Here is a statistic that should reorganise your entire content plan.
The overlap between the top ten links on Google and the sources AI assistants actually cite has fallen from around 70% to under 20%.
Read that again. Four years ago, winning Google essentially meant winning everywhere, because everything downstream drew from the same pool. Today, ranking first and being cited are two largely separate outcomes decided by two different systems.
Which means the SEO report that says you are doing well may be describing a game your customers have stopped playing.
The split is not uniform, and that matters
Not every assistant behaves the same way, and the differences tell you exactly where to spend effort.
Google's AI Mode and Perplexity both draw roughly 90% of their brand citations from Google's conventional top ten results. For those two, classic SEO is still most of the battle. Rank well and you get pulled into the answer.
ChatGPT pulls only about 30% from that same pool.
That single divergence explains most of the confusion in the industry right now. A business ranking well can be highly visible in Perplexity and completely absent from ChatGPT, and both observations are correct at once.
Two systems, two shelves. You can be a bestseller on one and not stocked at all on the other, and the report you are reading only covers one shop.
Where the other 70% comes from
If most ChatGPT citations are not coming from the search top ten, where are they coming from?
From training data, which is a compressed impression of the whole web over years rather than a live index. From places that never ranked well but were widely discussed. Forums, community threads, industry publications, documentation, roundups, review platforms, transcripts, and long tail pages that answered something precisely.
In other words, from the part of the internet that talks about businesses rather than the part optimised to rank.
This is why a business can be invisible in search and yet turn up in an assistant's recommendation, purely because a satisfied customer wrote three paragraphs about them in a busy community thread two years ago.
What this changes about the work
You now need two strategies running at once, and they are not the same strategy.
Strategy one, for the retrieval based systems. Conventional technical and content SEO still works and still matters. Rank for the terms tied to money. Fix your site speed, your structure and your internal linking. For Google AI Mode and Perplexity, this is roughly 90% of the job.
Strategy two, for the training and consensus based systems. This one is unfamiliar and most businesses are doing none of it.
- Be described consistently, everywhere. Same name, same services, same coverage, same specialisation across every profile, directory and mention. Contradiction is the fastest way to become the business a model is not confident enough to name.
- Get mentioned outside your own domain. Industry publications, supplier sites, association listings, roundups, podcasts, community answers. Independent corroboration is what these systems weight.
- Publish things worth quoting. Original data from your own work, specific numbers, named quotes, honest comparisons. Cited sources, hard statistics and quotations all lift citation rates materially.
- Show up where humans actually discuss your category. Not with promotional posts. With genuinely useful answers, under a real name, over time.
How to find out where you stand
You cannot fix what you have not measured, and this measurement is free.
Write down the ten questions a customer would ask before choosing a business like yours. Not keywords. Actual questions, phrased the way a person types them into a chat window.
Then ask all ten in ChatGPT, Google AI Mode, Perplexity and Gemini. Record who gets named, how they are described, and which sources are cited.
Do this once a month. It takes twenty minutes and it will tell you more about your competitive position than any rank tracker, because it is literally the answer your prospects are receiving.
Three patterns usually show up. You are absent everywhere, which is an entity and corroboration problem. You appear in the retrieval based systems only, which means your SEO is working and your reputation footprint is thin. Or you appear inconsistently, described differently in each, which means the internet holds contradictory facts about you.
Each has a different fix and you cannot tell which you have without looking.
The strategic read
Most businesses will keep optimising for the one system they can measure, because rank trackers produce a chart and AI visibility mostly does not yet.
That is the opening. The gap between ranking and being cited is where a smaller, sharper business can get named alongside competitors ten times its size, because reputation footprint is not proportional to marketing budget. It is proportional to how many people have said something specific about you in public.
A business with 200 detailed reviews, four industry mentions and a founder who answers questions honestly in the places customers gather will outperform a business with a bigger site and a bigger budget in exactly the systems that are growing fastest.
Do not abandon what still works
One correction worth making, because the reaction to this statistic is often too extreme.
Some businesses read that the overlap collapsed and conclude that search optimisation is finished. It is not. Two of the largest assistants still draw roughly 90% of their citations from conventional search results, and ordinary Google search remains an enormous source of buying traffic.
The correct response is addition, not replacement. Keep the technical foundation, the rankings and the content that earns them. Then add the reputation footprint work that the training based systems rely on.
Businesses that abandon one for the other end up losing on both shelves at once.
Start with the twenty minute audit
Write your ten questions. Ask them in four assistants. Read what comes back about you and about the competitor who gets named most.
The difference between those two descriptions is not a mystery. It is a to do list, and it is usually shorter than you expect.