You can rank number one on Google and still be completely invisible in ChatGPT. In my study, we found ten companies that ranked first on Google but were never mentioned by ChatGPT in any of the runs where the corresponding question was tested. We studied Swedish companies across several industries and locations. For each question, we collected Google’s results and then compared the companies appearing there with the companies recommended by ChatGPT. The AI part of the study used five different versions of each question, and each version was run three times.
Of the 269 company listings from Google’s results, 159 were mentioned by ChatGPT. That is 59.1 percent, which means that more than four out of ten companies visible on Google did not appear in ChatGPT. What I found most interesting, however, was the rankings. Companies in positions 1 to 3 on Google were mentioned by ChatGPT in 60.2 percent of cases, while companies in positions 4 to 10 were mentioned in 58.6 percent. The difference was just 1.6 percentage points.
I had expected a much clearer relationship. If Google considers a company the best answer to a question, I would at least expect ChatGPT to be considerably more likely to recommend it as well. That was not what we found. There is clearly some relationship between visibility in Google and visibility in ChatGPT, but a strong Google ranking is far from a guarantee that you will also become visible in an AI answer. This would have been less interesting if ChatGPT were still a fairly small channel, but it is not. OpenAI stated in August 2026 that ChatGPT is used by around one billion people every week. The company has also said that use of its search function has almost tripled in a year. The point is not that Google has disappeared or that you should move all your work from Google to ChatGPT. The point is that people now have another way to find companies, products and services, and our study shows that the two systems do not choose the same companies particularly often.
If you work with SEO and only measure what happens in Google, you are therefore missing part of how people actually find information. How large that part is will vary between industries and audiences, but the development is moving quickly enough that it no longer makes sense to wait for AI search to become a major channel sometime in the future. It is already large enough to be part of the work.
Overlap by industry
Overlap by city
AI needs to become part of SEO
Several new names have appeared for optimization aimed at language models. GEO, AEO and LLMO are some of them. I am not particularly convinced that we need a new profession every time search results change. Much of what appears to influence visibility in AI is already part of the work of a good SEO specialist, including content, technology, links, brand, external mentions and authority. The difference is that the work needs to become broader. A strong Google ranking is still valuable, but our study shows that it does not automatically make you a good answer for ChatGPT. You do not need to throw away traditional SEO and start again. You need to adapt parts of the work so they perform better both in traditional search results and when a language model decides which companies, products or sources to recommend.
Traditional SEO often starts with keywords. You might identify the search accounting firm Gothenburg and investigate how many people search for it, which companies rank and what kind of content is needed to compete. In ChatGPT, the same need might be expressed as Which accounting firms in Gothenburg are good for smaller SaaS companies that sell internationally? It is still the same customer and the same market, but the question contains much more information about what the person is actually trying to choose.
International studies also show that ChatGPT often performs several searches of its own behind the scenes before putting together an answer. In one study of 15,000 prompts, more than 43,000 underlying searches were generated, and 95 percent of those follow up searches had no traditional search volume. That is not particularly surprising. Many questions are too specific for enough people to search for exactly the same wording for them to appear as individual keywords in traditional tools. This does not mean search volume has become irrelevant. It means keyword research needs to be complemented by a better understanding of the decision the customer is trying to make. You need to understand which problems the customer is trying to solve, which alternatives they are comparing and which criteria determine which company becomes relevant. Some of those questions will never have impressive search volume, but they may still play a major role when a language model puts together a recommendation.
Content needs to give clearer answers
Some of what international studies point to is fairly undramatic. Focused content that actually answers the question appears to be easier to use in AI answers than long articles that try to cover everything around a subject. This is really just good SEO even without ChatGPT. If someone wants to know how interest on a business loan works, they do not first need to read 800 words about how running a business can be both exciting and challenging. They need the answer.
Definitions, numbers, comparisons, prices, examples and clear processes are useful because they help the reader understand the subject. The same things also make it easier for an AI system to extract relevant information when it puts together an answer. That does not mean you should start writing strange texts for robots or split every article into a hundred tiny questions and answers. It means every page needs to have a clear job and actually do that job.
If you have your own data, it becomes even more interesting. Publish the study, show the prices, compare the alternatives, describe who the service is suitable for and who it is not suitable for, and explain how the process actually works. Original information gives people a reason to read and gives other websites a reason to refer to you. At the same time, it creates clear facts that both Google and AI systems can use.
You need to exist outside your own website
This is where I think AI makes an old SEO question more interesting. SEO has long focused heavily on links. Who links to you, how strong is the page and what kind of link do you get? Those are still relevant questions, but for AI it also matters what information exists about you, not just what information comes from you.
Several international studies have found relationships between AI visibility and external brand mentions. Ahrefs analyzed 75,000 brands and found much stronger relationships between AI visibility and mentions on other websites and YouTube than between AI visibility and the amount of content on the company’s own website. This is correlation and does not show that a mention itself causes greater visibility, but the pattern is interesting enough to affect how I would work.
If a company is described in industry media, tests, comparisons, guides, videos and other credible sources, there are more external data points that can help an AI system understand what the company is and what it is known for. That makes digital PR more relevant to SEO. It makes brand work more relevant and it also makes mentions without a perfect SEO link more interesting. I still want the link, but I would not automatically dismiss a good publication just because the link is nofollow. If the company is mentioned clearly, in the right context and on a relevant website, that publication may have value that is not captured by the old question of link equity.
It also affects how you need to think about brand. An AI system is not simply trying to find a page. It needs to be able to answer questions about which companies are good at something, what distinguishes them and which company is suitable for a particular type of customer. To do that, the system needs to understand more than what is written on the company’s own homepage. You can write that you are the best as many times as you like. It becomes much more interesting if independent sources repeatedly describe the company as strong in a particular area. That is why I think the boundary between SEO, PR and brand work will become less clear. It is difficult to optimize your way into making other people talk about you. You actually need to give them a reason to do it.
The technical work is still fairly ordinary SEO
The technical part is much less dramatic. Google itself says that its AI features rely on the same basic technical requirements as ordinary search. The page needs to be crawlable and indexable, important content needs to be readable, internal links need to work and structured data needs to describe what is actually on the page. Those are things that should already be part of a normal technical review.
OpenAI similarly has a specific bot for ChatGPT Search called OAI SearchBot. It is also worth distinguishing it from GPTBot. GPTBot is used for model training, while OAI SearchBot is used for search, which means you can choose to block training without simultaneously blocking the possibility of appearing in ChatGPT Search. I would check that kind of thing as part of normal technical SEO work.
I would, however, be cautious about technical products claiming that websites need some completely new AI structure. Google has explicitly said that you do not need special AI markup or an llms.txt file to appear in its AI features. That is a fairly good reason not to start with that type of solution. If the website already has problems with crawling, content or internal linking, there are more fundamental things to fix first.
AI visibility needs to be measured differently
The biggest practical difference from traditional rank tracking is that AI answers vary. If you rank fourth on Google, you can check again and get roughly the same result. ChatGPT does not work like that. The same question can produce different companies in different runs, a different wording can produce a different answer and another language model can choose completely different companies. That is why we did not ask ChatGPT once in our study. Each question was tested with five different prompt variations and each variation was run three times. That gave us 15 AI answers for every question studied. It is still not an absolute truth, but it provides a much better basis than asking once, taking a screenshot and treating it as a stable search result.
I therefore think AI visibility needs to become part of normal SEO reporting. The most interesting things to track are how often the company is mentioned, which competitors are mentioned, which external sources are used and where there are large differences between Google and AI. Those differences are particularly interesting because they can reveal things that a traditional ranking report misses.
If you rank poorly on Google but are still frequently recommended by ChatGPT, there is something else the system is picking up. It could be external sources, a clear brand, reviews, specialization or some other signal that does not appear in your ordinary ranking report. If you rank first on Google and are never mentioned by ChatGPT, you have the opposite problem. Those differences are where I would start looking for what actually separates the systems.
We still know relatively little
There are already plenty of lists claiming to contain proven GEO factors, but I would be cautious with them. A research review from 2026 examined 45 studies on optimization for generative search systems. The most important conclusion is that we still do not have a universal method that has been proven to deliver consistently better AI visibility across several models, questions and platforms. The results vary too much for us to say that there is a simple recipe.
That does not mean you should ignore the field. It means you should not throw away what we know works because someone has invented a new abbreviation and a new package to sell. AI usage is growing extremely quickly and people are already using language models to find companies, compare alternatives and make decisions. At the same time, our own study shows that a strong Google ranking is far from a guarantee that ChatGPT will choose you.
That is why I do not think AI optimization should sit next to SEO as a separate project. It should be a natural part of SEO.
How to do AI optimization
Measure before you start optimizing
Start by measuring how often the company is mentioned, which competitors are recommended and which sources are used. Test several versions of the same question and repeat the tests, since AI answers vary. The goal is to identify a clear pattern, not an exact position.
Add prompt analysis to keyword research
Keywords are still important, but complement them with the questions customers actually ask AI. Try to understand what problem they want to solve, which alternatives they compare and which criteria they use to choose.
Give clear answers
Make the answer easy to find. Show prices, differences, definitions, examples and who the service is suitable for. Good content does not need to be written for robots, but it should be clear enough for both people and AI systems to understand.
Publish your own information
Original data is particularly valuable. Publish studies, statistics, prices, tests, comparisons and your own experiences. This gives both people and other websites a reason to refer to you.
Make sure others talk about the company
AI systems use many external sources. Try to appear in industry media, tests, comparisons, podcasts, YouTube and other relevant contexts. It is more valuable when others describe what the company is good at than when you are the only one saying it.
Build links where the context is right
Good links are still important, but context matters more. I would rather have a relevant mention and a nofollow link from a strong industry publication than a perfect follow link from an irrelevant website.
Check the technical foundation
Make sure important pages can be crawled and indexed and that internal linking works. Also check that OAI SearchBot is not blocked if you want to appear in ChatGPT Search. Start with the technical foundation before chasing new AI solutions.
Look for differences between Google and AI
Pay particular attention to questions where Google and AI choose different companies. If you rank first on Google but are never mentioned by ChatGPT, compare yourself with the companies that are actually recommended. Those differences are often where you find the most interesting things to improve.
