Table 5.

Guidelines for developing engaging chatbots.

Tip	Definition	Rationale
Focus on the first module	Emphasis should be placed on the development of the first module to increase the chance of participants continuing to engage with the chatbot.	Most participants use only 1 module (or 2). It is important to provide the best intervention possible from the beginning.
Start small	Developers should focus on creating a few engaging and effective modules at the beginning rather than developing a large variety of untested modules.	On the basis of participant’s flow through Tess, most individuals tend to discontinue after the second module.
Develop comparable modules	To compare module utilization, modules should be built with similar characteristics (eg, length, number and type of questions, and the inclusion of links).	Utilization of modules that differ in characteristics may not allow for meaningful comparisons.
Balance length and complexity	Modules with fewer interactions may have better completion, but more complex topics may need longer modules. Developers should strive to find a module length that enhances intervention fidelity without compromising engagement.	Modules that had more interactions from the artificial intelligence had lower completion rates.
Be aware of personalization and standardization	Standardization provides streamlined data and requires less work, whereas personalization has more complexity and consequently requires more effort both in development and analysis. However, personalization may provide richer interactions.	Without understanding the efficacy of the overall modules, it is difficult to assess whether branching promoted engagement.
Holistic assessment	There is no single variable that can provide an accurate measure of utilization, rather a combination of such variables can provide a broad idea of general utilization and specific information regarding specific aspects of the intervention.	Number of messages, characters used, time spent, and completion should be considered alongside helpfulness and satisfaction in the assessment of chatbots.