What we measure
| Engine | What it is |
|---|---|
| ChatGPT | OpenAI’s assistant, including its search-backed answers. |
| Claude | Anthropic’s assistant. |
| Gemini | Google’s assistant. |
| Perplexity | An answer engine built around cited sources. |
| Google AI Overviews | The AI summary that appears above traditional Google results. |
| Google AI Mode | Google’s dedicated conversational search experience, distinct from AI Overviews. |
Google appears twice on purpose. AI Overviews and AI Mode are different surfaces that answer differently, and being present in one does not mean being present in the other.
Why the same page performs differently
Two principles explain most of what you will see in a report.
Shared structure, different sourcing
Good writing structure helps across every engine — a clear, self-contained passage that answers the question is easier for any of them to lift. What differs between engines is sourcing, authority weighting and how much recency matters. So per-engine advice is a matter of emphasis and re-weighting, not of writing six different pages.
Crawler access is not citation
Unblocking training crawlers does not improve citation. GPTBot, ClaudeBot and Google-Extended are training crawlers. Citation runs through separate search and browsing bots. Allowing or blocking the training crawlers affects model training, not whether you get cited in an answer.
This trips up a lot of people, and it is the most common false fix in the category. The free Site Check reports both kinds separately so you can see which is which.
What a pass is, and why more than one
A pass is one complete run of your question set across the engines. Runs are multi-pass because AI answers vary between attempts — ask the same question twice and you can get different sources.
- One pass tells you what happened once. Useful, but a single result can be noise.
- Multiple passes tell you what is stable. A source cited in every pass is a real position; one that appears once is not.
- Stability is only claimed when the run supports it. A single-pass run will not be described as showing a stable pattern, because it cannot.
The same logic applies across runs over time. Movement between two runs is only reported when the runs are comparable; where they are not, the report says so rather than drawing a flattering line.
How we talk about what engines say
Two habits you will notice in reports, both deliberate.
- “Companies mentioned by AI engines” — never “verified competitors.” Models can name companies that do not exist or misattribute products. We report what was said without implying we checked it.
- Emergent competitors are surfaced, not endorsed. If an engine keeps naming a company you have never heard of, that is worth knowing. It is not a recommendation, and it is not proof the company is real.
Common questions
Do I pay extra for particular engines?
No. All six are included in every run. Runs are priced by size - how many questions and how many engines - not by which engines you pick.
Why are Google AI Overviews and Google AI Mode counted separately?
They are different surfaces. AI Overviews is the summary above traditional results; AI Mode is a dedicated conversational search experience. They answer differently and being cited in one does not mean being cited in the other.
Will unblocking GPTBot get me cited by ChatGPT?
No. GPTBot is a training crawler. Citation runs through separate search and browsing bots. This is the most common false fix in the category, and the free Site Check reports the two kinds separately.
How many passes should I run?
More than one, always. A single pass shows what happened once; repeated passes show what is stable. Runs are multi-pass by default for this reason.
What if an engine cannot be reached during a run?
It is excluded from that run and reported as excluded - not counted as a zero against you. A missing measurement and a bad result are different things.