Cited But Not Chosen: How AI Decides Which Brands to Name

Publisher: Chirag Goswami

Think of the last time you looked for a good place to eat or a course to learn a new skill. Many people now skip ten websites. They ask an AI tool and trust the short answer. If a brand is not in that answer, that person may never hear about it.

Search is changing

Earlier, a search gave you a list of links. Now it often gives you a ready answer. SparkToro's 2026 study used Similarweb data and found that 68.01% of US Google searches from January to April 2026 ended without a single click.

There is one catch. SparkToro says it measured 2024 with a different data panel (Datos) than 2026 (Similarweb), and calls the comparison "a bit of apples and oranges." So the exact rise is not certain. But many people do stop at the answer.

Buyers are changing too. G2's "Answer Economy" report, covered by Demand Gen Report in May 2026, says 51% of B2B software buyers now start research with an AI chatbot more often than with Google. It surveyed 1,076 buyers in March 2026. The survey is self-reported and covers software buyers only. G2 also sells software-review services, so it has its own interest here.

Look at your own friends. Many ask ChatGPT first, check Instagram for opinions, and use Google later to confirm. This is only what I see, not a survey. But it fits the studies.

What is GEO?

SEO means getting a higher spot on Google's list. GEO means getting your brand named inside the AI's answer.

Think of the chai shop outside a college gate. It may sit on the main road. But if students ask a senior "where is the best chai?" and he names another shop, that shop gets the crowd. Being easy to find and being recommended are two different things.

How content gets picked

Kevin Indig's study came out in February 2026 (Growth Memo and Search Engine Journal). It looked at 1.2 million ChatGPT answers and 18,012 citations, and checked where on the page each citation came from. The first 30% of a page got 44.2% of the citations. The middle got 31.1%. The last part got 24.7%. The lesson is simple. Put your main answer first.

Conversion summarised the same study. Text that got cited used more direct, firm language (36.2% against 20.2% in text that was not cited). A LinkedIn summary adds that cited text had many more named things like companies, products and people (20.6% against about 5% in normal English).

Here is an example. "There are many good tools" is vague. "Top tools include Salesforce and HubSpot" names real things. The second kind gets picked more. This study looked at ChatGPT only.

AirOps research, reported by Conversion, shows what happens in the background. ChatGPT breaks a question into parts and runs extra searches before it answers. Say you ask for the "best online course to learn Python." The AI may check price, course length, certificates and student reviews one by one. I only read this second-hand. Check the AirOps original before you quote it.

You can try a small test. Ask an AI tool the same question in two different ways. Compare the names it gives. The list may change a little. This is one small test, not proof. But it shows that the answer is not fixed.

Being cited is not the same as being chosen

This was the most interesting finding for me. Lily Ray's study was published in June 2026. She checked 100 B2B "best software" searches on three dates between April and June 2026. Google's AI Overviews often cited lists where a company ranked itself as the best. But in 69% of cases, they left that company out of the recommendation. Of the 80 searches that showed an AI Overview, such lists were cited 323 times.

In simple words, a company wrote "Top 10 tools" and put itself first. The AI used the article as a source. But it recommended someone else. Being a source did not make the company the winner.

There is a limit. The study covers only one area, B2B software, and a small set of searches. It shows a pattern. It does not set a rule for every industry.

To me, a "Top 10" list written by the company in first place feels like a student checking his own paper. So I understand why AI does not trust it fully.

What the evidence supports

Put together, the studies point to a few small lessons:

• Give the main answer early on the page.
• Be specific. Name real tools, places and people.
• Write in a clear and direct way.
• Don't just praise yourself. Being cited does not mean being recommended.

I am not claiming a secret formula. Most of these studies look at one AI tool or one industry. And AI tools keep changing.

This applies to small creators too. On an Instagram page, a clear first line that answers the viewer's question helps people. It may also help AI tools understand what the page is about.

Takeaway

I don't think this is magic. It only shows that people now ask machines first and websites later. For students and small creators, the safe plan is simple. Be clear. Be specific. Don't just praise yourself. Companies that sell GEO tools want this to sound urgent. So keep some distance and test things yourself.

References

1. SparkToro, "The Click Is Now Optional": sparktoro.com/blog/zero-click-search/
2. G2 "Answer Economy" report, via Demand Gen Report: demandgenreport.com/industry-news/news-brief/half-of-b2b-software-buyers-now-start-their-research-with-ai-chatbots-g2/52737 3.Kevin Indig, Search Engine Journal: searchenginejournal.com/the-science-of-how-ai-pays-attention/567597/ 4.Conversion (Indig and AirOps summaries): conversion.com.br/blog/citacoes-chatgpt 5.Lily Ray, "Why Calling Yourself the Best Could Be Helping Your Competitors Win in AI Search": lilyraynyc.substack.com/p/why-calling-yourself-the-best-could 6.Search Engine Land (news report on Lily Ray's study): searchengineland.com/google-ai-overviews-cite-self-serving-listicles-recommend-competitors-480573