Ask Anna
Making an AI experience easier to understand, use, and trust
Overview
Seerist helps analysts, security teams, and government organizations make sense of global risk across a large amount of intelligence.
Ask Anna gave them a faster way into that information. Analysts could ask a natural language question and receive a synthesized answer grounded in Seerist’s intelligence, with sources they could trace back and verify.
I led the UX redesign of the experience, working closely with Product, Data Science, engineering, intelligence experts, and Customer Success.
The challenge was to make Ask Anna easier to understand and trust while working within technical constraints and a tight delivery timeline.
- Lead Product Designer · sole designer
- Nov 2025 – Jul 2026
- 1 PM · 5 engineers · 3 data scientists · 7 customer success managers
Outcome
3h → 20m
Time to produce an intelligence analysis
Self-reported by analysts
25%↑
Analysts reused AI generated answers more often
Measured through copy interactions
3×
Increase in citation clicks, indicating deeper source investigation
Measured from users clicking the citations
The signal that made me look closer
As part of my regular product tracking, I noticed Ask Anna’s usage beginning to decline. I had not been deeply involved in the earlier design, so I took a closer look at the live experience and started documenting the usability issues I was seeing.
What I found pointed to a bigger problem. Ask Anna looked like chat, so users expected a flexible, conversational experience like ChatGPT. In reality, it worked best with one focused question, clear context, and a single synthesized answer.
The interface was setting expectations the technology did not support.
- Ask anything, in any way
- The system fills in missing context
- Follow-up questions build on the conversation
- Ask one question at a time
- Provide enough context in the question
- Each question stands on its own
“I assumed it worked like ChatGPT. When it didn’t, I thought the AI was broken.”
That mattered for intelligence analysts. They needed to understand how the system worked and trust its output before using it in their work.
The problem
At first, the obvious solution seemed to be helping users ask better questions.
But as I dug deeper, I realized users were trying to understand the system itself:
- What can Ask Anna actually do?
- What information does it need from me?
- What does it remember?
- Where does the answer come from?
- How much should I trust it?
Trust came up repeatedly in customer feedback. Users wanted more visibility into where answers came from and how reliable they were.
Research
I combined analyst interviews, user feedback, usability testing, and synthesis of user conversations to understand where trust was breaking down.
That shaped the redesign around three goals: explain how Ask Anna works, guide users to give it the context it needs, and make responses easier to evaluate.
Three goals guided the design
- 01
Design for how it works
Shape the interaction around what Ask Anna actually does well.
- 02
Guide in context
Help users give Ask Anna what it needs, right when they need to.
- 03
Help users verify
Organize answers so sources, context, and limitations are easy to find and assess.
Each goal maps directly to a design decision.
Design
- 01
From chat to focused search
The biggest shift was the interaction model itself.
A chat interface suggested an ongoing conversation, where each question built on the last. Ask Anna could not reliably support that behavior.
How it wasChatWhat’s happening in Eastern Europe?Here’s what I found. A long, unstructured response with no clear source.Ask another questionWhat about the economy?An unexpected answer. It did not carry over what was just asked, and nothing on screen says so. Is the AI wrong, or broken?How it becameSearchDrone incursions over the last 6 months in EuropeAnswer
Structured into sections, with every claim traced to a Control Risks report.
→ 3 sources, with datesThe question stays in the field. Asking something new means editing it, one question at a time, and the interface says so.Search became a stronger mental model: users asked a focused question, received a result, and refined their query when needed.
That better matched how the system worked and set clearer expectations from the start.
- 02
Guidance became part of the product
Ask Anna worked differently from more familiar AI tools, so users did not always remember how to work with it when they returned.
I explored lighter onboarding approaches, including a minimal start screen and first visit guidance. Both looked cleaner, but still relied on users remembering what they had seen.
I pushed to keep the most useful guidance in the experience itself, so it was there when users actually needed it.
One dense About paragraph that disappeared once users started typing became four clear steps they could scan and return to. The landing experience kept the essentials visible: ask one focused question, include the context that matters, treat each question independently, and check the sources behind the answer.
The goal was to make Ask Anna easier to work with without turning the experience into a set of instructions.
- 03
Make answers easier to judge
Getting an answer was only part of the job. Analysts still had to decide: can I use this?
Customer feedback raised concerns around hallucinations, generic responses, and inconsistent citations. That made the context around an answer just as important as the answer itself.
I redesigned the response around what analysts needed to judge an answer: clearer hierarchy, traceable sources, source dates, and clearer limitations.
- 01Limitations shown upfront
- 02Answers organized for fast scanning
- 03Citations tied to specific claims
- 04Older sources clearly flagged
The goal was to make Ask Anna easier to read, verify, and evaluate.
Constraints & tradeoffs
Designing for what the technology could support.
The product vision extended beyond what Ask Anna could reliably support at launch.
Customers wanted report generation, graphs, scheduled analysis, reusable prompts, asset aware responses, and more personalized outputs. The strongest experience we could deliver at the time was focused search and question answering.
- Flexible output formats (timeline, report)
- Send answers to Report Builder and save to Content Folders
- Preview citations within answers
- Multi-turn contextual conversations
- Clear question-and-answer model
- Readable, structured responses
- Visible evidence and citations
- Persistent in-workflow guidance
- Recoverable, explicit error states
Validation changed how I thought about guidance
Testing with analysts confirmed that people skimmed instructions and rarely remembered guidance they had seen earlier.
Rather than adding more onboarding, I focused guidance where decisions actually happened: when users were forming a question, evaluating an answer, or trying to understand what went wrong.
Copy set expectations, guided more focused questions, and clarified limitations.
| Before | After | |
|---|---|---|
| Value proposition | Access AI powered insights from 12 months of expert analysis on global risks, emerging threats, and geopolitical trends. | Get answers based on 12 months of analyst reports, with sources linked |
| First step | Ask a question. | Ask a focused question. Start with a specific question. Add a timeframe or location when relevant. |
| Input field | Enter your question here. | Ask in plain language |
| Limitation | Not stated. | One question at a time. Anna does not remember the last one. |
The final experience
The redesign brought those ideas together into a clearer, more focused experience.
Focused questions, relevant context, scannable answers, visible sources, and clear limitations gave analysts more information to understand and evaluate Ask Anna’s responses.
Overall, the interface did more to explain how the AI worked, reducing the need for users to learn through trial and error.
Try it
Reflection
What I shaped
This project pushed me beyond designing an AI interface.
I helped define how AI should behave inside Seerist, balancing user needs, technical constraints, business goals, and the transparency required for intelligence work.
What I would push further
The experience still required users to remember what made a strong question.
If I continued the work, I would help them build questions directly in the interface. Instead of asking users to remember to include a topic, timeframe, and location, the product could guide them through each:
What I learned
As AI gets more capable, the design challenge becomes helping people work with it well.
For me, strong AI UX creates a clear relationship between the user and the system, one where the AI adds speed and intelligence while the user retains context, judgment, and control.
