The major consumer AI discovery surfaces, and how their retrieval, citation and commercial mechanics differ — from validated evidence, with every gap left explicit rather than filled.
Registry facts (type, geography, discovery modes) are marked as registry metadata. Reach is a single evidenced figure where one exists; mechanics coverage and confidence come from validated claims only. Market share is context, not the product. Explore the full 35-surface registry →
OpenAI
Discovery modes (registry): web search · answers · shopping · deep research
Mechanics coverage: 7/13 evidenced
For ChatGPT, sources are cited, crawler/index controls are documented, its retrieval path is partly evidenced.
Open evidence & mechanics →Discovery modes (registry): web grounding · answers · deep research · shopping
Mechanics coverage: 0/13 evidenced · unknown is a valid answer
No evidenced discovery mechanic for Gemini yet; treat any marketing assumption as unverified.
Open evidence & mechanics →Discovery modes (registry): search · query fanout · shopping · local
Mechanics coverage: 4/13 evidenced
For Google Search AI Mode, sources are cited, its retrieval path is partly evidenced.
Open evidence & mechanics →Anthropic
Discovery modes (registry): web search · answers · research
Mechanics coverage: 0/13 evidenced · unknown is a valid answer
No evidenced discovery mechanic for Claude yet; treat any marketing assumption as unverified.
Open evidence & mechanics →Perplexity
Discovery modes (registry): search · citations · shopping · research
Mechanics coverage: 3/13 evidenced
For Perplexity, its retrieval path is partly evidenced.
Open evidence & mechanics →DeepSeek
Discovery modes (registry): web search · answers
Mechanics coverage: 0/13 evidenced · unknown is a valid answer
No evidenced discovery mechanic for DeepSeek Chat yet; treat any marketing assumption as unverified.
Open evidence & mechanics →ByteDance
Discovery modes (registry): web search · answers · recommendations
Mechanics coverage: 0/13 evidenced · unknown is a valid answer
No evidenced discovery mechanic for Doubao / 豆包 yet; treat any marketing assumption as unverified.
Open evidence & mechanics →Comparing 6 surfaces (maximum 6). The selection is stored in the URL, so the comparison is shareable.
| Dimension | ChatGPTOpenAI | GeminiGoogle | Google Search AI ModeGoogle | ClaudeAnthropic | PerplexityPerplexity | DeepSeek ChatDeepSeek |
|---|---|---|---|---|---|---|
| Search / retrieval triggerWhen and why the surface decides to retrieve from the live web rather than answer from model weights. | Known Mobile.de citations in German-language ChatGPT responses decreased by 66% after the GPT-5.5 rollout. evidence → | UnknownNo validated claim covers this dimension. | Known Google employs thousands of quality raters to test and improve search and AI results. evidence → | UnknownNo validated claim covers this dimension. | Known Perplexity Research publishes Q2D-Web, a first-stage retriever benchmark built on agent queries and web documents, indicating its search stack is evaluated at the retrieval stage. evidence → | UnknownNo validated claim covers this dimension. |
| Retrieval provider / indexWhich index or upstream search provider backs retrieval. | Known Mobile.de citations in German-language ChatGPT responses decreased by 66% after the GPT-5.5 rollout. evidence → | UnknownNo validated claim covers this dimension. | Known Google employs thousands of quality raters to test and improve search and AI results. evidence → | UnknownNo validated claim covers this dimension. | Known Perplexity Research publishes Q2D-Web, a first-stage retriever benchmark built on agent queries and web documents, indicating its search stack is evaluated at the retrieval stage. evidence → | UnknownNo validated claim covers this dimension. |
| Query rewrite / fan-outWhether and how the user query is rewritten, decomposed or fanned out into multiple sub-queries. | Known Mobile.de citations in German-language ChatGPT responses decreased by 66% after the GPT-5.5 rollout. evidence → | UnknownNo validated claim covers this dimension. | Known Google employs thousands of quality raters to test and improve search and AI results. evidence → | UnknownNo validated claim covers this dimension. | Known Perplexity Research publishes Q2D-Web, a first-stage retriever benchmark built on agent queries and web documents, indicating its search stack is evaluated at the retrieval stage. evidence → | UnknownNo validated claim covers this dimension. |
| Citation / source presentationWhether sources are shown, and in what form (inline links, cards, footnotes, absent). | Known The share of ChatGPT responses with at least one inline element (image embeddings) rose from approximately 15% to 35% between May 7-11, 2026. evidence → | UnknownNo validated claim covers this dimension. | Known While generating an AI response, Google's models identify additional supporting web pages and display a wider, more diverse set of links than classic web search. evidence → | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. |
| Crawling and indexing controlsCrawler user-agents, robots directives and site-side controls that govern what is fetched and indexed. | Known It can take approximately 24 hours for OpenAI's systems to adjust for search results after a site's robots.txt update. evidence → | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. |
| Freshness / recrawlHow quickly changed content is refetched and reflected. | Known It can take approximately 24 hours for OpenAI's systems to adjust for search results after a site's robots.txt update. evidence → | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. |
| Shopping / product feedProduct, catalogue or feed behaviour in commercial results. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. |
| Local retrievalUse of local/place retrieval or map data. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. |
| Social / community retrievalUse of social, forum or community sources (e.g. Reddit). | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. | UnknownNo validated claim covers this dimension. |