Ask the same question in ChatGPT, Claude and Gemini and you will often get three different sets of companies named.
Not slightly different. Different.
Most teams read that as noise. It is not noise. The three systems find sources in different ways and trust different kinds of sources once they have found them, which means being visible in one of them tells you almost nothing about the other two.
Here is what actually separates them, and what it changes about the work.
They do not look for you at the same rate
The first difference is how often a question sends the system to the web at all.
ChatGPT searches for nearly everything. A typical prompt fans out into several queries, skims a couple of dozen domains and settles on a handful. It is a wide, shallow reader.
Claude searches far less. Most prompts are answered from what the model already knows, and only about a third trigger a search. When one does, it tends to run a single query, read a small number of sources properly and cite a few of them. It is a narrow, deep reader.
Gemini mostly does not search in the way the other two do. It leans on what Google has already indexed and ranked, plus its own knowledge graph and its own properties.
This one distinction explains most of the rest. A system that reads twenty domains per answer will find a well-structured page on your own site. A system that reads five will mostly find the sources everyone already links to.
They do not trust the same sources
The second difference is what gets cited once it has been read, and this is where the strategies genuinely diverge.
In ChatGPT, your own site performs well. Better than review sites, better than aggregators, considerably better than news. It is the one major assistant where owned content is the strongest single channel.
In Claude, that flips. Editorial coverage and third party comparisons convert best, and your own site sits in the middle of the pack. Forums sit at the bottom by a wide margin, which is worth knowing if someone has sold you a Reddit strategy.
In Gemini, the question is barely about content type at all. The overwhelming majority of what it cites is already ranking in Google’s top ten. Video and encyclopedic sources take most of what is left.
So the same asset does three different amounts of work depending on where it lands.
A note on the numbers
Figures like these circulate widely and they are useful for direction, not for precision.
They come from samples of prompts at a moment in time, they move whenever a model or its retrieval layer is updated, and they vary a lot by category. Treat them as a description of shape: ChatGPT rewards owned content, Claude rewards being written about, Gemini rewards conventional ranking. Do not build a forecast on the decimal points.
What this means you actually do
Three systems, three jobs. They overlap less than anyone would like.
For ChatGPT, publish properly on your own domain. Cover the whole topic rather than scattering thin posts across it. Say what you are in plain language on every page, not only the homepage. Keep your best pages current, because recency is one of the clearer signals here.
For Claude, get written about. Editorial and comparison coverage is where it converts hardest, and no amount of publishing on your own blog substitutes for it. This is slower work: analyst briefings, contributed pieces, getting into the comparison articles that already rank in your category. Keep pricing and dated facts current too, because price and year language is often what tips a prompt into searching in the first place.
For Gemini, fix your indexing before anything else. If you are not ranking, you are not being cited, and everything else is premature. Then build the entity record, publish video with transcripts, and make sure your structure is machine readable.
The uncomfortable version: if you only have budget for one of these, you are choosing which assistant to be visible in.
The mistake that costs the most
Assuming that visibility in one surface carries to the others.
It is the most common failure we see, and it happens in both directions. Teams with strong SEO assume Gemini has them covered and are surprised to find themselves absent from AI answers even inside Google. Teams with a good content engine assume Claude will find it, when Claude is the system least likely to read their blog.
The related trap is optimising for the wrong crawler. Each system has its own, and blocking one while allowing another produces exactly the asymmetry you would expect. It is worth checking what your robots.txt actually permits, because this is usually decided once by someone who has left and never revisited.
What works everywhere
Underneath the differences, three things hold across all of them.
Be describable. Every one of these systems has to reduce you to a sentence before it can recommend you. If your own copy will not give it one, it will construct something from whatever else it found. We wrote about this at length in what makes an assistant quote you.
Be corroborated. A claim that appears only on your own site is treated as a claim. The same claim in three places is treated as fact, and this is true whether the system reads twenty sources or five.
Be dateable. All three discount sources they cannot place in time when the question involves anything that moves.
How to know whether any of it is working
Citation tracking only shows you the answers that included a link. That is a fraction of the picture, and in Claude’s case a small one, because most answers never cite anything at all.
The measure that matters more is unprompted recall. Take ten questions a buyer would genuinely ask, with no brand names in them, run each one five times in a fresh conversation, and count how often you get named before a competitor does. It takes an afternoon, it needs no tooling, and it tells you something citation dashboards cannot.
Run it separately in each assistant. The results will not match, and the gaps are the brief.
Our visibility audit method sets out the fuller version of this, and GEO vs SEO covers why the underlying logic differs from search.
The short version
ChatGPT reads widely and rewards your own site. Claude reads narrowly and rewards being written about by others. Gemini reads Google and rewards ranking there first.
One strategy will not cover all three, and pretending otherwise is how companies end up strong in one assistant and invisible in the rest.
If you want to know where you currently stand across the three, that is the kind of thing our GEO strategy work starts with.
