An answer instead of a list of links: what it means for traffic on a company website

As of this week, ChatGPT responds with charts, maps and simple apps, and Google DeepMind researchers have shown that a single model can create a ranking without a separate index. We checked which marketing decisions to make right now and which ones to hold off on.

For two decades, the customer journey from a search engine looked much the same. They typed in a question, saw a list of links, clicked, and landed on a company website, where an offer description, comparison tool or calculator was waiting. SEO budgets, sales funnels and traffic reports are built on this model.

Today, there are two signs that the list of links is losing its role as the first point of contact. The first is a feature that OpenAI has just made available to ChatGPT users. The second is a research paper showing where Google Search may be heading. They carry very different weight, so we discuss them separately.

ChatGPT builds a tool in the chat window

On October 7, OpenAI released the GPT-6 model and the Intelligent UI feature. The chatbot chooses the response format itself: plain text, chart, form, map, interactive diagram or simple tool. Examples shown included a Sunday lunch plan that recalculates ingredients based on the number of guests, a driving route on a map with suggested stops, a retirement savings calculator and a restaurant bill splitter. The interface is created gradually while the answer is being generated, and the user does not have to go to any external website.

The feature first reached the Plus, Pro, Business and Enterprise plans (in the latter depending on the administrator's settings), and from October 8 also the free version and the Go plan. Paid plans run on the GPT-6 Sol model, free ones on GPT-6 Luna. According to OpenAI, more than 1.2 billion people use ChatGPT every week.

For companies, this means one thing: calculators, configurators and comparison tools that many websites keep as a magnet for traffic and sales leads can now be assembled by the chatbot on the spot. This coincided with the development of ads in ChatGPT. Two days earlier, on October 5, OpenAI announced a visual ad format, first tested during image generation. The company says sponsored content will be clearly labeled and separated from answers. Our conclusion: the chatbot conversation is becoming the place where the purchase decision is made, and being present there will increasingly require payment.

Ranking without an index: for now, proof on paper

The second signal comes from the paper “Autoregressive Ranking: Bridging the Gap Between Dual and Cross Encoders”, published on January 9, 2026 by Google DeepMind researchers. Today, search engines work in two stages: a fast mechanism selects candidates from the index (an organized database of pages), and a slower one puts them in order. The authors show that a single language model can rank any number of documents without a separate index. In it, a document is not a URL address but a short code that the model generates like words.

The experiments are modest. The researchers fine-tuned Mistral-7B, not a top-tier model. The candidates were provided to the model in the prompt, so this was sorting a ready-made list rather than searching the whole web. The shopping test covered 310 examples. On the WordNet base (about 82,000 English concepts), the share of correct results in positions two through five (recall) rose from 50-60 percent to about 95 percent, but one configuration did a worse job of identifying the best result in first place. No one has replaced a search index with this mechanism.

The paper is, however, another step in the line of Google research from 2021 and 2022. Duane Forrester, who discussed it in Search Engine Journal, points to the consistency: the model generates a short list using beam search, meaning it keeps only a few of the most probable results. If the beam has ten slots, the eleventh position is not weak - it simply does not exist. Visibility becomes binary. Forrester himself calls this informed speculation and discloses that he founded CitationIQ, a company that measures visibility in AI, so he has a stake in this change. We found no critical voices toward his claims, but we also found no comments from the Google researchers themselves.

The drop in clicks is already visible in the data

Even without a new architecture, search traffic is already declining. According to SparkToro, in the first four months of 2026, 68.01 percent of Google searches in the US ended without a click to any website, compared with 60.45 percent in 2024 and 49 percent in 2019. Ahrefs, comparing 300,000 keywords in December 2023 and December 2025, calculated that with an AI Overviews summary (an AI-generated answer above the results), the CTR, or click-through rate, of the first result drops from 7.3 to 1.6 percent. Pew Research Center, on a sample of about 68,000 queries, recorded clicks in 8 percent of cases with an AI summary and in 15 percent without it.

Publishers feel this most strongly: according to Chartbeat, traffic from Google Search to news and media websites fell by 40.2 percent between July 2025 and July 2026. Methodologies and periods differ, but the direction is the same. At the same time, Alphabet reports that Google's revenue from Search and other services (the “Search & other” category) in the second quarter of 2026 amounted to $63.27 billion, 17 percent more than a year earlier. The money stays in search, while the traffic it used to pass on to websites is shrinking.

What to do right now

Intelligent UI is a fact, the DeepMind paper is a direction. That is why low-cost, reversible steps that will help in both scenarios make sense now:

• measure brand presence in ChatGPT answers and AI Overviews alongside rankings and visits, so you have a reference point for later,
• review near-identical pages, for example descriptions of similar products, and give them clearly different content,
• organize product data, specifications and prices so that a machine can read them unambiguously,
• maintain technical SEO, because whether AI systems reach for the content at all depends on it,
• invest in content that the chatbot cannot generate on its own: your own data, research, customer experiences.

It is worth waiting before rebuilding your entire strategy around a single research paper and before buying expensive services promising a guaranteed place in AI answers. Giving up on SEO would also be a mistake: it is still the foundation on which generative answers rely.

So the answer to the question in the title is calmer than headlines about the end of SEO suggest. The company website is not disappearing, but it is no longer the only place of first contact. The winners will be the companies that start measuring presence in answers before the list of links becomes only an internal step of the algorithm.