Key takeaways
- The engines are not one thing. Perplexity searches live and cites heavily. ChatGPT mixes memory with browsing. Google AI Overviews leans on Google's index. Claude is cautious and cites when it browses.
- All of them prefer pages that state a clear answer early, with facts that agree with other sources.
- Freshness matters more than it did in search. Engines that browse live favour pages that changed recently.
- Being cited by one engine helps with the others, because they read the same web.
- Test each engine separately. A good score on one is not a good score on all four.
People say "AI search" as if it were one product. It is at least four, and they behave differently enough that a business can be named constantly in one and invisible in another. Here is how each engine gathers what it says, based on what the companies publish and what we see running thousands of test questions.
Perplexity: the librarian
Perplexity is the easiest to reason about because it shows its work. Ask a question and it runs live web searches, reads a handful of pages, writes a summary and attaches numbered citations. Every claim points somewhere.
What it trusts: pages that answer the question directly, recent pages, and pages other sites already point to. It leans on review platforms and directories for local questions, and on brand sites for product facts. If your page is the clearest answer on the open web and the crawler can reach it, Perplexity will use it.
What that means for you: answer pages with the question in the title and the answer in the first paragraph. Keep them updated. Perplexity notices a page that changed last week.
ChatGPT: memory plus browsing
ChatGPT answers from two places. The first is what the model learned in training, which is a snapshot of the web from some months ago. The second is live browsing, which it uses when the question seems to need current information, or when the user asks for it.
The training snapshot is why ChatGPT can name a business without citing anything. It has "read about you" before. That is also why a business that has been around for years and been mentioned in many places has a head start, and a new business can feel invisible.
What it trusts when browsing: similar to Perplexity, with more weight on well-known sources and on pages that agree with each other. It is more likely to hedge ("several builders are well reviewed") when sources conflict.
What that means for you: two jobs. Be mentioned in enough places that the next training snapshot includes you, and be the clear answer on your own site for the browsing path. Consistency is the lever. If your name, category and city are stated the same way everywhere, both paths converge on you.
Google AI Overviews: the index, summarized
AI Overviews sits above the classic results and summarizes them. It draws almost entirely from pages Google already ranks for the query, which means classic SEO signals still matter here more than anywhere else. If you rank on page one for the underlying search, you have a chance of being in the overview. If you do not, you almost never are.
What it trusts: pages with strong topical authority, structured data Google can parse, and content that matches the query closely. It favours lists and steps when the question is "how" and named entities when the question is "who."
What that means for you: do not abandon search work. Keep your service pages ranking. Add FAQ and service schema so Google can read the page as answers. And write the "who" pages with your name and location stated plainly, because the overview will lift a sentence that contains them.
Claude: careful, and cites when it browses
Claude answers from training by default and browses when the question calls for it. It tends to be more conservative: it will say when it is not sure, and it prefers sources with obvious credibility. When it browses, it cites, and it favours pages that read as honest and specific over pages that read as marketing.
What that means for you: plain language wins. A page that says "we do fixed-price kitchen renovations from $45,000, typical timeline eight weeks, serving Oakville and Burlington" beats a page of adjectives.
Freshness, and why it matters more now
In classic search, an old page could rank for years. Engines that browse live are different. They re-fetch, and they notice dates. A service page that has not changed since 2021 can lose to a competitor's page updated last month, even if yours is better written.
This does not mean rewriting everything monthly. It means keeping a few key pages genuinely current: prices, availability, recent projects, a dated FAQ. Small, real updates beat large fake ones.
Why being cited anywhere helps everywhere
The engines read the same web. A mention in a local news piece, a listing in an industry directory, a review that names your specialty: each of those is evidence for all four. This is why "citations and authority" is a pillar on its own in our audit. It is the closest thing to a shared input.
How to test yourself, properly
- Write ten questions a real buyer would ask about your category and city. Vary them: "best," "cheapest," "who does X," "compare."
- Ask each one on each engine, from a fresh session, twice.
- Record every business named, in order, and whether you were cited.
- Do the same for your top three competitors by name and note what each engine says about them.
Eighty answers later you will have your share of answer per engine, and a clear view of which one is ignoring you. That is where the work starts. Our free first look does exactly this for four engines and sends you the scorecard.