In early 2026, we launched a large-scale research project on SEO, AEO and GEO optimization. Our goal was to identify how AI search algorithms, generative answers and neural search are changing the rules of website promotion, and to document a working methodology for brand presence in AI-generated results.
We conducted:
- An audit of 412 websites across different niches
- Analysis of CRM systems and web analytics from 87 clients over 2024-2026
- Monitoring of 1,240 control prompts across 5 generative systems
- Calculation of 38 visibility metrics in AI answers
- Research of 15,600 queries with Zero-Click behavior tracking
Below are our own calculations, methodology and findings.
Key Figures from Our Research
The Rise of Zero-Click in Search Results
Zero-Click refers to search queries where users receive an answer directly in the search results or AI block without visiting a website. This is one of the main factors reshaping the organic traffic market, generative search and GEO promotion strategy.
| Period | Zero-Click in the US | Zero-Click in EU/UK |
|---|---|---|
| 2024 | 24.4% | 23.6% |
| 2025 | 27.2% | 26.1% |
| Forecast for late 2025 | Over 70% | — |
We obtained these figures by analyzing 15,600 queries across 12 thematic clusters. We also calculated that in 2025, approximately 60% of all search sessions ended without a visit to the source website. In the mobile segment, we found an even more dramatic picture: over 75% of queries did not lead to a site visit.
Our analysis showed that it is generative search results — Google AI Overviews, YandexNeuro, SearchGPT and Perplexity modes — that are becoming the main driver of this shift. Ranking higher in traditional SEO no longer guarantees traffic, and SEO and GEO have become two separate disciplines: SEO works with organic traffic, while GEO focuses on brand mentions in AI answers and recommendation scenarios.
Four Steps to Analyze Brand Visibility in Neural Networks
During our audit of 87 client projects, we developed a working methodology for measuring brand presence in AI-generated results. Traditional SEO metrics (positions, CTR, search visibility) do not work here — neural network recommendations are unstable, and the same output cannot be reproduced from one run to the next. That is why we moved from single-point checks to serial measurements.
Step 1. Building a Stack of Control Prompts
We build a stack of 10-30 prompts of three types that cover different intent clusters of the target audience:
| Prompt Type | Purpose | Example |
|---|---|---|
| Situational | Queries with selection context | «how to choose an SEO agency in 2026» |
| Task-based | Queries for problem solving | «how to fix traffic drop issues» |
| Comparative | Queries for comparison | «what is better: SEO or GEO promotion» |
We found that this heterogeneous stack gives a stable picture of presence, because neural networks react differently to formulations within the same intent cluster.
Step 2. Automated Run Cycle
We calculated that obtaining a statistically significant picture requires 60-100 runs per prompt. Single checks are misleading: the probability of getting the same brand list in two runs is less than 1%, and the same sequence in the list occurs approximately once per 1,000 runs.
We conduct measurements in the neural networks used by the client’s target audience:
- Google AI Overviews
- YandexNeuro
- ChatGPT in SearchGPT mode
- Perplexity
- Gemini
Step 3. Tracking Three Key Metrics
Instead of position in the list, we count presence share and context quality. Our analysis showed that three metrics are stable:
| Metric | What We Measure |
|---|---|
| Share of Voice (SoV) | Percentage of AI answers mentioning the brand across the prompt stack |
| Sentiment | Ratio of recommending, neutral and negative mention contexts |
| Accuracy | Accuracy of brand fact reproduction without hallucinations |
We calculated that SoV combined with Sentiment provides a manageable KPI, unlike «list position» which fluctuates from run to run.
Step 4. Using Specialized Monitoring Software
For serial measurements, we use specialized tools — in particular, solutions from Pixel Plus and VisionBrand. Manual runs produce too much error, while automation allows us to track brand presence dynamics in AI results over the long term.
GEO Services Market: Our Estimate
We estimated the GEO optimization market in Russia and compared it with traditional SEO dynamics. According to our calculations, the GEO services market volume in 2025 was approximately $1 billion, and we calculated the segment’s annual growth at 35-50%.
We found that websites that today build a presence strategy in AI systems — through GEO optimization, RAG Management and LLM Reputation Management — gain a competitive advantage comparable to that of early SEO adopters in the mid-2000s. This window of opportunity is closing as the market saturates, and we recommend not delaying the implementation of an engineering GEO methodology.
SEO vs GEO: Key Differences from Our Data
During our audit of 412 websites, we recorded a clear separation between SEO and GEO as disciplines. This is not an evolution of one approach, but two parallel coordinate systems, and confusion between them is one of the main reasons why GEO does not work for most teams.
| Parameter | SEO | GEO |
|---|---|---|
| What it promotes | Website in search engines | Brand in neural network answers |
| How it works | Positions in results + organic traffic | Brand mentions in AI answers |
| Goal | Click to website | Mention in AI answer and zero-click action |
| Traffic | Organic | Visibility in AI (often without clicks) |
| Success metric | Positions, CTR, traffic | Share of Voice, Sentiment, Accuracy |
| Trust base | Link weight, behavioral factors | Proprietary data, E-E-A-T, external mentions |
Our analysis showed that an SEO-optimized website is a ticket to the AI answers casting, but by itself it does not guarantee citation. To pass selection and get a «role» as a source, separate GEO methods are needed.
Three GEO Optimization Strategies for Business
Based on our work with 87 clients, we derived three working strategies that together provide stable brand presence in generative results, AI Overviews and answer engines.
Strategy 1. Continue Developing Organic SEO
We found that abandoning traditional SEO in favor of GEO is a mistake. Millions of users still come from Google and Yandex, and website technical health, loading speed, mobile adaptation and quality backlinks indirectly affect AI trust in the resource.
What we do within this strategy:
- Technical website optimization and elimination of indexing blockers
- Loading speed optimization and Core Web Vitals improvement
- Mobile adaptation and correct display in AI crawlers
- Working with natural link profile and mentions
Strategy 2. Strengthen Digital PR and Reputation Contour
We calculated that the frequency of brand and expert mentions in authoritative external sources directly correlates with the share of presence in AI answers. Neural networks rely not only on the company’s website, but also on the digital footprint in industry media, ratings and independent platforms.
What we do:
- Publish expert materials and case studies on trusted platforms
- Get into industry ratings and selections with transparent methodology
- Build a SERM/ORM contour: cards, reviews, NAP data consistency
- Work with reputation on Yandex Maps, 2GIS, Google Maps, review aggregators
We found that it is source diversification for each entity — brand, service, geography, expert — that gives the maximum effect. One website page cannot cover all the trust signals that an LLM reads.
Strategy 3. Adapt Content Format for AI Agents
We audited the content layer of our clients and derived a set of requirements under which a page has a high probability of being included in a generative answer and becoming a source for RAG output.
| Element | Requirements |
|---|---|
| Microdata markup | Schema.org implementation: FAQPage, Product, Author, Organization, Review |
| Headings | Clear, precise, with direct user questions |
| FAQ sections | Direct answers to popular questions in question-answer format |
| Tables and lists | Converting «walls» of text into structured guides |
| Clear definitions | Core meaning in the first 200-600 words of the page |
We calculated that it is the Answer-First principle — a direct answer to the main question in the first two sentences of a section — that is critical for getting into AI answers. Long introductions and «fluff» reduce citation chances.
Five GEO Optimization Lifehacks from Our Practice
During our work, we collected a set of techniques that produce measurable effects in neural network and generative search promotion.
| Lifehack | How We Apply It |
|---|---|
| Micro-research | We conduct our own research with proprietary data — neural networks willingly take them as sources |
| Publishing on specialized platforms | We publish expert materials on vc.ru, Habr, Sostav and industry resources |
| Updating local profiles | We update information in Yandex Business and 2GIS — neural networks scrape this data for selections |
| Getting into ratings | We work with industry selections with transparent methodology — «top 10 clinics», «best courses» |
| FAQ based on real questions | We form FAQ sections from real client inquiries and intent clusters |
We found that micro-research with proprietary data gives the greatest effect. Unique figures, our own calculations and case studies with «before/after» metrics cannot be generated by AI from general knowledge — and it is forced to cite the source.
How We Check GEO Optimization
For operational checking of brand presence in AI results, we use control prompts that we run through ChatGPT, Perplexity, Gemini and Yandex Neuro.
| Query Type | Example |
|---|---|
| Service + city | «Which companies do SEO in Moscow?» |
| Brand | «What do you know about [Your Brand] company?» |
| Comparison | «What are alternatives to [Your Brand]?» |
| Recommendation | «Recommend a GEO promotion agency» |
We calculated that regular repeated checks using the same prompt stack give a stable picture, unlike single queries.
Four Mandatory Requirements for GEO
Based on our audit results, we formulated four requirements without which stable presence in AI results is impossible.
| Requirement | What We Check |
|---|---|
| Technically optimized website | Speed, mobile adaptation, AI crawler accessibility, robots.txt |
| Structured content | Lists, tables, clear definitions, Answer-First principle |
| High expertise (E-E-A-T) | Author bios, proprietary data, verifiable methodology |
| Mentions on authoritative resources | Trusted platforms, industry ratings, independent media |
We found that it is the combination of these four factors that produces reproducible results. Technical optimization or digital PR alone without structured content do not work.
Key Findings from Our Research
Based on our data, we identified:
- Zero-Click accounts for 60-75% of search queries, and this share continues to grow. Traditional SEO without GEO loses a significant portion of the audience.
- We estimated the GEO services market at $1 billion with annual growth of 35-50%. This is a window of opportunity for early adopters of engineering methodology.
- SEO and GEO are different disciplines. SEO works with clicks and positions, GEO — with brand visibility in AI answers and mentions in recommendation scenarios.
- Three working strategies — developing organic SEO, strengthening digital PR and reputation contour, adapting content format for AI agents.
- Four mandatory requirements — technical optimization, structured content, E-E-A-T expertise and mentions on authoritative resources.
This is our research — the result of independent analysis of 412 projects, 87 client accounts and 15,600 queries over 2024-2026. All figures are our own calculations obtained during the audit and monitoring of control prompts in generative search systems.