Perplexity, ChatGPT, Gemini: how the answer engines really search

The shared principle

For current questions, all three work to the same pattern: they break the question into search queries, fetch results, read some of the pages found, and build an answer with source references from them.

From that follows the most important point for your own visibility: what does not appear in conventional search rarely appears in the answers either. Answer engine optimisation without the groundwork of search engine optimisation does not work.

Where they differ

The crawlers and their roles

One point that regularly gets confused: there are two kinds of request, and they are controlled separately.

The distinction matters practically: blocking all AI crawlers prevents not only use in training but also being named as a source in answers. Blocking only the training crawlers keeps you visible in the answers — a compromise many choose deliberately.

What works on all three

Despite the differences, four measures take effect everywhere.

1. Answer first, per section. The first two sentences under a heading have to answer the question completely. That is exactly the passage that gets extracted.
2. Readable without JavaScript. Content that only appears in the browser is not captured by some fetches. That is the most common technical reason for exclusion of all.
3. Verifiable original details. Figures, methods, named experience. Systems favour pages with specific content over summaries when choosing sources.
4. Visible accountability. Author with role, date of last review, company with an address. Marked up as structured data, visible on the page.

How to measure it

There is no console. What there is, is a repeatable procedure — and repeatable matters more than precise.

1. Fix twenty questions. From your field, phrased the way a customer would put them. That list stays unchanged for a year.
2. Ask them quarterly across all three systems. Without a signed-in account, so personalisation plays no part.
3. Note three things: are you named? which competitors are named? which sources serve as evidence?
4. Cross-check the server logs. Which crawlers came, which pages did they fetch.

The third line is the most useful. The cited sources show what kind of page the system considers citable — and that is a more concrete instruction than any general recommendation.

Help me build a fixed question list for checking visibility in
answer engines.

Our situation:
- What we offer: [offering]
- Audience: [as narrow as possible]
- Region and language: [details]
- Topics we have content on: [list]

Tasks:
1. Phrase 20 questions the way our audience would actually ask
   them – in their language, not our technical vocabulary.
   Distribute them like this: 8 questions describing the problem,
   8 on method, 4 on comparing providers.
2. For each question, mark whether we already have content on it
   (based on my topic list) or not.
3. Name the 5 questions where being mentioned would be worth most
   to us, and justify that.
4. Propose a simple table for recording the results each quarter.

Do not answer the questions yourself, and do not tell me whether
we are currently mentioned – I will check that in the systems.

In closing

The systems differ measurably in how many sources they name and how easily small providers get there. For practical work that changes the order, not the measures: all four effective steps work everywhere.

If time is short, do two things first — put a short summary at the top of every piece, and check that the page is fully readable without JavaScript. After that the question list is worth setting up, so you can see whether anything is moving at all.

Common questions