4 a thorough life cycle approach to computerized systems provides a robust foundation for safe and effective support of GxP-regulated processes. Integrating life sciences validation concepts and data science methods to achieve fitness for the intended use of the system, the guide’s Appendix D11 has established a dedicated life cycle model, including three key phases: concept, project, and operation.

This case study’s main focus is the concept phase, with an outline of the subsequent project phase. Taking a patient-centered design and implementation approach is required in the context of informed consent, keeping in mind the particular needs of patients as direct end users of the system. In addition, regulatory requirements in a GCP context need to be considered, including Institutional Review Board (IRB) review and approval.

A particular point of interest in this case study is the close collaboration between a site owner and a solution provider collaboratively exploring the potential of synthetic avatars in informed consent. Following this collaborative approach, an implementation roadmap was created. These insights may serve as input for further innovation initiatives in GCP and the use of AI.

Concept Phase

The concept phase includes the business opportunity or challenge to define the intended use, identify data and modeling options, perform initial risk assessment activities, and may include a prototype to demonstrate technical feasibility.

Project Phase

The project phase includes the specification of model requirements, preparation of data including data engineering and data splitting, and iterative experimentation and fine-tuning of models to derive a release candidate and perform formal acceptance testing.

Operation Phase

The operation phase, which guides the AI-enabled system throughout subsequent iterations, leverages data for ongoing monitoring and change management to maintain a validated state and maximize potential for improvement.

Understanding Participant Needs in Informed Consent

Building on the previous section’s exploration of challenges in the informed consent process, a more participant-centered, adaptable approach is required to address the key issues identified.14In summary, participants often face barriers to understanding the trial due to complex language, static formats, and a lack of personalized engagement, which can diminish their ability to make fully informed decisions and reduce overall trial retention.

Taking a patient-centered view, key challenges that participants face are described in more detail in this section. Understanding these barriers is essential for developing a patient-centered approach to avoid possible confusion and mistrust or hesitance in participation.

Excessive Length

ICFs are frequently criticized for being too lengthy, which discourages thorough reading and leads to poor comprehension. Studies show that longer documents can overwhelm participants, reducing their understanding and retention of key information. For example, a study published in the Journal of Patient Experience15 found that rural patients are underrepresented in cancer trials, and a study comparing short (2,000-word) vs. long (6,000-word) consent forms found that among 240 patients, 89 (37%) were rural. Seventy-one (80%) rural and 117 (77%) nonrural patients signed a consent form of any length (P = .68). Forty-one of 47 (87%) rural patients signed a short consent form; in contrast, 30 of 42 (71%) signed a long form. This underscores the importance of how simplifying ICFs could improve accessibility, understanding, and engagement, particularly among underrepresented populations.15

High Complexity and Readability

In addition to length, the complexity and readability of ICFs present a significant barrier. Most ICFs are written at or above a tenth-grade reading level, far exceeding the recommended eighth-grade level, which makes them difficult for many participants to understand (further information on grade systems can be found at “Good Calculators.”16) This challenge is further compounded in low- and middle-income countries, where high illiteracy rates and the use of vernacular languages add another layer of difficulty.

Legal Protection Focus

Although simplifying language can aid in readability, it is equally important to shift away from the legal-eavy focus that dominates many ICFs. Often, these forms are primarily designed to protect institutions, investigators, and sponsors legally, rather than to ensure clarity for participants. This focus on legal protection leads to overly complex clauses that may seem confusing or even unfair to participants.

Mistrust and Anxiety About Clinical Trials

This need for transparency becomes apparent through the lack of familiarity with the process, and prior negative experiences contribute to participant anxiety and hesitation. This stigma around clinical trials, particularly for those who have not previously participated, often acts as a barrier to recruitment. For example, a study on different types of distrust in clinical research among White and Black Americans found that Black Americans were significantly more likely than White Americans to believe they could be “used as guinea pigs” without their consent (54% vs. 28%).17

Common to these challenges is the divide between sponsors, CROs, and other healthcare organizations on the one hand and patients on the other, with differing levels of experience and domain expertise. Thus, proactively understanding these challenges and addressing the diverse needs of participants through new approaches to the communication of information during informed consent holds high potential for fairer clinical trials.

Consent Strategies for Diverse Populations

Addressing challenges and needs identified in the previous section is at the core of selecting a solution approach, addressing key challenges in understanding and agreeing to risks, obligations, and responsibilities as provided in ICFs. In addition to general aspects, such as the use of simpler language, a more engaging presentation of information in informed consent through video and audio elements may help demystify complex aspects of clinical trials. This could enable participants to better understand the purpose, risks, and protocols involved.

Such methods of presentation may also involve personalized elements, tailored to the needs of subpopulations or even on a patient level. With the goal of providing further, more engaging and interactive means of presentation of content, synthetic avatars were identified as a way to achieve a more engaging experience in providing content. These avatars can articulate content, including mimics and gestures, and allow for a multi-channel experience instead of plain text.

In this context, a study published in JMIR Formative Research [18] found that participants who used short-form, video–based eConsent demonstrated higher satisfaction (83.2% vs. 76.3%, P = 0.047) and improved understanding of risks, privacy, and procedures compared to those using traditional written consent forms. This supports the potential value of incorporating multimedia tools, such as synthetic avatars, to further enhance engagement and comprehension in the informed consent process.

The complexity and heterogeneity of approved study materials and cultural preferences need to be reflected to promote inclusivity and reduce the stigmas that may discourage participation, and to help achieve diverse patient populations in trials. Therefore, the avatar should match the target population in ethnicity and gender, whereas further configuration elements, such as the voice and tone, should be selected to mirror the professional environment of clinical trials. Further design elements and requirements were established:

• Provide information in sections, with stepwise coverage of the ICF’s content.
• Patients must not be able to skip content. They must confirm understanding of information before proceeding.
• Patients may be able to repeat passages and sections, as per their understanding and the complexity of content.

Considerations for Business Process Integration

This solution is applied at the intersection of participant recruitment and the informed consent process, extending to boundary activities such as registration and participant education or interviews with an investigator or clinician. To effectively integrate the digital solution within the clinical trial process, the following aspects were considered:

• Synthetic avatar videos are prepared based on the ICF, tailored to a set of patient subpopulations. These videos are embedded in a solution that guides the patient through education on the informed consent information. To this end, avatar synthesis was used as a service, configuring the synthesis behavior without fine-tuning the AI components generating the avatar.
• The solution enhances initial participant engagement by simplifying registration when individuals first encounter information on the clinical trial, leading to a more streamlined participant experience.
• Following registration, the approach aims to support education through interactive modules, including controls such as ensuring coverage of the complete ICF. This approach not only facilitates a deeper understanding, but also ensures that all interactions are grounded in clarity and compliance.
• The informed consent process is concluded by an interview with an investigator or clinician, with patients being better prepared so that the focus is shifted to more personally relevant elements like potential benefits and risks for the patient.

Stakeholder Engagement and Perspectives

Integrating synthetic avatars offers a wide range of benefits for both patients and clinical trial stakeholders.

• Patients: ICFs facilitated by avatars can make complex information more accessible and can help patients better understand study protocols, thereby increasing engagement and comprehension. As avatars can explain procedures at the user’s preferred pace, patients’ anxiety may be lowered.
• CROs and Site Management Organizations (SMOs): By educating patients more effectively, avatars streamline the consent process, reducing the time and resources typically spent on participant education. This efficiency improves participant retention and allows CROs and SMOs to manage trials more effectively, showing strong support for sponsor goals and minimizing delays or risks due to misunderstandings.
• Clinical staff: When patients are well-informed about trial procedures, clinical staff can dedicate less time to addressing patient concerns or confusion. This enables them to focus on more personalized aspects in support of the participant or productive research activities rather than clarifying procedural details.
• Private site owners: Following innovative approaches may offer smaller, diverse private sites a chance to differentiate themselves from the competition in a competitive marketplace.
• Sponsors: For sponsors, AI-driven consent solutions like avatars lead to cost savings and resource efficiencies by accelerating the trial pipeline and improving participant retention.

Grounded in GCP guidelines and the ISPE GAMP® Good Practice Guide: Validation and Compliance of Computerized GCP Systems and Data – Good eClinical Practice (Second Edition),4 this multi-stakeholder approach illustrates how avatar-supported informed consent can enhance trial quality, participant experience, and operational efficiency across clinical research settings. In particular, the risk of participants having incomplete or misunderstood information may be reduced, which promotes patient safety, increases efficiency, and mitigates the technical risks of the study.

A Risk-Based Approach to Avatar Implementation

A risk-based approach allows for the focus of resources on relevant areas, taking a forward-looking approach based on process understanding. The possible impact on patient safety at the initial field study level with a limited number of patients, hazards, potential risk impacts, and mitigation strategies has been identified for the synthetic avatar implementation, and examples are displayed in Table 1. Here, we follow a user-centric approach with particular focus on the patient, as recommended by the eClinical Good Practice Guide.5

Ethical Considerations and IRB Submission

Given the selection of risks described in the previous section, implementing an avatar-based informed consent solution requires careful attention to ethical concerns and adherence to GCP principles. As with standard advertising processes, scripts for avatars are developed following rigorous GCP guidelines and are submitted to institutional review boards for review. These scripts are quality controlled to align with approved materials, ensuring that content remains accurate, clear, and ethically sound. Only after institutional review board approval are the final synthesized videos delivered and ready for patient use. In this context, the following aspects are of high relevance.

• Transparency and disclosure: Patients should be fully informed of the avatar’s purpose and role in the consent process, emphasizing that this approach supplements, rather than replaces, discussions with clinical staff.
• Accessibility options: To accommodate diverse needs, the avatar solution should provide both audio and video formats, allowing patients to choose their preferred way of accessing information. This multimodal approach addresses participants with varying levels of literacy, language skills, or accessibility requirements.
• Patient diversity and representation: Recognizing the importance of patient diversity, avatars should be tailored to reflect the cultural and demographic characteristics of study populations. For instance, a Black avatar with an authentic, culturally resonant voice can be used to better connect with Black patients to match their background, fostering a sense of familiarity and reducing perceived barriers to participation.

Incorporating these considerations, the integration of synthetic avatars in the informed consent process supports both ethical transparency and patient-centered communication, ensuring that the technology meets the highest standards of accessibility and cultural compatibility.

Prototype Implementation

Building on previous design, ethical, regulatory, and cultural considerations, the concept phase continues with the development of a prototype. This prototype combines selected components designed to enhance patient understanding and engagement while adhering to GCP guidelines. The selection process, quality assurance measures, and site preparations were highly relevant to ensure that the prototype is fit for purpose. The following elements have been considered.

• Selecting and customizing components to meet patient needs: Key design components—such as avatar’s appearance, script language, and interactive features—were carefully chosen to support clarity, accessibility, and patient-centered communication. Special attention was given to verifying that the avatar’s mimicry and pronunciation meet high standards of authenticity and clarity. Pronunciation, tone, and facial expressions were chosen to convey empathy and professionalism, improving participant trust and comprehension. Thorough testing ensured meeting stated objectives, considering risks as outlined previously.
• Maintain alignment and consistency with approved ICF content: The avatar’s spoken content and visual representations were aligned with the approved ICF. Every script follows the exact language of the approved materials, with adjustments made for simplicity and clarity where appropriate. The outcome was reviewed by the quality assurance unit to verify completeness and accuracy, thus providing a reliable resource for participants consistent with subsequent steps in the process and the original ICF.
• Site preparations and testing: Before deployment, site preparations were conducted to ensure that the prototype integrates smoothly within the clinical process. This includes configuring the system to support seamless access to the avatar solution, training the site staff on effective use, and establishing supporting information to address participant questions or technical issues. Regular checks were conducted to ensure that the avatar performs reliably, with responsiveness to any errors or inconsistencies that could affect patient comprehension and safety.

The prototype implementation resulted in an avatar solution that meets high standards of accuracy, accessibility, and ethical transparency. By embedding these measures into the development of the avatar tool, the prototype solution lays a strong foundation for patient-centered communication that supports participants’ understanding and trust throughout the informed consent process. Data collection for initial use of the solution is described in the next section.

Field Study Results

Data were collected to evaluate the effectiveness of the avatar-supported prototype in supporting informed consent processes. Both quantitative and qualitative data were gathered, including comprehension levels, engagement metrics, retention intentions, and efficiency in the consent process to assess how well the avatar tool supports patient understanding, satisfaction, and retention in clinical trials.

As displayed in Figure 2, the synthetic avatar explanation provided benefits to patients:

• Patients preferred the synthetic avatar, with no reports of a “fair” or lower rating, unlike the paper-based informed consent.
• Patients reported the avatar to be more engaging, with a majority of responses expressing that the avatar was very engaging, whereas the majority of patients reported the paper-based informed consent to be either not engaging or neutral.
• 100% of respondents found the AI avatar format to have an adequate length, compared to 60% who did so for the paper format.
• All participants expressed they felt very confident in deciding whether to participate in the study when having access to the synthetic avatar, whereas 40% reported some level of hesitance (“somewhat confident”) when only having access to the paper-based ICF.

Even though the number of patients is rather low (n = 5), the data align with findings from the JMIR Formative Research study.18 This study highlights the role of short form video–based eConsent tools in enhancing patient-centered communication and informed decision-making in clinical trials. The preliminary data and prototype approach have demonstrated significant potential to refine and expand the solution, maximizing its effectiveness in the informed consent process.

Challenges and Considerations

As the AI-powered avatar solution for informed consent progresses toward broader implementation, several challenges are anticipated and require careful planning.

Coverage Across Diverse Trial Complexities

One significant challenge is ensuring that the AI-powered avatar solution can adapt to various levels of trial complexity. Different protocols may necessitate unique or adaptive consent elements that require the same level of flexibility in the avatar approach to ensure accuracy in conveying the trial information. Application across a range of protocols will help improve its adaptability, enabling it to support diverse informational needs and foster greater retention by providing clear, protocol-specific information. Combining the use-case scenarios with a comprehensive functional risk assessment form should create the basis for a testing strategy. This will demonstrate the fitness for intended use for wider application of the synthetic avatar solution.

Language Barriers

Providing language support for less frequently spoken languages poses a challenge for the avatar synthesis technology in ensuring inclusivity and accessibility in the consent process, given the high-quality required. If these language needs are unmet, there’s a risk of alienating participants from specific demographics, potentially affecting trial enrollment and participant diversity. Integrating translation capabilities and partnering with language experts should help the avatar provide clear communication for all subpopulations and patients.

Consent Management

Managing multiple versions of the consent process for various subpopulations is essential for maintaining compliance and patient understanding. Thorough version control is required to manage the baseline synthesization model, content, and configurations for various trials. Monitoring should ensure that the chosen configuration for a trial yields the expected outcomes. This should allow for the leveraging gathered data for planning the application of the synthetic avatar in further trial settings.

Established eConsent Platform Integration

For the AI-powered avatar solution to enhance rather than disrupt the informed consent workflow, it must integrate seamlessly with existing eConsent platforms.

Supplier and Partner Relationship Management

While providing a specific application for use in a clinical trial setting, the synthetic avatar solution will be dependent both on supplied components (like the synthesization model) and integration with eConsent platforms. Robust supplier and partner management, including change management and incident management processes, are important to maintain the fitness for intended use.

Legal Considerations for Data Capture and Storage

The avatar’s interactions with patients involve patient-specific data, which raises legal considerations around data capture, storage, and usage. Failure to comply with privacy regulations could undermine patient trust and compromise trial compliance. Furthermore, such data is of high relevance to assess the effectiveness of the avatar or identify weaknesses.

Jurisdictional Variability in Regulatory Expectations

With clinical trials often spanning multiple jurisdictions, each with unique regulatory requirements, the synthetic avatar solution must be adaptable to meet these diverse standards. Failure to address jurisdictional variations could lead to regulatory noncompliance, which could potentially impact trial outcomes. By proactively addressing these challenges and drawing on successful approaches from similar projects, the team is laying the foundation for a scalable, adaptable, and compliant AI-powered avatar solution.

Conclusion

The integration of AI methods into informed consent processes—coupled with expert oversight and thorough quality assurance processes aligned with standards set by the respective institutional review board and ethics committee—offers substantial potential in clinical trials. With the goal of improving participants access to ICF information, participant comprehension, engagement, and satisfaction can be improved. These objectives were verified in a field study, providing quantitative and qualitative insights into the real-world benefits of the use of synthetic avatars in the informed consent process. These findings indicate that the approach may support the success, diversity, and ethical integrity of clinical trials, contributing to more inclusive and representative research that benefits a broader range of participants.

In addition to the use case explored in this article, AI-enabled approaches may not only elevate the participant experience but also open avenues for more dynamic engagement throughout the entire clinical trial process. For instance, interactive avatars can support participants throughout the trial by guiding them through study protocols, explaining risks and benefits, and answering common questions. This interactive model may help participants better understand and adhere to the study, ensuring they feel supported at every stage.

However, thorough data and risk management are required in environments such as clinical trials. Control measures should be considered (e.g., involvement of human experts when interactive avatars hit limitations) and avatar interactions should be regularly reviewed to reduce the risk of errors and implications for patient safety or data integrity. Considering further innovative approaches like interactive avatars will require thorough evaluation, demonstrating fitness for intended use across a broad range of scenarios, thus necessitating a representative test data set.4

Adopting these innovative approaches requires a practical implementation approach and rigorous quality considerations, e.g., by leveraging recently published guidance as in the ISPE GAMP® Good Practice Guide: Validation and Compliance of Computerized GCP Systems and Data – Good eClinical Practice (Second Edition).5

Looking forward, the potential of AI in clinical trials extends beyond informed consent. AI-driven systems may support more patient-centered clinical trials by playing a central role in trial engagement, safety monitoring, and participant support, paving the way for more effective, ethical, and inclusive trials that benefit all stakeholders involved.

Acknowledgments

The authors would like to thank the ISPE GAMP Global Software Automation and Artificial Intelligence Special Interest Group for its support in the creation of this article. We would like to thank Keith Berelowitz, Ebone Jenkins, Lia Hunter, Jennifer Lechner, Ulf Menzler, and Laila Rasmy-Behket for their review of this article and insightful discussions during the writing process.