I see more and more SEOs, and those working around our industry, trying to take learnings from one AI Tool and assuming they apply to all of them.
The result is that people are confused about how the different Generative AI tools, all of which we want to have our brands and products showing in, actually work and what we can and can’t do.
Which Generative AI Tools are really search engines, and which just use search engines?
As SEOs we need to think differently about different Generative AI, even if users might think of, for example, ChatGPT as a direct replacement for Google.
How to separate out AI Search from AI Searchers.
One of the easiest ways to split them is to consider which use a search index (in the same way as ‘Traditional’ Google)? And which are just LLMs using a search engine in a similar way to how we would?
ChatGPT, Claude, Deepseek and others are really just like your colleague at work who if they don’t know the answer to your question quickly Googles it and pretends they knew.
Perplexity in contrast works like a search engine, (an AI powered Search Engine), it crawls and has an index, though it may also do searches on other search engines for grounding (seems fair).
Google’s AIOs it seems do rely on finding candidate docs (pages) in Google’s search still to answer a question, so similar to Perplexity.
Interestingly the Bing equivalent to Google AIOs, powered by CoPilot, doesn’t seem to work in the same way, and appears, based on what little documentation there is, to look at pages from Bing’s (traditional) search results to answer the question (I would assume with a query fan out, not just from the accompanying SERPs).
Gemini, at least the basic version right now, is closer to ChatGPT and Claude, it searches on Google, though it then pulls the pages it wants to look at closer from Google’s Index.
Google AI Mode, including Deep Research, I have bundled into ‘Google AI’ here as they use a mix of techniques and this is moving fast, but checking Google’s results seems to be a key part of the process, albeit with a much wider query fan out.
Of course this is all moving fast and what I’m saying could be out of date soon (or even in the time it took me to write this), but based on what I have been able to find (documentation can be limited) this is the current state of Gen AI search.
What this means when we come to optimise
Ok don’t expect a full overview of how to optimise for each tool, those are (several) other posts. But let’s look at what can actually be optimised: as us SEOs think of optimising – i.e. for ranking factors.
Optimising for Google AIOs and Perplexity while not completely different from traditional SEO is still something that makes sense: ranking factors have become ‘inclusion factors’ to get chosen as a source, (but with AI these will never be completely black and white).
Optimising for LLMs like ChatGPT and Claude, right now, has to start with optimising for the search engines they use. (Further optimisation is to improve the chance they will pick your result to look at further and ensure that they can then see your content when they take a look at it).
Deepseek isn’t shown above as it uses a number of different search engine, which partly depends on where you are and what language you are speaking it seems.
n.b. Since this graphic was created ChatGPT has confirmed that it sometimes uses Google instead of or as well as Bing.