Case study — X / Twitter redesign

less chatbot, more context. AI in search

A redesign of how AI shows up on X. Instead of sending people off to a separate Grok chat, we put it inside the two places they already go to make sense of things: their topic tabs and search.

Team

Mike & Pooja

Type

Existing-product analysis + redesign

Focus

AI in topic tabs and search

Deliverables

Analysis, hi-fi prototype

X app screens showing Grok summaries inside topic tabs and search

Tools used

What I used, start to finish.

Figma

Wireframes & hi-fi prototype

Claude

Research & writing support

01 — The product

the internet’s
town square

X is the fastest place to find out what’s happening, and what people think about it.

Before designing anything, we mapped what X is for, who uses it and what the business needs from it.

Purpose 01

Know what’s happening

Breaking news, live events, sports and culture,
as they unfold.

Purpose 02

Hear every side

An open, public conversation where anyone can reply, quote or push back.

Purpose 03

Reach people directly

Creators, experts, brands and officials talk straight to their audience.

Who uses it

Readers want to stay informed. They mostly scroll and lurk: news, trends, sports, entertainment and the replies.

Creators and voices want to build an audience. They post takes, threads and video, and earn through revenue sharing and subscriptions.

What the business needs

Turn attention into revenue, while keeping X worth paying attention to.

Grow daily usage: more time on X means more ad inventory and data. Diversify revenue: grow Premium and data licensing so ads aren’t the only income.

02 — Where AI fits today

grading grok

Grok is capable, but it’s bolted on. X pushes it as a destination instead of using it to improve what people already do.

Pain point

What AI does today

Verdict

Information overload

Grok summarizes trending stories in Explore and explains any post from the Grok button.

Works

The most useful part of Grok today.

Hard to know what’s true

People tag @grok to fact-check. Community Notes uses an algorithm to surface notes people agree on.

Partly

Grok can be confidently wrong.

A noisy feed

The For You feed ranks posts by predicted engagement.

Partly

Optimizes for engagement, not relevance.

Search takes work

The Grok tab answers questions in a separate chat.

Misses

Lives outside search, so it’s a detour.

Good

Usefulness

Summaries of trending topics and posts genuinely save time.

Weak

Placement

A whole nav tab plus a logo on every post feels pushy.

Weak

Fit with why people use X

People come for news, people and conversation, not a separate LLM.

Mixed

Trust

Answers can be confidently wrong, and sources aren’t always easy to check.

03 — The takeaway

AI on X works best when it helps people make sense of what’s already in front of them, and least when it sends them somewhere else.

Summaries in context are useful. Grok as a separate destination doesn’t fix the core pain points: overload, trust and noise.

04 — The pivot

we changed the brief

A better chatbot wouldn’t fix the real issue: people don’t come to X for a chatbot.

So instead of redesigning Grok, we designed a new AI experience that lives where the need already is.

Considered

Make the Grok chatbot better

Better answers, but still a separate place people have to go. It doesn’t match why they open X.

→
Our choice

Introduce a new AI experience

AI shows up inside topic tabs and search, right where people are already trying to make sense of something.

Why 01

Highest intent

People search when they want to make sense of a topic, not just pass time.

Why 02

Biggest pain

Search results are a wall of posts with many viewpoints and no overview.

Why 03

Proven elsewhere

Threads and Bluesky show that subtle, in-context AI lands better than a chat tab.

Two entry points: Home topic tabs and Explore search
The two moments we designed for: catching up on the topics you follow (Home → Sports & Tech tabs), and searching for something specific (Explore → “nyc news”).

05 — Scenario 1

catch up on my topics

“What did I miss?” Grok turns the Sports and Tech tabs on Home into a quick catch-up that fits the topic.

We followed one person through two everyday moments. This is the first: opening X to catch up on the things they follow. Path: Home → Sports or Tech tab.

Generative UI summaries for NBA, NFL and tennis

Generative UI

One summary format can’t fit every topic, so Grok picks the layout that fits the content.

  1. 1

    The content decides the layout

    A live NBA game gets a scoreboard and win probability, an NFL game gets an injury report, tennis gets a set-by-set score.

  2. 2

    Visual first, text second

    Numbers, status and trends come before sentences, so the story reads at a glance.

  3. 3

    Personalized, with controls

    Based on accounts you follow and posts you engage with, and it says so. Show less or turn it off per topic.

Tech tab showing viewpoints and follow-up answers

Further interactions

Follow-ups turn the catch-up into answers about you, and you can always shrink it back to posts.

  1. 1

    Summarizes the viewpoints

    A sentiment bar shows the split: excited, mixed, skeptical, and where people disagree.

  2. 2

    Answers follow-up questions

    “Will my iPhone get it?” becomes a checklist for the devices you own.

  3. 3

    Leads back into content

    Suggests creators’ videos and posts to keep exploring, rather than ending the conversation in a chat.

06 — Scenario 2

search for something specific

“What actually happened?” Grok sums up the search results and shows where each claim comes from.

Path: Explore → search “nyc news”. The summary sits above the results, with numbered citations linking to the posts it used. And it changes with each results tab.

Top

What’s the story?

An overview of the top posts, with sources.

Latest

What’s new?

A live timeline that updates as posts come in.

People

Who’s worth following?

Says why each account is relevant.

Videos

What’s in these clips?

Describes the footage and flags re-uploads.

Photos

Are these real?

Describes the images and flags old photos.

Search summary across all five result tabs

Designing for trust

Our own analysis said Grok can be confidently wrong, so every claim in a search summary carries a numbered citation, and the Photos and Videos tabs actively flag recycled media.

Hi-fi prototype showing both entry points

07 — Prototype

A clickable hi-fi prototype of both entry points.

  1. 1

    Home → tap Sports or Tech

    Open the catch-up, expand it, hide it or turn it off.

  2. 2

    Explore → search “nyc news”

    See the summary change across Top, Latest, People, Videos and Photos.

  3. 3

    Ask Grok a follow-up

    “Which ones were on the subway?” narrows the results to the posts that match.

08 — Process

sketches, flows and ideas we set aside

We sketched both directions on paper, mapped the user flows,
then pushed early dark-mode search screens before the hi-fi pass.

Early paper sketches and dark-mode screens
AI in search and personal topic summaries, sketched before any screens.

Ideas we explored

Sketched + flowed

Summaries on high-traction posts

Open a popular post and Grok summarizes it and the replies.

Flowed

A “More like this” button

From a post, Grok finds similar posts, or you tell it which topics you want more of.

Rejected

Redesign the existing Grok chatbot

A better chatbot doesn’t change why people open X, and Grok already has its own app.

User flows for AI in search and personal summaries

09 — What’s next

taking it to the final

This was our midterm. These are the questions we want the final version to answer.

Tie it to the business

A good summary could mean people leave sooner. We want to show how follow-ups and “shrink back to posts” lead people deeper into X, and whether this belongs in Premium.

Design for uncertainty

What happens when sources disagree, Grok isn’t sure, or a Community Note contradicts the summary? The Tech tab’s viewpoint bar is a start.

Bring creators in

If Grok summarizes six creators’ videos, those creators should still get the credit and the views.

Test with real people

Short interviews to confirm the pain points, then usability tests of both flows in the prototype.