1. Starting Point: Answers Instead of Link Lists

For more than 25 years, search was a game of ranking as high as possible: users type in a keyword, receive a list of options and click on a result. In 2026, that process is definitively no longer the norm. ChatGPT, which has become an integral part of how users search, reported around 900 million weekly users in February 2026 – more than twice as many as a year earlier. Google’s AI Overviews reach over 2.5 billion monthly users as of May 2026, and the newer AI Mode passed one billion monthly users one year after its launch. Answers instead of link lists are the new standard.

The loudest narrative of the year – that buying directly in the chat is the next logical step – has, however, recently lost some momentum: in March 2026, OpenAI discontinued “Instant Checkout” in ChatGPT, about five months after its US launch and after only around 30 Shopify merchants had gone live. Connecting merchants and keeping product data up to date remained difficult, and Walmart measured a conversion rate roughly three times lower for purchases made directly in the chat than for clicks through to its own website. Conversational commerce is coming anyway – it’s just that the checkout channel isn’t ready yet in 2026. The infrastructure for it keeps growing, but not in a straight line: OpenAI is betting on the Agentic Commerce Protocol, Google on the Agent Payments Protocol (AP2), Perplexity on “Instant Buy” with PayPal, and Shopify is activating “Agentic Storefronts” to syndicate product data to multiple AI systems. Committing to a single system today is an unnecessary risk given this dynamic.

For a brand to achieve conversions through AI search at all, the right foundations have to be laid first, before thinking about checkout in the chat. That means the mention and recommendation in the AI answer must come first, before any purchase is made, regardless of where the transaction takes place. That is what Generative Engine Optimization (GEO) stands for: the targeted optimization of content and data so that AI systems such as ChatGPT, Gemini or Perplexity find, understand and recommend a brand. GEO does not replace classic search engine optimization (SEO); in large part, it builds on it.

2. Deep Dive: What Is GEO Technically?

To use GEO (Generative Engine Optimization) strategically, it helps to look at the technology – and at the names, which have changed several times since 2024. OpenAI’s search prototype “SearchGPT” was integrated into ChatGPT for all users as early as February 2025 and has been called “ChatGPT Search” ever since. At Google, “Gemini” is the underlying model, but not the product that answers in search: “AI Overviews” is the answer box above the classic results, and “AI Mode” is a separate, conversational search tab. Both are powered by Gemini models.

Technically, ChatGPT Search, AI Overviews/AI Mode and Perplexity are based on Retrieval-Augmented Generation (RAG): instead of merely indexing and ranking pages, the systems pull facts from the web in real time and synthesize an answer from them. For this to work, content must still be crawled, indexed and understood: classic SEO fundamentals therefore remain the prerequisite for appearing in AI answers. So what can you do, and what should you avoid?

What the Research Actually Shows

The most frequently cited academic study on GEO (Princeton, IIT Delhi, Georgia Tech and the Allen Institute for AI, KDD 2024, around 10,000 search queries) tested nine optimization tactics. It measured how visible a source is in the generated answer (word share and position). The authors tested in a purpose-built, controlled AI search environment and additionally on Perplexity; ChatGPT Search and AI Overviews were not part of the study. The result is more concrete than most GEO guides suggest (increase in visibility compared with unoptimized content):

FactorSignificance for GEOStrategic approach
Statistics in the text+31% visibility in the AI answerInclude your own figures, studies, benchmarks
Direct quotes from named sources+41% visibility, the strongest single lever in the studyExpert and customer quotes with name/title
Source citations (cite sources)+27% visibility, particularly effective for factual questionsLink to primary sources and data
Fluency optimization (fluent text)+28% visibility through more fluent textsClear, easy-to-read wording instead of filler
Keyword stuffing−8% visibility: the classic SEO tactic tends to hurtNo keyword density; use concrete facts instead

Table 1: Scientifically proven optimization tactics for GEO

Structured data (Schema.org) was not part of the study. Google stresses that the same fundamentals apply to AI Overviews and AI Mode as to classic search, and that neither separate optimization nor special markup is necessary. In our view, it is still worth doing: when products, services and expertise are clearly structured through Schema.org and clean feeds, an AI is less likely to output incorrect or outdated information about a brand.

3. The New Customer Journey: Fewer Clicks, More Purchase Intent

The decline in classic clicks is no longer a forecast but measured reality: for informational queries with an AI Overview, the organic click-through rate fell from 1.76% to 0.61% between June 2024 and September 2025, according to Seer Interactive – a drop of slightly more than 60%. By contrast, brands that are cited in the AI Overview itself receive significantly more clicks than uncited ones, according to the same analysis. At publisher level, Semrush data (analyzed by the Wall Street Journal) for the period June 2025 to June 2026 show declines in Google traffic of between 23% (Politico) and over 85% (Business Insider), although not every decline can be attributed solely to AI Overviews.

At the same time, Adobe – based on more than one trillion visits to US retail sites – shows the opposite when it comes to quality: AI-referred traffic to retail sites converted 54% better than non-AI traffic in May 2026, generated 53% more revenue per visit and grew by 138% within a year. A year earlier, AI traffic had still converted significantly worse than other channels.

The conclusion: there are fewer visitors, but they are significantly more ready to buy. Classic click tracking is thus increasingly misleading as a measure of success. The recommendation happens in the chat; the purchase happens on the merchant’s site.

The more fitting metric is called “Share of Model” (SoM) – a term coined in 2024 by Jack Smyth and Tom Roach (Jellyfish). Originally, it describes how a language model perceives a brand; in practice, what is usually measured is the share of relevant AI answers in which the brand is mentioned or recommended. There are no uniform benchmarks yet, because the values depend heavily on category, prompt set and AI system. What is most meaningful is above all the comparison with direct competitors over time. Unlike classic rankings, there is no “position 4” here – a brand is either cited or invisible.

4. GEO Implementation Roadmap: Five Tactics for 2027

So how can a company position itself for the future, so that it is mentioned in AI systems just as it has been in traditional search? We present five tactics that have proven themselves in our client work and have also been scientifically validated.

A. Content Structuring: From Running Text to “Atomic Content”

The most important information belongs in the first two sentences of a paragraph (direct-answer principle). Lists and tables make comparisons easier to extract. This is a proven rule of thumb from practice, not a requirement confirmed by the providers. In practice, that means: every page answers one specific customer question, headings are clear statements or questions, and core facts such as services, prices, terms and contact options appear as short, unambiguous text on the page rather than in graphics, PDFs or behind tabs. Contradictory or outdated information on different pages should be cleaned up, because systems that cross-check multiple sources otherwise run into contradictions. This builds on classic SEO principles such as search intent and clear page structure, but applies them more consistently to individual, extractable statements.

B. Text Level: Substantiate, Write Clearly, Don’t Over-Optimize

The wording itself also helps determine whether content appears in an AI answer. Substantiated statements work best: back up core statements with concrete figures, link primary sources, and quote expert statements with name and title. Add fluent, easy-to-read language. You should avoid keyword stuffing, i.e. the frequent repetition of search terms: this outdated SEO pattern brings no advantage here.

The tactics work best in combination, for example clear language plus substantiated figures. Which tactic has the strongest effect depends on the topic; for factual questions, source citations are naturally especially important.

C. Technical Foundations: Structured Data for Parallel Systems

According to Google, Schema.org markup (Organization, Person, Product, Review) and clean product feeds are not a prerequisite for AI Overviews, but they are a solid foundation for helping AI systems assign offers correctly. Because of the fragmentation described in the introduction (AP2, Shopify feeds, OpenAI’s Agentic Commerce Protocol), it pays to rely on open, platform-independent data standards rather than tying yourself to a single protocol. Specifically, this means: feed product, price and availability data from a single maintained source, publish it via structured data, and regularly check that the website, feed and third-party systems match. Accessibility also counts: important content should be retrievable by crawlers and not loaded exclusively via script.

D. Brand Authority (E-E-A-T 2.0)

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness. It is the quality framework from Google’s guidelines for quality raters and has been part of the standard SEO repertoire for years. It is an important basis for GEO optimization, because AI validates information by cross-checking multiple sources. Mentions on highly authoritative third-party sites, as well as linking content to real, verifiable author and expert profiles, are considered important trust signals. This means identifying authors with name, title and expertise, publishing your own data and studies that others can cite, and being present on the third-party sites that AI answers in your own category draw on (trade media, comparison portals, industry directories, review platforms).

E. Monitoring: Measure Share of Model Regularly

Prompt testing with the 50 most important customer questions in your own category, manually or tool-supported, in ChatGPT, Gemini and Perplexity. The prompt set is organized by customer intent (inform, compare, decide) and repeated regularly, because AI answers fluctuate. The evaluation looks at whether the brand is mentioned, in what tone, and how large the gap to the most important competitors is. The list of measures for content, data foundation and third-party sources emerges from the deviations. Prompt tracking complements classic SEO reporting made up of rankings, visibility index and clicks; both views belong together.

We recommend starting in this order: first monitoring (E), because it shows where the brand stands today; then the data foundation (C) as well as the structure (A) and wording (B) of the content, which close the gaps; and on an ongoing basis, building authority (D), which takes the most time.

All five tactics target organic visibility, that is, recommendations a brand earns through its data foundation. Alongside this earned visibility, a second channel is emerging: paid placement in the chat. Since the end of August 2026, there has also been a paid route into the AI answer in Germany.

5. Paid Visibility in Chat: ChatGPT Ads in Germany

On August 24, 2026, OpenAI launched ChatGPT Ads in Germany, together with 30 other European markets. Since August 31, companies have been able to book their campaigns directly themselves in Ads Manager, without going through the OpenAI sales team. The ads appear as specially labeled “Sponsored Cards” below an answer, aligned with the context of the conversation; personalized ads, however, are not yet available in the EEA and Switzerland. They are shown exclusively to users on the unpaid Free and Go plans. Anyone paying for Plus, Pro, Business, Enterprise or Edu still sees no advertising.

This follows a familiar pattern from classic search: a purely organic channel becomes a hybrid channel of organic visibility and paid placement – SEO gets its SEA counterpart, only inside the chat rather than next to the results list.

Paid visibility reaches only the free user segment, not the paying users, who are presumably more often high-spending. And a “Sponsored Card” is not the same as an organic recommendation from the AI itself – the difference between bought and earned visibility remains real for a brand’s credibility, just as it always was between a search ad and an organic ranking. OpenAI itself emphasizes that advertising does not influence ChatGPT’s answers. Anyone who relies on ads alone, without creating the data foundation for organic citability described in section 4, is buying reach with only part of the users. However, bought reach does not create a genuine recommendation, as organic content could.

6. Conclusion: The Data Foundation Decides, Not the Checkout Channel

Today, SEO is an interface between data science, brand management and sales. The events of 2025/26 also show, however, that not every bet on conversational commerce pays off immediately. What remains is the need for a solid, structured and up-to-date data foundation – because regardless of which checkout protocol ultimately prevails, this data foundation determines whether an AI knows, understands and recommends a brand at all. Paid placements in the chat do not replace this foundation either: bought reach is no substitute for an earned recommendation. Brands without clean data and a solid SEO base will not be mentioned by AI chatbots.

Artefact helps companies systematically build their visibility in AI answers and classic search: from analyzing how ChatGPT, Gemini and Perplexity portray the brand today, through concrete recommendations for content and data foundation, to the ongoing measurement of KPIs such as Share of Model.

Sources