QuillBot's
AI Detector
World's first free sentence level AI-detector which can distinguish between AI-generated & AI-paraphrased.
Company
QuillBot
Project Category
UX Design (P0)
Year
Jan-Apr, 2024
Team size
1 Prod. designer (me, IN)
1 Prod. manager (NL)
2 Front & back-end (IN)
1 Copy (CA)
OKR: Grow to 1.5M MAU coming from organic search & rank top 3 in Tier-1 countries
Status Quo: ~1.3M MAU and counting | 2nd rank globally | ~6% conversion rate
Responsibilities
-
Understand data, product need, objective, strategy, and use cases with respect to user's writing journey.
-
Brainstorm, Ideate, User journey mapping, Wire-framing, Prototyping & Visual design with stakeholders.
-
Craft Hi-fidelity UI screens, user flows, and blueprint to seek leadership alignment.
-
Work with business intelligence & strong user advocacy in a cross-functional team.
-
Support development efforts by pro-active reviews and quality assessment of specs.
-
Flexible yet resilient with rationale & clarity with solid presentation & communication skills.
-
Advocate best design practices, and facilitate team activities with a sense of inclusion.
-
Monitor data analytics to build hypotheses with an objective to incrementally improve product metrics.
Scenario
With growing usage of ChatGPT in the writing journey, AI detector tools have become widely popular among users.
Based on Ahrefs, search volume for AI detectors is increasing rapidly as per SEO experts. This became an opportunity to build the QuillBot's AI detector MVP for free users to drive traffic.
The solution
Based on all product considerations, the new QuillBot's sentence level AI Detector is launched on 1st April, 2024, with an objective to rank Top 3 in Tier 1 countries and reach a MAU of 1.5M. This launch is aimed at providing users a seamless and intuitive tool that detects 4 different types of text from the input within seconds, as shown below.
Dark modes
It is observed from UXR reports that 30-40% users prefer to use dark modes after using the tool for long to avoid strain in eyes. Hence, this tool is made to accommodate the user need of switching to dark mode at a click of a button.
Responsive design
30% users prefer to switch to mobile as it provides liberty to roam around socialising while also working on some assignments in parallel and discuss them in the social setting.
It all started with a sketch
Began with a few rough scribbles to visualise how the product landing page would look like, which further evolved incrementally with more sophistication.
Seeking clarity through chaos
Structuring the thoughts into robust frameworks to approach the problem with exhaustive manner and address all the cases.
Slept on "How might we's?"
Ideating with teams to address multilingual capacity of AI Detector for German, Spanish, French and English languages by designing the usage experience and Top of funnel awareness drive.
Conducted workshops to design the AI Detector icon
Option-A chosen as per the poll conducted (Sample size 30)
Based on user behaviour of toggling between AI Detector and other QuillBot portfolio products, the touchpoint is added to the side navigation bar.
New design of side navigation bar
Old design of side navigation bar
Strategised SEO to improve organic search
Below fold landing page
Based on usability findings, it is evident that Samsung account set-up wizard is placed towards the end of the OOBE (Out of box experience) when the user is exhausted. As the OOBE is Android centric, Samsung OneUI do not have legal rights to alter the positions.
Full page desktop view
Full page mobile view
Conducted a smooth dev handoff followed by quality assessment
Promoting async collaboration with development team
Basic interaction guide is introduced to use its annotations to make a blueprint of the tool inclusive of all conditions, user choices and error states.
The blueprint
The tool blueprint helps developers to have a overview yet detailed picture of all scenarios, error states and user decisions. This blueprint not only reduces the transactional conversations, but also improves async collaborations for a remote team working in different time zones globally.
.png)
Recovered from false positives right after launch
We managed to recover from false positives in 5 days by addressing model and setting right expectations at UI level.
Right after the launch user feedback came pouring in at a rate of 1:98, among which majority were around false positives. As the model need some time to build real time accuracy, the product POD decided to manage expectations at UI level followed by rendering the engine with a more conservative algorithm. This 5 day sprint was executed on an urgent basis, reducing the false positives significantly, which further reduced the number of feedbacks.
Day 0
Feedback on false positives due to usage of liberal AI model in the release
Day 1
Adding ‘likely’ to manage expectations & re-categorising using MECE framework
Day 3
Using a conservative model to reduce false positives. Added ‘Like’ & ‘Dislike’ for feedback
Day 5
Cross selling of QuillBot’s relevant portfolio products to improve visibility
We witnessed a significant traction in analytics dashboard
500K+ weekly active users in the first week of launch
Not only traffic has significantly increased but also the tool made its ranking to 3rd from 5th as per the Ahrefs.
Significant decline in the ratio of feedback provided
Due to the quick response of the team to make amendments at both UI level and model level, feedbacks reduced significantly.
Positive conversion from a Non-Premium tool
Conversion from AI Detector raised from 139 users to 201 users in the first week of launch
Paraphraser is highest used by AI detector users
Based on analysis, Paraphraser has been the winning candidate among Grammar Checker and Plagiarism Checker.
Next steps:
We plan to go big, but incrementally
QuillBot's AI detector will evolve in multi-lingual geographies, accuracy, monetisation strategies, paraphrasing capabilities, and effective cross pollination with other products.
Optimise for Non-English text at model level
Close AB tests & start cross selling contextually
Make the chunks interactive with more specificity
As QuillBot has began its journey in the European market, optimising the tools for Non-English is imperative considering its high traffic.
AB test for cross selling 'Paraphraser', 'Grammar checker' & 'Plagiarism checker' needs to be closed with more data, and introduce backend engine to detect issues and nudge the user to use the right product.
Add hover effects on highlighted chunks for user to take specific actions like 'Rewrite', 'Like' and 'Dislike' touchpoints.
Demo videos to understand the Usage experience of AI Detector
Click here to try and test the AI detector and provide your valuable feedback.
Elements of landing page
The video explains the elements of the landing page of Free AI detector tool and how this page acts as both as a web application and a marketing page for SEO generation.
Text input user flow
Due to the model's best performance, text input should be at least 80 words and not more than 1,200 words. This video shows how error messages are being shown to user subtly while pasting less than 0 or 80 words
Consuming results
The legend on the right clearly depicts the colour coding of the 4 different types of content pasted by the user, with tool tips explaining through examples for better clarity.
Making edits & re-analyze user flow
Due to its great speed of analysing, users can make quick edits in their detected content on the left hand side and re-analyse again and again. As there are legal constraints to have 'Paraphrase' CTA in the tool, the team decided to cross sell the QuillBot's portfolio products opening in a new tab.
Appendix
Pre-emptive launch of first MVP on Jan 4, 2024
QuillBot witnessed a significant rise in the traffic of AI Detector MVP in a short span of time.
MVP analytics
QuillBot witnessed a significant rise in the traffic of AI Detector MVP in a short span of time.
Product & design debts
MVP was launched with velocity postponing certain product & design aspects to release in the subsequent phases.
Competition analysis
Competitors are analysed for their overall experience for first time as well as regular usage in every step of the UX flow.
Outdated & uninviting interface
Plethora of text OR advertisements
Specificity & better expectation management
Premium pricing
Harshly opinionated that AI is bad
Clumsy usage experience
Chunk level detection with emphasis on Humanize
Login to use more
Exhaustive product brief
Design an AI detector tool for QuillBot free & un-logged users, which can provide a likely score for overall AI-generated content and able to surface AI-generated & paraphrased chunks at sentence level specifically. The tool should prominently cross-sell QuillBot’s portfolio products like Paraphraser, Plagiarism Checker & Grammar Checker for its relevance after running AI detection.
With a trendy look & seamless UX, the tool should have a neutral stance while highlighting AI-generated content. Feedback touchpoints should be easily accessible so that the AI model can refine itself for false positives.
Key product considerations
Based on data, heuristics analysis and competition study, following product considerations are imperative to execute, in order to achieve OKRs.
Thank you