CharacterChat: Supporting the Creation of Fictional Characters through Conversation and Progressive Manifestation with a Chatbot

• CharacterChat targets a gap in existing tools, which mainly track established characters rather than supporting ideation or bringing new characters “alive”.Source: §1 · p. 1
• Progressive Manifestation has writers turn the chatbot into a fictional character by chatting with it.Source: §1 · p. 2
• Guided Conversation Mode uses rule-based prompts and concept-network suggestions to help writers define character attributes and values.Source: §1 · p. 3
• Open Conversation Mode uses a dialogue model conditioned on elicited character information, allowing writers to explore the character and seek inspiration.Source: §1 · p. 3
• Writers can interweave the two modes to define a character and explore questions about its traits, experiences, and actions.Source: §1 · p. 3
• The system was developed through a survey of 30 writers and two qualitative user studies with 7 and 8 participants, producing user-centred design insights.Source: §1 · p. 3

2 INFORMING AND DEVELOPING THE CHARACTERCHAT CONCEPT

• The concept was developed iteratively by combining Media Science and Narratology with creativity-support tools and conversational HCI, then informing design choices through a writer survey.Source: §2 · p. 3

2.1.1 Creating Characters in Storytelling.

• Character-creation theory motivated prompts covering physiology, psychology, sociology, multiple qualities, inner life, development, goals, and decisions.Source: §2.1.1 · pp. 3–4

2.1.2 Creativity Support Tools for Writing.

• Its design addresses thinking through associations, exploring a conversational character simulation, and composing a character step by step.Source: §2.1.2 · p. 4
• CharacterChat extends creativity-support tools by using an AI agent for ideation and character simulation rather than automatically generating stories.Source: §2.1.2 · p. 4

2.2 Empirical Background: Initial User Research

• The survey included 30 writers and gathered writing backgrounds, character-creation approaches, and attributes beyond those recommended in the literature.Source: §2.2 · p. 4, §2.2.1 · p. 4

2.2.2 Results.

• Writers described diverse starting points and character information, including social life, appearance, motivation, personality, quirks, background, thoughts, interests, values, and story function.Source: §2.2.2 · p. 5, §2.2.3 · p. 5
• Reported problems included stereotypical or indistinct characters that were too close to templates or inspirations, or too flat and shallow.Source: §2.2.2 · p. 5

2.2.3 Discussion & Implications.

• Because writers cover many attributes but do not all consider everything, the findings indicate an opportunity to support exploration of attributes for creating rounder, more complex characters.Source: §2.2.3 · p. 5

3.1 User Interface and Interactions

• The interface contains a chat view, a character view listing defined attributes, and a pinboard for bookmarked dialogue pieces.Source: §3.1 · p. 5

3.2.1 Understanding Intents.

• DialogFlow recognises user intents for guided interactions and switching between guided and open conversation modes.Source: §3.2.1 · p. 6
• Guided Conversation Mode randomly suggests undefined attributes from a 31-item list based on the survey and literature, while allowing users to enter values freely.Source: §3.2.2 · p. 6
• ConceptNet supplies possible attribute values by using relations such as “is-a” to suggest related concepts.Source: §3.2.2 · p. 6

3.2.3 Open Conversation with a Neural Language Model.

• The open mode uses a persona-based neural language model that generates text from chat history and personality statements.Source: §3.2.3 · p. 6
• Conditioning generation on both conversation and character attributes technically realises manifesting the fictional character as interaction progresses.Source: §3.2.3 · p. 6

• The study gathered writers’ first feedback on CharacterChat’s design choices and usefulness for character creation.Source: §4 · p. 6

• Remote sessions combined observation, think-aloud, interviews, chat-log analysis, and questionnaires to prioritize qualitative detail over sample size.Participants used the web app while sharing their screens with a researcher in a video call.Source: §4.1 · p. 6

• Seven amateur writers were recruited via university networks; all stated that they wrote as a hobby, and all but one prepared a writing project context for developing a character.Their mean age was 25 years, with a range of 24–27 years.Source: §4.2 · p. 6
• Each participant spent 20 minutes creating a character, followed by questions about inspiration, usability, chatbot conversation, and future use.Sessions lasted about 40 minutes and included researcher observation, think-aloud prompts, and the CSI questionnaire.Source: §4.2 · p. 6

4.3.1 Interviews & Feedback During Use.

• Everyone reported getting character ideas from the tool, including new events, possessions, and personality-revealing statements.Guided questions also helped writers consider overlooked aspects such as religion or fear.Source: §4.3.1 · p. 7
• Writers valued both modes because guided prompts structured attribute definition while open conversation enabled more communication and exploration.The combination was described as difficult to replace with either mode alone, although suggested attribute values were less consistently useful.Source: §4.3.1 · p. 7
• Five participants could imagine longer chatbot conversations, but responses often lacked context and did not sufficiently reflect the character’s developed personality.These limitations could become counterproductive as character development progressed.Source: §4.3.1 · p. 7
• Opinions about deeper character understanding were divided: some writers felt unchanged, while others gained a stronger or more nuanced character image.Participants reported additions such as a love interest or reconsideration of a stereotypical heroine after chatbot interactions.Source: §4.3.1 · p. 7
• Writers viewed the tool mainly as an early-stage aid for creating or elaborating characters, including secondary characters, inexperienced creators, writer’s block, and low motivation.Most participants could imagine incorporating it into their creative process, while several wanted more interaction controls and character visualizations.Source: §4.3.1 · p. 7

4.3.2 Creativity Support Index.

• The prototype achieved a CSI score of 67.78 out of 100, with strong Enjoyment, good Results worth effort and Exploration, and high importance assigned to Exploration and Expressiveness.Collaboration was rated very low, although the prototype did not include collaboration with other people.Source: §4.3.2 · p. 8

4.3.3 Chat Logs.

• Chat logs revealed robustness issues, repeated-attribute bugs, unsuitable suggestions, unclear starts, and confusing switches between guided and open modes.Despite these problems, the logs indicated that the chat system generally worked as intended.Source: §4.3.3 · p. 8

4.4 Prototype Improvements Based on the Study

• Descriptive ratings improved for all CSI aspects in the comparison with the first study, although the small sample prevented statistical tests.Table 2 reports the CSI questionnaire results for both studies.Source: Table 2 · p. 9

4.4.1 Giving Users More Choice.

• The revised chatbot displayed three suggested open-mode replies, allowing users to ignore unsuitable responses while exposing more options for inspiration and exploration.Only the user-selected reply was added to the chat history.Source: §4.4.1 · p. 9

4.4.2 Giving Users More Control.

• A delete function for collected attributes let users remove entries and explore changes when they reconsidered an attribute.Source: §4.4.2 · p. 9

4.4.3 Making Mode Switches More Explicit.

• Mode switching became explicit rather than relying on on-the-fly intent recognition, improving intent recognition and reducing confusion.Suggested inputs guided transitions after three defined attributes or when returning from open conversation.Source: §4.4.3 · p. 9

4.4.4 Further Improvements.

• The team also fixed bugs and made smaller user-interface, backend, logic, and input-handling adjustments based on study insights.Source: §4.4.4 · p. 9

5 USER STUDY II: IMPROVED PROTOTYPE

• Eight amateur writers evaluated the improved prototype after receiving 10 additional minutes of interaction time.The study used the same design and procedure as Study I, with one participant having taken part in both studies.Source: §5 · p. 9

5.1.1 Inspiration.

• All interviewees reported receiving ideas from the bot, while four explicitly described its inspiration and three described new approaches to their characters.One returning participant found the system less helpful when entering the study with a more specific character image.Source: §5.1.1 · pp. 9–10

5.1.2 Feedback on Specific Features.

• Three writers found guided conversation useful for discovering overlooked aspects, while open conversation and suggested values supported exploration and reduced author-typical characters.Participants described guided prompts as a “red thread,” valued open conversation’s expansion of supplied material, and said suggested values helped avoid unrealistically perfect characters.Source: §5.1.2 · p. 10

5.1.3 Conversation with the Chatbot.

• Three writers felt they were chatting with the character, and four said the interaction changed their character image or prepared-character aspects.Five could imagine incorporating the character into their stories, but some noted missing typical utterances or that characters develop during writing.Source: §5.1.3 · p. 10, §5.1.4 · p. 10
• Conversational quality improved but remained inconsistent: two participants found interactions unsmooth, three encountered answers contrary to defined attributes, and one viewed the system as an interactive notepad.Participants also criticized dialogue that did not reflect a character’s age or personality, despite being willing to have longer chats.Source: §5.1.3 · p. 10

5.1.5 Further Ideas and Use of the Tool.

• All writers could imagine using the tool again, especially for exploratory initial character development or supporting characters, while later writing stages required improvements.Six participants gave positive feedback on deeper understanding, and four requested notification if the tool became available online.Source: §5.1.5 · p. 10, §5.1.3 · p. 10, §5.1.4 · p. 10

5.2 Creativity Support Index & Chat Logs

• The prototype scored 76.96 CSI points, improving over the first prototype’s 67.78; no statistical tests were conducted because of the small sample.Descriptively, ratings improved for all aspects.Source: §5.2 · p. 10

6.1 CharacterChat Supports Early Stages of Character Creation

• CharacterChat was particularly useful in early character creation, helping writers capture fundamental aspects, discover further attributes, enrich secondary characters, and tolerate unexpected replies.The authors report that early-stage users were less constrained by occasional odd chatbot responses, which could remain inspirational rather than contradictory.Source: §6.1 · p. 11

6.2 Mode Switches in Creativity Support Should be Explicit for Users

• Different conversation flows for question-answer prompts and open dialogue can confuse users unless the interface provides explicit pathways for switching modes.The revised prototype made mode-switching phrases visible through interface suggestions.Source: §6.2 · p. 11

6.3 Guided Prompts Support Round Characters yet Limit Impressions of a Conversation

• Guided prompts supported discovering character attributes and escaping author-typical tendencies, although their fixed structure limited the impression of conversing with a character.Writers still valued the mode for attribute discovery, and simple, bounded interaction can be inspiring when aimed at one-word discoveries.Source: §6.3 · p. 11

6.4 Open Conversations are Inspiring yet Limited by Technical Capabilities

• Open conversation was inspiring, but generated text did not consistently sustain realistic dialogue or meet expectations and could contradict specified character attributes.Despite this limitation, the improved prototype produced an enjoyable conversational impression for some writers, and every writer pinned potentially interesting bot statements.Source: §6.4 · pp. 11–12
• Presenting multiple generated utterances as suggestions gave writers greater control and enabled exploration of model output as a pragmatic response to limited text-generation capabilities.The improved interface presented three suggestions, which users received positively.Source: §6.4 · pp. 11–12

6.5 Reflections on our Interaction Concept: Conversational AI as both Co-Creator and Creative Artifact

• CharacterChat realises a co-creative, mixed-initiative concept in which users and AI take turns shaping the character, while the chatbot is both collaborator and creative artifact.User actions define attributes and pin statements; bot suggestions hand initiative back to the user, whereas dismissed statements do not alter the character.Source: §6.5 · p. 12

6.6 Reflections on Methodology & Limitations

• An iterative user-centred process combining a survey and two user studies adapted the prototype to varied writing practices and revealed benefits, challenges, and opportunities.Study 1 informed interface and system changes that improved the prototype based on feedback and CSI scores.Source: §6.6 · p. 12
• The evaluation was limited by task, sample size, and study duration, motivating larger samples, professional writers, and longer use to assess writing processes beyond novelty.Writers generally worked on personally relevant characters, and CSI quantified experiences; writers valued exploration and expression over immersion.Source: §6.6 · p. 12

7 CONCLUSION

• CharacterChat supports character creation through Progressive Manifestation, in which writers progressively shape the chatbot into a fictional character through conversation.The concept combines guided prompts for defining attributes with open conversation for exploring the character.Source: §7 · p. 13
• CharacterChat was particularly useful for ideation when writers were creating new characters.This finding comes from two user studies evaluating the tool as support for writers.Source: §7 · p. 13
• Improved conditional dialogue language models could provide a better conversational impression in future versions.The conclusion identifies limited conversational capabilities as a current challenge.Source: §7 · p. 13
• The system realises co-creative AI by making the chatbot both a co-creative actor that suggests character elements and the creative artifact representing the character.This combines AI suggestions with the chatbot’s role as the character being shaped.Source: §7 · p. 13
• CharacterChat combines a creative tool for writers with an emulation of the resulting character.The work positions this combination as a contribution to discussions of AI tools and emulation.Source: §7 · p. 13
• Its two linked modes span tool-like guided prompts and partner-like open conversation, offering a reference point for interactive AI between these interaction concepts.The authors released the prototype and related material to encourage further exploration.Source: §7 · p. 13