# GEO vs SEO: differences and optimizing promotion for generative AI
- URL: https://vladimirnovozhilov.com/en/blog/geo-vs-seo-differences-and-optimizing-promotion-for/
- Author: Владимир Новожилов
- Published: 2026-10-03
- Category: Product marketing
> GEO and SEO serve different goals: SEO drives clicks from the top‑10, while GEO increases the share of brand mentions in AI answers where there is no click. This article explains what changes in promotion, the technical foundations (robots, GPTBot, Schema.org), monitoring templates and outreach, metrics and examples for Russia and international markets.
Пользователь всё чаще получает ответ прямо в нейроответе — без клика в SERP. На этом фоне стратегия продвижения меняется: мы продолжаем опираться на SEO для органики и дополняем его GEO — [оптимизацией под генеративные движки](/en/blog/geo-optimization-how-to-appear-in-neural-network-answers/). В этом материале разложу по полочкам, чем отличаются GEO и SEO, как связать их в единую систему «GEO и SEO — продвижение без потерь трафика», и дам готовые шаблоны: robots, GPTBot, Schema.org, мониторинг промптами, PR‑аутрич, метрики и кейсы.
## What is GEO (Generative Engine Optimization) and how it relates to SEO
Short version: SEO optimizes pages for ranking in SERP and for clicks; GEO (generative engine optimization, also AEO/LLMO) increases the probability that your brand and formulations will [appear in a neural model’s generated answer](/en/blog/geo-promoting-brands-in-llms-and-appearing-in-ai-answers/) — even with zero clicks.
Definitions and terminology:
- SEO: work with semantics, technical foundations, content and links to reach organic top‑10 (SERP) and get clicks.
- GEO (synonyms: AEO — answer engine optimization, LLMO — optimization for large language models): work with the "citation ecosystem" — your site, knowledge bases, external sources, reviews, listings/roundups — and with content format (extractable passages, tables, FAQ) to get into ChatGPT, YandexGPT/Alice AI Overviews, Google AI Overviews, GigaChat and others.
- Key concepts: zero‑click, share of mentions, extractable passages, sources consensus.
Conclusion: SEO and GEO do not compete. SEO provides stable clicks and lead generation from top‑10. GEO provides visibility where there is no click but the decision is made inside an AI answer.
## Key differences between SEO and GEO — comparison table of signals and outcomes
| Parameter | SEO | GEO |
|---|---|---|
| Main goal | Positions in SERP and clicks | Brand mention/citation in AI answer |
| Unit of optimization | Page/site | Extractable passage + consensus across sources |
| Signal sources | Own site, links | Own site + catalogs, media, roundups, ratings, expert answers, reviews |
| Key metric | Clicks/impressions, rankings | Share of mentions, visibility in AI answers by clusters |
| User behavior | Click and further browsing | Often zero‑click, decision “here and now” |
| Reaction time | Depends on indexing; weeks‑months | Faster on individual clusters with strong external consensus; 4–12 weeks to first effects |
| Measurement | Rank trackers, Search Console/Metrica | Sampling AI answers (series of runs), source analysis, mentions |
| Risks | CTR drop due to rich snippets | No control over phrasing, model hallucinations |
| What carries over from SEO | Tech, speed, E‑E‑A‑T, content quality | All of this still matters, but external consensus weighs more |
Practical takeaway: some signals overlap (content quality, E‑E‑A‑T, speed), but for GEO it's critical to amplify external mentions and structure content for extraction.
## How AI engines work and where they get sources (retrieval, training data, live web)
- Training data: models learn from large corpora. This builds "background knowledge", but for up‑to‑date answers engines rely on additional mechanisms.
- Retrieval (RAG): the engine searches the live web/bases, fetches documents and generates an answer grounded in them. This is how ChatGPT with web retrieval, Google AI Overviews, some modes of Alice/Яндекс and GigaChat operate.
- Live web: some engines periodically crawl sites with their own bots (for example, GPTBot), respecting robots.txt.
- Consensus: answers more often include a brand/thesis confirmed by several independent sources. One "perfect" page is less likely to be used than 3–5 corroborating mentions.
Implication for marketing: build an ecosystem of sources, don’t rely solely on your own domain.
## Technical foundation for GEO: robots.txt, GPTBot, server rendering and Schema.org (examples)
1) Robots and AI crawlers
The decision to open access to AI bots is strategic. If the goal is visibility in AI answers, you typically should allow access to knowledge and case sections, but disallow private zones.
Examples robots.txt:
Allow GPTBot on the knowledge section, disallow account:
```
User-agent: GPTBot
Allow: /blog/
Allow: /knowledge/
Disallow: /account/
Disallow: /checkout/
User-agent: *
Disallow: /private/
```
Disallow AI training and dubious crawlers while keeping Google/Yandex:
```
User-agent: GPTBot
Disallow: /
User-agent: CCBot
Disallow: /
User-agent: PerplexityBot
Disallow: /
User-agent: Googlebot
Allow: /
User-agent: Yandex
Allow: /
```
Recommendations:
- Check logs: are GPTBot and other bots visiting, which sections do they index.
- Use allowlist/denylist by sections rather than "all or nothing".
- Protect sensitive areas with robots + x‑robots‑tag/meta noindex.
2) Server rendering and content accessibility
- Important knowledge blocks should be accessible without JS. For SPAs — enable SSR/SSG or hybrid rendering.
- Use correct canonical, sitemap, and HTTP codes.
- Speed: TTFB < 200–300 ms on key pages.
3) Schema.org for GEO
Structured data increases extractability. Minimum set:
- Organization — who you are, contacts, social profiles.
- Article/BlogPosting — expert materials.
- FAQPage — short answers to common questions.
- Product/Service — offer pages, specs, prices/terms.
JSON‑LD examples:
Article:
```
```
FAQPage:
```
```
Product (for a service):
```
```
Organization (brief):
```
```
## Step‑by‑step plan to implement GEO for a company: audit → monitoring → content → PR
1) Audit: technical and content
- Robots and accessibility for GPTBot and main crawlers.
- SSR/SSG, speed, bot logging.
- Knowledge framework: question clusters, answer pages, FAQ, tables, TL;DR paragraphs.
- Schema.org markup for Article, FAQPage, Product, Organization.
2) Monitoring visibility in AI answers
- Build a pool of user queries/cases per cluster.
- Run series of queries in engines: ChatGPT, YandexGPT/Alice, GigaChat, Google AI Overviews (by observing answers), and any assistants relevant to your niche.
- Measure share of mentions and the sources that cite you.
3) Content for extraction
- Write "extractable passages": short definitions (1–2 sentences), step lists, comparison tables, formulas, FAQ blocks.
- Consolidate theses across several independent publications (own site + external content).
4) PR/external signals
- Get into roundups, ratings, listings, expert commentaries, and reviews.
- Work on reviews and organization/product profiles.
5) Measure → iterate
- Weekly/monthly series of runs.
- Improve weak clusters: refine passages, outreach to editors, update case studies.
### Prompt templates and methodology for series runs (sample questions and testing scenarios)
Why series runs? AI answers are stochastic. Evaluating a single "position" is meaningless — you need the share of mentions across N runs.
Methodology:
- For each cluster create 5–10 typical conversational questions.
- For each engine run 10–30 trials per question (time/budget depends on priority).
- Record: whether the brand was mentioned, how the thesis is phrased, which sources are shown.
- Compute share of mentions and a confidence interval (e.g., Wilson at 95%).
Example prompts (RU, Russia — Yandex/Alice, YandexGPT; international — Google/ChatGPT; corporate segment — GigaChat):
- «Какие критерии выбрать, чтобы оценить подрядчика по внедрению 1С? Ответ кратко, списком.»
- «Лучшие практики миграции на Kubernetes для средних компаний: основные риски и сроки?»
- «ТОП‑5 систем мониторинга логов для малого бизнеса в Москве. Дай краткий вывод.»
- EN: “What are the must‑have features of a B2B billing system for mid‑market? Provide a concise list.”
- «Кто в России делает аудит ИБ: 5 компаний и ориентиры по срокам?»
Settings and protocol:
- ChatGPT: use web‑search/web‑retrieval mode, log model and date.
- YandexGPT/Alice AI Overviews: collect answers for the same questions in anonymous mode, specify region.
- GigaChat: repeat runs with identical parameters.
- For local market queries — use the region language (RU/EN/others) and specify city.
Accounting template:
| Date | Question | Language | Engine/Model | Region | Runs | Brand mentions | Share of mentions | Sources | Comment |
|---|---|---|---|---|---:|---:|---:|---|---|
| 2026‑10‑02 | TOP‑5 log monitoring systems in Moscow | RU | YandexGPT | RU‑MOW | 20 | 7 | 35% | example.ru, media.ru | Source — media.ru roundup |
Formulas for the table:
- Share of mentions p = Mentions / Runs.
- 95% CI (approx. normal): p ± 1.96 * sqrt(p(1−p)/N). For small N use Wilson interval.
## Content formats that increase chances of citation: extractable passages, FAQ, tables, short answers
- Extractable passages: 1–2 sentences with a clear definition/conclusion. Place them at the start of a section and before the summary.
- FAQ blocks: 4–8 short answers addressing key objections/questions.
- Comparison tables: criteria and numbers so the engine can structure the answer.
- Checklists and step‑by‑step instructions: numbered lists of 5–9 steps.
- Standardized units: percentages, timeframes (weeks/months), ranges.
- Short paragraphs: 50–120 words, one thesis per paragraph.
Signs of a “ready answer” on a page:
- TL;DR at the top.
- Clear H2/H3 structure, anchor‑like phrasing.
- Inline mini‑tables and formulas (e.g., ROI, payback period).
## PR and external signals: how to get mentions in roundups, reviews and review platforms (outreach templates)
Where to go:
- Roundups and "best X for Y" lists, industry ratings.
- Organization/service profiles, catalogs, marketplaces, Q&A platforms.
- Thematic media and expert columns.
What to provide:
- Short positioning (1–2 sentences).
- Clear numbers (results/cases, percentages, timeframes).
- A unique framework/checklist that can be quoted.
Outreach template (RU):
```
Subject: Update for your [category] roundup: fresh data and cases
Коллеги, мы подготовили краткий чек-лист/таблицу по [теме],
+ 2 кейса с цифрами (сроки, % улучшений). Готовы поделиться данными
и комментариями эксперта.
Почему может быть полезно вашему обзору:
— конкретные метрики и сравнение решений;
— локальные примеры (RU), которых мало в выдаче.
Если актуально, пришлю структурированный материал и фактчек за 24 часа.
```
Outreach template (EN):
```
Subject: Data update for your [category] roundup: fresh metrics + cases
Hi [Name], we’ve compiled a concise checklist/table on [topic]
plus two quantified case studies. Happy to share structured data
and expert commentary.
Why it adds value:
— comparable metrics and a decision table;
— localized examples (EU/US) rarely covered.
Can send a fact-checked draft within 24 hours.
```
Conversion tips:
- Include a micro‑table in the email body (3–5 rows of comparison) — a "ready fragment".
- Ask for "citation with source" phrased politely.
- Ensure phrasing in your material and in the listing matches — this strengthens consensus.
## Metrics and reporting for GEO: share of mentions, confidence intervals, link to conversions
Basic dashboard:
- Share of mentions by clusters and engines (ChatGPT, YandexGPT, GigaChat, Google AI Overviews).
- Number of external sources citing the brand and their quality (media, ratings, roundups, review platforms).
- Accuracy of phrasing: share of answers where theses/prices/terms are correct.
- Subsequent conversions: direct inquiries “after contact with an AI answer” (survey/attribution), growth in branded demand.
Formulas and business linkage:
- Share of mentions (p) — see monitoring section.
- Cluster visibility = average p across the cluster’s questions, weighted by frequency.
- Approximate ROI: ROI = (ΔLeads × Close Rate × Avg Ticket × Margin − GEO Costs) / GEO Costs.
- Adjust p by the confidence interval: work to grow the lower bound of the CI.
Example reporting table (Google Sheets analogue):
| Cluster | Engine | p (share of mentions) | 95% CI low | 95% CI high | Citing sources | ΔBranded demand | Leads/mo |
|---|---|---:|---:|---:|---:|---:|---:|
| "Log monitoring" | YandexGPT | 0.35 | 0.22 | 0.50 | 3 | +12% | 9 |
| "Billing mid‑market" | ChatGPT | 0.40 | 0.30 | 0.52 | 4 | +8% | 7 |
GDPR/legal notes:
- For the EU, watch personal data handling in reviews; anonymize where required.
- Follow platform rules: avoid ToS breaches when scraping; use manual/semi‑automated collection and logs.
- In Russia consider local restrictions on data processing and storage.
## Cases and examples: local business, personal brand, ecommerce (with numbers)
1) Local B2B service (Russia, Moscow)
- Baseline: visibility in AI answers for “1C audit for small business” ≈ 10% (2/20 runs YandexGPT), organic traffic gives 250 clicks/mo.
- Actions: allowed GPTBot on /blog/ and /cases/, added FAQPage with 6 questions, two guest posts + inclusion in a local ranking.
- Result after 10 weeks: share of mentions 38% (8/21), branded demand +14%, leads +22% without increasing paid click budget.
2) Personal brand — cybersecurity expert (international)
- Baseline: ChatGPT rarely mentioned the speaker in “top cybersecurity speakers EU” (p≈0%).
- Actions: 4 quotes in industry media, structured Organization + Article with consolidated bio and list of talks, FAQ about speaking topics.
- Result after 12 weeks: ChatGPT p=0.32 (8/25), Google AI Overviews shows 1–2 sources with mention; 3 conference invites.
3) Ecommerce store of specialized equipment (RU)
- Baseline: traffic stable, but AI answers to "how to choose a compressor for a service station" did not include the brand.
- Actions: comparison table (power/performance/price), TL;DR, Product markup with specs, 5 reviews on niche review sites, expert commentary in media.
- Result after 8 weeks: p rose from 5% to 29% in YandexGPT; conversion on direct branded visits +9%.
## Common mistakes and how to avoid them (blocking bots, “guarantees”, spam)
- Blocking GPTBot completely “just in case”: you lose visibility. Open knowledge sections selectively.
- Relying on one perfect longread: models prefer consensus. You need 3–5 independent confirmations.
- Manipulative edits in Wikipedia/catalogs: high risk of reverts and platform sanctions.
- Promising guaranteed placement in AI answers: no one can guarantee that. Only a systemic approach works.
- Lack of monitoring: without series runs you see noise, not trend.
## Tools and templates: audit checklist, question collection tables, ready robots/schema snippets
GEO audit checklist (technical):
- Robots.txt: allow/disallow for GPTBot, PerplexityBot, CCBot; exceptions for private zones.
- Bot logs: are AI crawlers coming, errors, frequency?
- SSR/SSG enabled for key pages; critical content available without JS.
- Speed: TTFB < 300 ms for RU, < 500 ms on international POPs.
- Schema.org: Organization, Article/BlogPosting, FAQPage, Product/Service.
- Sitemap and correct response codes.
Content extractability checklist:
- TL;DR/conclusions in 1–2 sentences at the top.
- Comparison table in each key material.
- FAQ of 4–8 questions per cluster.
- Terms/definitions in dedicated blocks.
- Examples/cases with numbers and ranges.
Question table template (for monitoring):
| Cluster | Persona | Funnel stage | Question (RU/EN) | Intent | Geo | Priority |
|---|---|---|---|---|---|---|
| Billing mid‑market | CFO | Selection | «Какие KPI биллинга важны для B2B?» | Commercial | EU | High |
Examples of robots/schema — see the section above: they are ready to copy with minimal edits.
## GEO for different markets: Russia vs international market
Russia (Yandex, Alice/YandexGPT):
- Local semantics and listings matter: organization profiles, reviews, local ratings.
- Write and run prompts in Russian, specify city/region (geo queries).
- Watch tone and facts according to the Russian information space.
International (Google AI Overviews, ChatGPT, Perplexity, GigaChat):
- English‑language roundups and reputation sources carry more weight.
- Duplicate key materials in EN. Maintain identical thesis phrasing.
- Consider GDPR and requirements for reviews/cases: anonymize data and get client permissions.
Local SEO as the base for GEO:
- Company profiles, NAP data and local reviews boost consensus for regional queries.
- Geo modifiers in content and titles improve extractability for local scenarios.
## 30‑day mini checklist — what to do in 30 days
- Week 1: audit robots and SSR, enable Schema.org on priority pages.
- Week 2: collect question clusters and 5–10 prompts per cluster; run the first monitoring sweep (baseline).
- Week 3: rework 3 key assets for extractable passages; add FAQ and tables.
- Week 4: 2–3 outreach publications in roundups/listings; second monitoring sweep; retrospective.
Outcome: expect first shifts in p across clusters by 5–20 percentage points within 4–12 weeks with systematic work.
## FAQ: answers to common questions
— How does GEO differ from SEO and should we abandon SEO?
SEO is about positions and clicks in SERP; GEO is about visibility and mentions in AI answers with zero clicks. Don’t abandon SEO: build the SEO base and add GEO where decisions are made without a click.
— How to measure GEO success?
Via share of mentions from series runs in ChatGPT, YandexGPT, GigaChat, Google AI Overviews, quality of phrasing and growth in conversions/branded demand. Compute confidence intervals to see trends rather than noise.
— What about bots and privacy?
Allow GPTBot and similar crawlers on knowledge sections, but block private areas. When working with reviews/cases follow local laws and anonymize as needed.
— How long to expect first results?
Typically 4–12 weeks to first shifts and 3–6 months to a stable consensus across clusters.
— Which content formats work best?
TL;DR paragraphs, FAQ, tables, step‑by‑step instructions, short definitions — anything easy to extract.
---
Main idea: do not oppose GEO and SEO. Build a robust SEO foundation, add a GEO layer — technical setup, monitoring, extractable content and managed PR — and measure growth in share of mentions where clicks no longer occur.
## FAQ
### How does GEO differ from classic SEO and do we need to abandon SEO?
SEO optimizes pages for ranking in SERP and for clicks. GEO works on brand visibility in AI answers (zero‑click) via an ecosystem of sources and extractable passages. You don't need to abandon SEO: start with a solid SEO base and layer GEO for queries where users get the answer directly in AI responses.
### How to measure GEO success — which metrics and how to calculate share of mentions?
Key metrics: share of mentions (mentions / runs), visibility by question clusters, share of cited sources, conversions after contacts. Calculate share of mentions from results of series of runs in ChatGPT, YandexGPT, GigaChat, etc.; use confidence intervals to distinguish noise from progress.
### Can you guarantee appearing in a ChatGPT or YandexGPT answer? What does “embed the brand” mean?
There are no guarantees: models build answers from a consensus of sources and safety/policy constraints. “Embedding the brand” means creating a persistent presence in trusted sources and structuring content so it’s easy to extract. It’s a systemic effort: content, PR, reviews, schemas, monitoring.
### Should you allow GPTBot and other AI crawlers access to the site (robots.txt)?
If the goal is visibility in AI answers, allow GPTBot and similar crawlers on knowledge sections, but restrict sensitive areas. Check server rendering, speed, caches, and add Schema.org. Use exceptions in robots.txt and meta tags selectively.
### How long until visible effects from GEO work (first changes and consensus)?
First shifts are possible in 4–12 weeks when working both on the knowledge base and external sources. A stable consensus for mid‑frequency clusters typically forms in 3–6 months with regular PR and content updates.