ANSWER ENGINES
7 answer engines —
what actually
differs
These are the 7 answer engines Navirang uses as the denominator for citation rate. They differ from crawler names to index sources to how citations are displayed, so results from one engine cannot be used to infer another.
SHORT ANSWER
Answer engines gather documents with different crawlers, draw on different indexes, and display sources in different ways. So “we show up in AI” or “we don’t” cannot be settled in one sentence — it has to be split by engine. This page collects the specifications of the 7 engines Navirang measures against, in one place.
Navirang fixes these 7 as the denominator for citation rate. 40 questions × 7 engines = 280 cells is the unit of our own baseline. Adding or removing an engine turns the same citation rate into a different number, so once the set is chosen it does not change.
COMPARISON
All 7 engines side by side
Only the axes where engines actually differ. Axes that return the same value everywhere are left out.
| Engine | Operator | Crawler / control | Index source | Citation format |
|---|---|---|---|---|
| ChatGPT | OpenAI | GPTBot · OAI-SearchBot · ChatGPT-User | Own index + third-party search | Inline links |
| Google AI Overviews | Googlebot (no dedicated AI crawler) | Google search index | Source cards | |
| Perplexity | Perplexity AI | PerplexityBot · Perplexity-User | Own index + live search | Numbered footnotes |
| Claude | Anthropic | ClaudeBot · Claude-SearchBot · Claude-User | Own index | Source list |
| Naver AI Search | Naver | Yeti | Naver search index (high domestic share) | Source list |
| Copilot | Microsoft | bingbot (no dedicated AI crawler) | Bing search index | Source list |
| Gemini | Google-Extended (control token, no UA) | Google search index | Source list |
ENGINES
Reference per engine
Each page carries a specification table, citation conditions, common misconceptions, and how to verify it yourself.
OpenAI
ChatGPT
Three separate crawlers — training, search, and user-requested
Google AI Overviews
No dedicated AI crawler — Googlebot is the whole story
Perplexity AI
Perplexity
The most visible sourcing and the fastest to reflect change
Anthropic
Claude
Separate crawlers for training, search and user requests
Naver
Naver AI Search
Heavy on domestic documents — the channel you publish on decides the outcome
Microsoft
Copilot
Runs on the Bing index — the entry point is Webmaster Tools
Gemini
Grounding is controlled by the Google-Extended token
PRINCIPLES
What holds across every engine
Specifications differ per engine; the measurement protocol does not.
- 01
We never infer one engine from another
Indexes and selection criteria differ, so appearing in ChatGPT does not mean appearing in Perplexity. Recording per engine is what tells you where to fix things.
- 02
We count mentions and citations separately
A mention is your name in the answer text; a citation is your domain in the source list. The two move independently and call for different responses.
- 03
A citation rate without a stated denominator cannot be verified
Different question counts and engine counts turn the same percentage into a different number. When Navirang publishes its own figures, the denominator is always stated with them.
- 04
We never judge from a single response
The same question yields different answers depending on the moment and the session. Fixing the wording and repeating in logged-out sessions is what makes a record reproducible.
- 05
No engine guarantees visibility
Selection criteria are undisclosed and change often. OpenAI itself states in its official documentation that there is no way to guarantee top placement.
FAQ
Questions about answer engines
Each answer is written to be quoted as it stands.
Why 7 answer engines?
It is a fixed set chosen from the engines where a brand in Korea can realistically check its visibility: ChatGPT, Google AI Overviews, Perplexity, Claude, Naver AI Search, Copilot and Gemini. That count produces the denominator for citation rate (questions × engines). Changing the set makes comparison with earlier measurements impossible, so it is not changed after a baseline is set. If a new engine is worth measuring, we start a separate set rather than touching the existing one.
Do we have to cover every engine?
Covering all of them is not inefficient, because most of the work overlaps. Opening crawler access, keeping body text in the HTML source, and writing paragraphs as citable units apply the same way everywhere. What differs per engine is roughly the crawler names and the entry path. If budget is limited you can order the work by the engines your customers actually use, and that judgement is made together once the audit results are on the table.
Can we see citation rates per engine?
Not yet. Navirang measures 40 questions × 7 engines = 280 cells against its own site, but we do not have enough publishable sample split by engine. A figure written without conditions cannot be verified — that is the standard we have asked of others in our own articles — so we will publish per-engine numbers together with the measurement conditions and denominator once the sample is there.
Where was the information on this page verified?
Each engine page states the verification method alongside its specification table. Some items are verified in robots.txt, some in server logs, some in a search console. Navirang’s own robots.txt is public and can be read as a configuration example, and the canonical crawler list is kept as a table in a separate article.
See how you appear in all 7 engines right now
Send us a URL and our free audit records mentions and citations separately, engine by engine. We agree the measurement conditions and the question set before we start.
We reply within one business day.