Bottom line
For chat products, copy ChatGPT progressive depth: chips on the claim, then popover, then sidebar. For research products, copy Perplexity audit path: Links tab, selection check, and wrong-source feedback.
At a glance
| Dimension |  ChatGPT |  Perplexity |
|---|---|---|
| Inline citation chips | Publisher favicon + name pills on claims; +N marks multi-source support. | Domain-first chips (northjersey +3) sit at the end of claims in the reading flow. |
| Claim-level preview | Chip opens a popover with publisher, headline, and snippet; 1/N arrows for each source. | Chip popover with 1/N navigation and snippet without opening the Links tab. |
| Full source audit | Sources row under the reply opens a sidebar of publisher cards beside the answer. | Links tab with full source cards; Sources sidebar from the answer bar for side-by-side audit. |
| Research transparency | No research-step UI; citations appear when the answer is web-grounded. | Collapsed “Completed N steps” expands to plain-language search and check steps. |
| Source quality feedback | Standard thumbs feedback; no first-class “wrong sources” failure mode in the citation surface. | Negative feedback includes Wrong sources; Check sources on text selection localizes verification. |
| Product bet | Chat-first: progressive verification depth when the web is involved, without making citations the brand. | Research-first: evidence in the reading flow is the product, with audit and source-failure taxonomy. |
Inline citation chips
Claim-level preview
Full source audit
Perplexity
Links tab with full source cards; Sources sidebar from the answer bar for side-by-side audit.
Research transparency
ChatGPTNo research-step UI; citations appear when the answer is web-grounded.
Source quality feedback
ChatGPTStandard thumbs feedback; no first-class “wrong sources” failure mode in the citation surface.
Perplexity
Negative feedback includes Wrong sources; Check sources on text selection localizes verification.
Product bet
ChatGPTChat-first: progressive verification depth when the web is involved, without making citations the brand.
PerplexityResearch-first: evidence in the reading flow is the product, with audit and source-failure taxonomy.
What to copy
- Copy ChatGPTYou ship a general chat product that sometimes grounds on the webPublisher chips on claims, claim-anchored popovers with 1/N paging, and a one-click Sources sidebar.Read the ChatGPT teardown
- Copy PerplexityYour product is research or answer-engine and trust is the brandDomain chips, research steps, Links/sidebar audit, Check sources on selection, and Wrong sources feedback.Read the Perplexity teardown
Frequently asked questions
What is AI citations UX?
Citations UX is how an AI product shows which sources support a claim: inline chips, hover previews, source lists, research steps, and feedback when sources are wrong. It is the primary trust surface for web-grounded and research answers.
ChatGPT vs Perplexity: which citations UX is better?
Neither is universally better. ChatGPT favors progressive depth inside chat (chips, popover, Sources sidebar). Perplexity favors citation-native research (chips, steps, Links audit, wrong-source feedback). Match the pattern to chat vs research.
Should citations be inline chips or footnotes?
Both ChatGPT and Perplexity put publisher- or domain-first chips on the claim in the reading flow. Footnote-only lists at the bottom force skeptical readers to leave the claim. Pair inline chips with a full audit surface for heavy verification.
When do you need Wrong sources feedback?
When sourcing quality is a first-class failure mode in research, news, finance, or health. Perplexity treats wrong sources as feedback taxonomy. ChatGPT focuses on inspection depth rather than source-specific failure labels.
How is this comparison different from the product teardowns?
Each citations teardown is a screenshot-backed walkthrough of one product. This page synthesizes the same trust job across ChatGPT and Perplexity into a bottom line and comparison table. Use it to pick a pattern, then open the linked teardown for evidence.