Bottom line
Pick ChatGPT if first-message habit matters most. Pick Claude if spend and latency need to be visible before send. Pick Perplexity if research depth is the default decision. Pick Gemini if Google files and Workspace depth are the hook.
At a glance
| Dimension |  ChatGPT |  Claude |  Perplexity |  Gemini |
|---|---|---|---|---|
| Default bar | Empty messaging bar; starter pills hint at jobs without mode chips. | Empty card with starter pills; model and effort already visible on the right. | Ask bar with Search and Computer chips inside; research depth is the default decision. | Empty pill bar, no chips; + left, Flash nickname and mic on the right. |
| Tool & mode disclosure | One + menu for attach, tools, and modes; removable chips show scope before send. | Web search on by default; + menu for skills and attach; no separate search mode chip. | Mode chips (Search, Computer) are primary; + handles uploads and connectors. | + menu leads with Drive; Deep research and creation tools live in flyouts. |
| Model & cost visibility | Outcome labels in menus (“Create image,” “Deep research”) over raw model names. | Model and effort on the composer for spend control before the first send. | Model picker on the bar; free tiers often show locks; Computer labeled as orchestrator. | Flash on the bar; picker uses outcome labels and plain-language thinking copy. |
| Attachments & context | Removable chips for tools and files so scope is legible before send. | Context chips for files and sources; input enrichment over mode picking. | File preview with name and size in the ask bar; source toggles in connector flyout. | Add from Drive on the first + screen; image preview with filename in the pill. |
| Voice | Two voice entry points; dictation vs conversation is easy to confuse. | Voice transcription with confirm before send; icons can look alike until tapped. | Mic and voice on the right of the bar beside the model control. | Mic on the compose row as a peer to typing. |
| Product bet | Mass-market habit: look like chat, reveal power when intent is clear. | Research-ready default with explicit cost/quality before send. | Research product: pick depth before the query, not after. | One front door; ecosystem retention via Drive and Google workflows. |
Default bar
Tool & mode disclosure
Model & cost visibility
Attachments & context
Voice
Product bet
ChatGPTMass-market habit: look like chat, reveal power when intent is clear.
ClaudeResearch-ready default with explicit cost/quality before send.
PerplexityResearch product: pick depth before the query, not after.
GeminiOne front door; ecosystem retention via Drive and Google workflows.
What to copy
- Copy ChatGPTYou need maximum first-message conversion on a consumer chat productCalm empty bar, one + menu, removable chips, and outcome labels instead of model jargon.Read the ChatGPT teardown
- Copy ClaudeUsers must understand cost and quality before a heavy runModel and effort visible on the composer; web search default on without a mode chip.Read the Claude teardown
- Copy PerplexityYour product is research or agentic depth, not open-ended chatPrimary mode chips for depth; split attach/connectors from mode selection.Read the Perplexity teardown
- Copy GeminiYou win by pulling users into an existing workspace or file ecosystemEmpty default bar plus Drive on the first + screen and plain-language thinking copy.Read the Gemini teardown
Copy GLM (Z.ai)The product ships sites, decks, or code from a chat homeBuilder headline plus job pills that pull into the bar and swap the gallery. Keep Deep Think on the flagship, not the casual default.Read the GLM (Z.ai) teardown
More design the composer teardowns
Same job, not covered in the sections above.
GrokNearly empty canvas for time-to-first-query; depth and upsells share the same band.Read teardown
Google AI ModeConversational search on google.com with an explicit AI Mode toggle, not a second app.Read teardown
LovableBuild vs Plan before the agent runs; project starter, not open chat.Read teardown
DeepSeekInstant/Expert tiering with Search on by default and DeepThink as an opt-in.Read teardown
KimiAgent pills rewrite the bar into job workspaces; Instant/K3/Swarm sets depth.Read teardown
ManusAssign-a-task framing with mode chips and Lite/Pro/Max agent tiers.Read teardown- GLM (Z.ai)Builder pills rewrite the gallery; GLM-5.2 swaps in Deep Think and artifact categories; Agent is a second home.Read teardown
Frequently asked questions
What is an AI chat composer?
The composer is where users type prompts and choose tools, modes, attachments, models, and voice before the model responds. It sets expectation, cost, and capability before the first token streams.
ChatGPT vs Claude vs Perplexity vs Gemini: which composer UX is best?
None is universally best. ChatGPT favors habit and first-message conversion. Claude favors cost and quality visibility with web search on by default. Perplexity favors choosing research depth before the query. Gemini favors a calm bar plus Google ecosystem depth. Match the pattern to your users.
Should tools live in the bar or behind a + menu?
Use bar chips when mode is the product (Perplexity Search vs Computer). Use a + menu when most users should just type (ChatGPT, Gemini). Claude’s middle path defaults web search on and keeps skills behind +. Parallel discovery paths without a bridge confuse users.
Where should model and cost controls appear?
Put model and effort on the composer when spend and latency matter before send (Claude). Use outcome labels in menus when casual users should not pick model names (ChatGPT, Gemini). Lock or gate pickers honestly on free tiers (Perplexity) rather than hiding limits until after send.
How is this comparison different from the product teardowns?
Each product teardown is a screenshot-backed walkthrough of one composer. This page synthesizes the same job across products into a bottom line, comparison table, and copy rules. Use it to pick a pattern, then open the linked teardown for evidence and anti-patterns.