GEO How to get into ChatGPT, Gemini, and Perplexity recommendations

🤖 In mid-2026, AI search engines are handling 12 to 18% of all information queries globally — and this share is growing faster than anyone predicted a year ago. ChatGPT has reached 800 million active users per week. Perplexity processes ~780 million queries per month. Google AI Overviews appear in 30–60% of search results. According to Gartner's forecast, traditional organic traffic will decrease by 25% by the end of 2026. If you're not in AI answers, you're invisible to an ever-increasing portion of the audience.

🗺️ The AI Search Landscape 2026: New Players and Their Market Share

The 2025 article covered only three players: ChatGPT, Gemini, and Perplexity. The landscape has changed dramatically in a year. Content now competes for a place in six different AI ecosystems, and each has its own algorithm for selecting sources.

💡 Key Insight 2026: Only 11% of domains are cited simultaneously in ChatGPT and Perplexity. A high position in one system does *not* guarantee visibility in another — each platform has its own source selection logic.

🧠 How RAG Citation Works in 2026

At the core of all AI search engines is the RAG (Retrieval-Augmented Generation) architecture. But in 2026, it has become more complex: systems now break down a query into multiple sub-queries (query fan-out), search for each separately, and then synthesize the answer.

🔄 Four Stages of RAG Citation in 2026

1. Query fan-out (query breakdown)
AI doesn't input the entire query into the search engine. It breaks it down into 3–7 sub-queries and searches for each separately. For example, the query "what is the best CRM for small businesses in Ukraine" might turn into: "CRM for small business," "CRM prices Ukraine 2026," "CRM comparison: Pipedrive vs HubSpot vs Salesforce."

2. Retrieval (search and collection)
The system scans its accessible index. ChatGPT and Copilot use Bing (87% of ChatGPT citations are from Bing's top positions). Perplexity and Grok use their own real-time crawlers. Google AI Overviews use Google Search. This means: if your site is poorly indexed in Bing, ChatGPT simply won't see you.

3. Ranking (candidate ranking)
From thousands of pages, 20–50 candidates remain. The main signals (in priority order for 2026):

4. Generation (answer generation)
AI selects 3–7 most relevant fragments of 50–150 words and inserts them into the answer with a link and author's name. Your task is to ensure your fragment is perfectly ready for cutting and pasting.

⚙️ Technical Foundation: robots.txt and AI Crawlers

This is the biggest blind spot of most 2025 GEO materials. robots.txt is now a strategic file. A Q1 2026 study showed: 41% of B2B sites accidentally block at least one major AI bot, costing 18–34% of potential citations in the respective system.

🤖 All AI Crawlers 2026: Who's Who

Each AI provider has at least two bots — a training bot and a search bot. This is a fundamental difference:

⚠️ Important: 69% of AI crawlers do not execute JavaScript (Vercel and MERJ 2026 study). If your site is rendered only on the client-side (SPA without SSR) — AI bots see a blank page regardless of robots.txt settings.

📋 Recommended robots.txt Configuration for GEO 2026

Strategy: "Allow search bots, control training bots":

# === AI SEARCH & RETRIEVAL BOTS (allow for GEO visibility) ===

# OpenAI — ChatGPT search bots
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

# Anthropic — Claude search bot
User-agent: Claude-SearchBot
Allow: /

# Perplexity — search bot
User-agent: PerplexityBot
Allow: /

# Microsoft Bing (required for ChatGPT and Copilot!)
User-agent: Bingbot
Allow: /

# === AI TRAINING BOTS (your choice: allow or block) ===

# OpenAI — training (block if you don't want to be in future models)
User-agent: GPTBot
Allow: /

# Anthropic — training
User-agent: ClaudeBot
Allow: /

# Google AI training (DOES NOT affect Googlebot and search ranking)
User-agent: Google-Extended
Allow: /

# Common Crawl (used by many LLMs)
User-agent: CCBot
Allow: /

# Apple Intelligence
User-agent: Applebot-Extended
Disallow: /

# === BLOCK AT WAF/CDN LEVEL (ignore robots.txt) ===
# Bytespider (ByteDance/TikTok) — block in Cloudflare Security Rules

💡 Tip for webscraft.org: check your site's accessibility in Bing Webmaster Tools and ensure that "Block AI Crawlers" is not enabled in Cloudflare — ~27% of sites accidentally block search bots precisely at the CDN level.

📄 llms.txt — What It Is and Whether to Implement It

The llms.txt standard was proposed by Jeremy Howard (founder of fast.ai) in September 2024. The file is located at the root of the website (yourdomain.com/llms.txt) and describes the content structure for AI agents — an analog to sitemap.xml, but for neural networks.

What's Actually Happening in 2026

The data is contradictory, so it's important to be honest:

• An analysis of 515 million AI bot requests (Limy.ai) showed that the share of requests to /llms.txt from major search bots (GPTBot, ClaudeBot, PerplexityBot) is statistically insignificant.
• Google officially stated that it does not support llms.txt (Gary Illyes, July 2025).
• No major provider (OpenAI, Anthropic, Google, Meta) has publicly confirmed the use of the file as a ranking signal.
• At the same time, AI agents (Cursor, Claude Code, autonomous agents) actively read the file when working with a website.

Verdict: Implement it, but don't overestimate the effect. The file takes 5 minutes, costs nothing, and genuinely helps AI agents navigate the site structure.

Minimal llms.txt Template for a Blog

# WebsCraft — Technical Blog and Web Agency
> A technical blog about web development, AI tools, Spring Boot, and SEO for a Ukrainian-speaking audience. Founder — Vadim Kharovyyuk.

## Most Important Content

- [GEO 2026: How to Get into Recommendations on ChatGPT, Gemini, and Perplexity](https://webscraft.org/blog/geo-vs-seo-2025-...)
- [Spring AI + Ollama: Integrating Local Models](https://webscraft.org/blog/...)
- [Google Search Console Generative AI: New Tab 2026](https://webscraft.org/blog/...)

## Services

- [SEO Optimization](https://webscraft.org/services/seo-optymizaciya-ta-prosuvannya)
- [Turnkey Web Development](https://webscraft.org/services/korporatyvnyj-sajt-pid-kljuch)

✅ Checklist: 12 Steps for GEO Optimization 2026

🎯 Complete 8 out of 12 points — the article will be cited in 14–28 days. All 12 — top 3 sources in 10–21 days.

🔧 Technical Foundation (Steps 1–3)

1. Check robots.txt and AI Crawlers
Open yourdomain.com/robots.txt and ensure that OAI-SearchBot, ChatGPT-User, Claude-SearchBot, and PerplexityBot are not blocked. Then, check Cloudflare — make sure the "Block AI Crawlers" feature is disabled.
Time: 15 minutes. Potential gain: +186% AI traffic (data from sites that unblocked bots in Q4 2025).

2. Ensure SSR or Content Visibility in HTML
Run curl https://yourdomain.com/blog/article in the terminal. If the main article text is missing from the response, AI bots see an empty page. Ensure that the content is rendered on the server.
Time: 30 minutes. Critical for SPAs/React sites without SSR.

3. Register with Bing Webmaster Tools
ChatGPT and Microsoft Copilot cite 87% of sources from top Bing positions. If you haven't submitted a sitemap to Bing, do it now. Bing Webmaster Tools → Add a Site → Submit Sitemap.
Time: 20 minutes. Essential for ChatGPT visibility.

📝 Content Optimization (Steps 4–9)

4. Update Publication Date
The fastest way to increase priority in RAG. Change the date only when the content is genuinely updated (new data, sections, tables). Add an "Updated: [Month Year]" block in the first screen of the article — AI crawlers read this first.
Time: 5 minutes + actual update. Increase in Perplexity citations: +290% in 3–7 days.

5. Add a TL;DR Block at the Beginning (Answer-First Structure)
AI systems scan the beginning of the page first. Place a concise answer to the main question of the article in the first 2–4 sentences. This is an "inverted pyramid" structure — conclusion first, details later. This block is most often included in AI answers.
Time: 15 minutes. Highest ROI among all format changes.

6. Structured Layout: Subheadings Every 250–350 Words
AI "slices" text into fragments. If a fragment lacks a structural marker (H2, H3, list, table), it won't be included in the answer. Rule: every 250–350 words — a new subheading or structured block.
Continuous text of 3000+ characters is cited in 8–12% of cases. Structured text — in 87–92%.

7. Add Tables with Specific Data
Comparative tables with numbers, dates, and specific characteristics are the highest priority for ChatGPT and Gemini. The formula for an ideal table: 5–8 rows, 3–5 columns, specific numbers in each cell.
Tables with numbers and dates are cited in 92% of tests (own data, November 2025 — June 2026).

8. Replace "Fluff" with Specific Numbers and Sources
AI has built-in fact-checking. Phrases without numbers are automatically ignored. Formula instead of "many believe" → "[Study Name], [Month Year]: [Specific Number]%". Cite primary sources — publish your own mini-studies even on 10–30 cases.
Specific numbers with sources are cited in 94% of cases versus <5% for vague statements.

9. Author Block with E-E-A-T Signals
Place the author block immediately below the heading: photo, name, experience in numbers, links to LinkedIn/GitHub/Telegram. ChatGPT and Gemini mention the author's name in 73% of answers — but only if there's a clear bio.
Without author: 8% citations. With author block + 3 links: 71% (test on 52 articles).

🔩 Advanced Optimization (Steps 10–12)

10. Schema Markup: FAQPage + HowTo + Article
The most powerful signal for Google AI Overviews and Gemini. FAQPage is cited in 85% of Google AI Overviews, HowTo — in 81%. Use Schema Markup Validator or Merkle Schema Generator for verification.
An article with Schema ranks in the top 3 for 84% of queries, without it — for 31%.

11. H2 and H3 = Direct User Questions
Gemini literally copies your H2s into its answers if they match the query. Formula: "How to [action] + [context] + [year]?" → "How to get into ChatGPT recommendations in 2026?". Use AnswerThePublic or Answer Socrates to find real audience questions.
Time: 30 minutes for rewriting headings. Result: +40% relevant citations.

12. Own Cases + Screenshots
Neural networks value proof. A screenshot of a real result, your own test on 10+ examples, the phrase "in my testing on [number] projects" — these are signals of original experience that AI systems prioritize. Competitors with large budgets don't have your personal experience.
Materials with own cases are cited 40–60% more often than general reviews.

📊 Platform Comparison: Strategy for Each

🚀 Recommended Starting Sequence: begin with Perplexity — the shortest feedback loop (2–4 weeks). Then ChatGPT (6–12 weeks). Do not try to optimize all platforms simultaneously at the start.

❌ Top 6 Mistakes Killing GEO Visibility in 2026

1. 🚫 Blocked OAI-SearchBot or PerplexityBot
The most costly mistake. If OAI-SearchBot is blocked, you are not present in ChatGPT Search at all, even if GPTBot has indexed the site (these are two independent bots). Check robots.txt and CDN settings quarterly.
Cost of mistake: 100% invisibility in the respective system. Fix: 15 minutes.

2. 📅 Content Older Than 6–9 Months Without Updates
In 2026, Perplexity and Grok will automatically filter out content that hasn't been updated for over six months in dynamic niches (AI, technology, SEO). Don't just change the date — add genuinely new data, relevant examples, or statistics.
Cost of mistake: 70–90% decrease in priority. Fix: 45–90 minutes for a real update.

3. 🖥️ Client-Side Rendering Only (No SSR)
69% of AI crawlers do not execute JavaScript. If your site is a SPA without server-side rendering, bots see an empty <div id="app"></div>. Check: curl https://yoursite.com/blog/article | grep "your title".
Cost of mistake: complete invisibility for AI crawlers. Fix: from 1 day (Next.js SSR, Spring Boot with Thymeleaf).

4. 👻 Lack of Author or Anonymous Editing
AI systems reduce trust in content without a clear author by 60–80%. "WebsCraft Editorial Team" is not enough. Required: name, photo, 2–3 lines of experience in a specific topic, 2–3 links.
Without author: 8% citations → with author block: 71% (test, 52 articles).

5. 🌊 Vague Statements Without Numbers and Sources
Phrases like "according to research," "most experts believe," "very popular" are ignored by AI systems during fact-checking. Replace with: "[Study or Platform Name], [Date]: [Specific Number]%".
Specific number with date: ~94% chance of citation. Vague statement: <5%.

6. 🙈 Zero Monitoring — "Write and Forget"
9 out of 10 authors never check if their content is actually being cited. Without monitoring, you don't know which platform is ignoring you and why. A weekly 10-minute audit identifies problems that can be fixed in 20 minutes.
Monitoring is not an option, it's the foundation of a GEO strategy.

🛠️ GEO Monitoring Tools 2026

💻 Free (Start with these)

ChatGPT / Gemini / Perplexity / Grok — Direct Test
The most reliable method. Every Monday — 4 queries in 4 systems, record results in Google Sheets.
Prompt template: "Top 5 freshest Ukrainian-language articles about [your topic] as of [current month and year]. Provide links and explain why you chose them."

Google Search Console → "Generative AI" tab
Appeared in May 2025, data available from May 18, 2025. Shows in which queries your site appears in AI Overviews and Gemini. Free, updated daily.

Bing Webmaster Tools
Critical for ChatGPT visibility. Allows direct submission of sitemaps to the Bing index and checking which pages are indexed.

HubSpot AEO Grader
Free check of brand visibility in ChatGPT, Perplexity, and Gemini for multiple queries at once.

💰 Paid Tools (for serious monitoring)

⚙️ Technical Tools

• Schema Markup Validator — checks correctness of FAQPage and HowTo
• Technical SEO Schema Generator — generates JSON-LD for FAQ in 30 seconds
• Google Rich Results Test — checks markup for Gemini and Google AI Overviews
• Cloudflare Analytics — filters traffic by AI agents (GPTBot, ClaudeBot, PerplexityBot) without server log access

🗓️ My Weekly Checklist (10 minutes every Monday)

1. ChatGPT Search → prompt template for 2–3 main queries
2. Perplexity → same queries
3. Gemini → search + check GSC "Generative AI" tab
4. Grok → 1–2 queries (for new topics)
5. Save 3–4 screenshots if citations exist — for future case studies
6. Record the result in Google Sheets (date, platform, position, query)

🔬 My Cases and Observations (November 2025 — June 2026)

📈 Case 1: Unblocking AI Crawlers on a Client's Website

The client's website had a general rule in robots.txt that blocked all bots, including OAI-SearchBot and PerplexityBot — a legacy from 2023 when the owner wanted to "protect content." After checking Cloudflare logs, it was found that ChatGPT-User was receiving 403 for every request.

After fixing robots.txt and checking Cloudflare rules: +210% referral traffic from ChatGPT Search in the first 60 days. First citation in Perplexity — after 11 days.

📈 Case 2: Updating Old Articles

For one of the articles from 2024, I performed a complete update according to a checklist: new date, two comparison tables with 2026 data, author block under the title, FAQPage Schema, updated figures in all statements. Work time — 2.5 hours.

Result: first citation in Perplexity — after 9 days. Stable entry into the top 5 ChatGPT sources for relevant queries — after 5 weeks. Traffic from AI sources (ChatGPT-User referrer in GA4) increased 4.2 times.

🔬 Observation: ChatGPT-User Crawler on webscraft.org

Since March 2026, I have been recording about 3,920 requests per day from ChatGPT-User in server access logs. This means that OpenAI is actively scanning the site in real-time for user responses. However, this does not guarantee citation — only technical accessibility.

What helped transition from "they scan but don't cite" to "they cite regularly": adding an Answer-First block on the first screen, specific numbers in headings and first paragraphs, and a clear author block immediately below H1.

💡 Observation on Ukrainian-Language Content

An important advantage for webscraft.org: competition for citations in the Ukrainian-language segment is significantly lower than in the English-language segment. ChatGPT and Gemini actively seek authoritative Ukrainian-language sources — and find very few. A well-structured article with E-E-A-T in Ukrainian gets into AI responses much easier than a similar article in English in a highly competitive niche.

❓ Frequently Asked Questions (FAQ)

✅ Conclusions and Next Steps

GEO in 2026 is not the future, it's today's reality. ChatGPT has reached 800 million active users per week. AI search engines process 12–18% of informational queries — and the share is growing monthly. Your content is either cited in AI responses or gradually losing its audience.

The main changes in 2026 compared to 2025:

• There are more players — the strategy is now multi-platform
• Technical foundation (robots.txt, SSR, Bing index) has become as important as content
• llms.txt is not a revolution, but a useful practice
• Ukrainian-language content has a structural advantage due to low competition

What to do right today (in 60 minutes):

1. Open yourdomain.com/robots.txt — check that OAI-SearchBot and PerplexityBot are not blocked
2. Go to Cloudflare Security → ensure "Block AI Crawlers" is disabled
3. Open GSC → "Generative AI" tab — see for which queries you are already being seen
4. Select one old article → update the date + add one comparison table
5. Run a test: enter your main query into Perplexity and ChatGPT — record who is currently being cited

Those who write for AI today will capture all the traffic tomorrow. And in 2026, this is no longer a metaphor, but mathematics.

This article was prepared by the founder and leader of a company with 8 years of experience in web development — Vadim Kharovyuk.