GEO Experiments 2026: What We Tested, What Failed, and What Actually Works

Everyone has opinions about Generative Engine Optimization. We have data. Over the past year, the OtterlyAI team ran dozens of controlled experiments - testing llms.txt, Schema Markup, AI-generated vs. human-written content, YouTube citation patterns, earned media outreach, and much more - tracking real outcomes across the largest AI Search platforms. Some results confirmed our hunches. Others completely surprised us. In this session, Thomas Peham walks through what worked, what flopped (looking at you, llms.txt), and the one experiment that got otterly.ai cited on ChatGPT within 24 hours. Expect raw data, honest failures, and a repeatable experimentation methodology you can take back to your team on Monday.

Transcript

1. GEO Experiments 2026: What We Tested, What Failed, and What Actually Works Thomas Peham OtterlyAI @thomaspeham
2. GEO Experiments 2026: What We Tested, What Failed, and What Actually Works brightonSEO 2026
3. I’m Thomas Peham, CEO & Co-Founder at > ex-VP Marketing at Storyblok > in B2B SaaS since 2014 > in Marketing since 2010 Hello. Otterly.AI
4. Doing SEO since 2010 Otterly.AI
5. Otterly.AI Trent Kennelly OtterlyAI is the Intelligence Platform for AI Search
6. We Analyzed 100M Websites. And we Publish Everything We Learned About AI Search & GEO.
8. 1% referral traffic from ChatGPT What you’ll see in your web analytics
9. 10% or more from ChatGPT What you’ll hear from your prospects and customers ChatGPT & AI
10. 12% of our Website Traffic is coming from AI Crawlers and Agents OpenAI GPTBot ChatGPT-User OAI-SearchBot Gemini Deep Research Perplexity-Bot Perplexity-User ClaudeBot Claude Searchbot Claude-User
11. Measurement of Success has already changed
12. The AI Search Funnel Otterly.AI Brand Coverage (supported by Website Citations) (Human) Referral Traffic Branded + Direct Traffic (indirect) Conversions AI Bot/Agent Traffic On Page AI Search
13. GEO Experiments Inside OtterlyAI’s GEO Research
14. Get All GEO Experiments Free GEO Sheet: Send me a LinkedIn DM or email (thomas@otterly.ai) - I’ll send you our GEO Experimentation Spreadsheet Otterly.AI
15. OtterlyAI Experiment Methodology Are we tracking AI bot visits, brand mentions, and AI citation rates over a fixed time window? Measure Relentlessly What GEO tactics & myths are we testing, and what result do we expect it to produce? Form a Hypothesis Did our experiment move the needle, and what can other GEO practitioners learn from it? Publish Results Otterly.AI Feedback Community
16. #1 llms.txt Experiment The Myth LLMs.txt Files - proposed as a "robots.txt for AI" - promoted as a necessary tool to increase AI visibility - help improve crawlability and citability on AI Search Otterly.AI
17. Otterly.AI LLMs.txt Results: Only 0.1% of AI bot traffic accessed /llms.txt - We created a llms.txt and embedded it in our footer - llms.txt performed 3x worse than average pages - No positive correlation between llms.txt presence and increased AI crawler activity
18. #2 Human vs AI Content Experiment The Test AI vs Human Published Content Create two landing pages on the same topic. Landing Page 1: otterly.ai/agencies Created by our team Landing Page 2: otterly.ai/agencies-ai-search-monitoring Created with AI Landing Page Creator Otterly.AI
19. #2 Human vs AI Content Experiment Otterly.AI
20. The Results: AI-assisted content outperforms human-generated content Visibility on Google Organic, ChatGPT, Google AI Overviews, Gemini Source: otterly.ai
21. When does AI content work? • AI content output is influenced by AI search patterns • Based on patterns we see work across industries, Google and AI Source: geo.otterly.ai/geo/ai-landingpage-creator/
22. But: Don’t do AI Slop AI content on Google Source: ahrefs.com
23. #3 Experiment YouTube AI Citation Study Otterly.AI
24. Otterly.AI The YouTube Study for AI Search • 100M citations over 30 days • YouTube 2nd biggest social media channel cited • 1.8M Youtube Citations
25. Otterly.AI Where Is YouTube Cited Most in AI Search?
26. Otterly.AI The Results Popularity metrics (likes, views, comments) don’t matter for AI Search Active channels have more videos = more chance of being cited
27. #4 Experiment Schema Markup Otterly.AI
28. Otterly.AI Schema lifted Google AIO 1,500% but ChatGPT citations dropped AI Search Platform AI Citation Increase or Decrease? ChatGPT decrease AI Overviews increase AI Mode increase Gemini decrease Perplexity No impact Copilot decrease
29. Otterly.AI Results: AI couldn’t answer the question despite the Schema being on the page
30. #5 Press Release Distribution Otterly.AI The Theory Press Releases are a great tool to: - keep AI updated on our brand - get content on high-ranking domains - to build brand awareness and brand search
31. Press Release Distribution: 10 Press Releases distributed between Sept 2025 - February 2026 Otterly.AI
32. Press Release Distribution: The Results Otterly.AI - First Website Citations from businessinsider.com, yahoo.com within days. - Overall increase in brand mentions since start of Press Release Distribution in October 2025
33. #6 HTML vs Markdown Otterly.AI The Experiment: We placed new HTML and .md URLs in our website footer. They had the exact same URL structure, giving all crawlers an equal structural opportunity to discover them.
34. HTML vs Markdown: Markdown versions received 0% AI Crawler traffic Otterly.AI 4.6% AI Crawler visits to HTML 0% to Markdown pages
35. #7 Localization Experiment Otterly.AI What We Did: We analyzed 100+ prompts in Finance, Insurance and Pharma in US, CA, AUS, UK, NL, IT. All prompts were written in English + local language.
36. The Localization Factors: Do AI Search Engines Localize by Country & Language? ✅ General Location ✅ System Settings ✅ Contextual Data ✅ Interaction History ✅ Demographics (indirect) ❌ Precise GPS Location ❌ IP Address ❌ Name, Age, Email ❌ Any internet browsing history ❌ Demographic details (gender, education,...)
37. Different Country Performance on Brand vs 3rd Party Websites Otterly.AI USA relies heavily on news/media; UK and Netherlands favor brand/comparison sites Insurance comparison sites dominate in UK/NL, while US relies on financial media (Forbes, NerdWallet, Bankrate)
38. AI Platforms Behave Differently Otterly.AI Percentage of citations going to local/regional sources - Perplexity fails to localize Avg Localization Index Copilot 77.4% HIGH Google AI Mode 68.5% HIGH Google AIO 65.7% HIGH ChatGPT 58.2% MED Perplexity 9.3% V.LOW
39. Prompt Language Changes Everything: Italian Prompts in Italy unlock local content.
40. Otterly.AI WINNERS What Actually Worked in AI Search The experiments that moved the needle on brand mentions and citations. 1 AI-Assisted Content 5x more visibility AI-generated landing page outperformed human-written version across Google, ChatGPT, AI Overviews & Gemini. 2 Press Release Distribution Days to first citation 10 releases → first citations from businessinsider.com & yahoo.com within days. Brand coverage trending up since Oct '25. 3 Wikipedia Presence #2 most cited page Our own Wikipedia article is the 2nd most cited Wikipedia page for GEO queries — even draft/declined pages get cited by ChatGPT. 4 Schema Markup (for Google) +1,500% AI Overviews Sitewide schema rollout correlated with +1,500% AIO and +377% AI Mode citations. No lift on ChatGPT / Gemini / Copilot. 5 YouTube Optimization #2 social channel cited YouTube is the 2nd most-cited social source (1.8% share). Popularity metrics don't matter — active channels with more videos win. Otterly.AI
41. OtterlyAI LOSERS What Wasted Our Time in AI Search The GEO tactics that sound smart on Twitter — but didn't hold up in the data. 1 llms.txt 0.1% AI bot traffic Performed 3x worse than average pages. No correlation between llms.txt presence and crawler activity. Major platforms ignore it. 2 Markdown (.md) Mirrors 0% crawler traffic HTML versions got 4.6% AI crawler visits. Markdown mirrors got exactly zero. Don't duplicate your pages as .md for GEO. 3 Pure AI-Generated Content ↓ on Google over time Pure AI content ranks short-term then collapses on Google. Use AI to assist — not to replace — human editorial. 4 Schema for ChatGPT & co. ↓ on 4 of 6 engines Schema citations dropped on ChatGPT, Gemini and Copilot; no impact on Perplexity. Schema couldn't even surface a unique fact we embedded. Otterly.AI
42. OtterlyAI THE 6 KEY TAKEAWAYS What We Learned From GEO Experiments 1 Measure beyond clicks AI referral traffic is <1% in analytics but up to 10%+ in agent traffic. Track brand coverage, citations & AI bot traffic - not just GA. 2 AI-assisted > human-only Our AI-assisted landing page drove 5x more visibility than the hand-crafted one. But pure AI content still loses on Google. 3 Earn third-party real estate Press releases, Wikipedia & news sites drive brand mentions within days. You don't control the narrative — they do. 4 Schema helps Google, not ChatGPT Sitewide schema delivered +1,500% on AI Overviews and +377% on AI Mode - but decreased citations on ChatGPT, Gemini & Copilot. 5 Skip the GEO folklore llms.txt got 0.1% of AI bot traffic. Markdown mirrors got 0%. Stop chasing rituals - focus on HTML quality and distribution. 6 Localize the prompt, not just the page Italian prompts unlocked up to +59pts of local citations on Perplexity. Every engine localizes differently - Google leads, Perplexity lags. Otterly.AI
43. Let’s connect Thomas Peham, CEO & Co-Founder at Free GEO Sheet: Send me a LinkedIn DM or email (thomas@otterly.ai) - I’ll send you our GEO Experimentation Spreadsheet Also at brightonSEO today: 5.5M AI Citations Later: What YouTube Teaches Us About AI Search Otterly.AI Thanks! You’ve been an otterly amazing audience