GEO is not "Just SEO" (or maybe it is?)

Is GEO (Generative Engine Optimization) just SEO with a new name? In this Search Central Live Deep Dive 2026 Barcelona lightning talk, Thiago Pojda (Director of SEO at SIXT, co-founder of https://wesearch.media) argues that both answers are right: for the few who always did the whole job, GEO is SEO. For everyone else, what we called SEO was only a fraction of it.

What the talk covers:
- Examples of what we got wrong: JSON-LD schema was sold as an AI citation lever, but a re-test of 1,885 pages found no meaningful gain (AI Overviews dropped 4.6%, the only statistically significant result). llms.txt and author bio boxes get the same treatment.
- What is plumbing, not gaming: AI crawlers like GPTBot and ClaudeBot don't execute JavaScript, so content must be in the initial HTML.
- What works but is fragile: listicles (+16.5pp mention chance across 200k AI responses), and why engines verify that a claim exists, not whether the source is honest.
- What's new: query fan-out. 43% of AI research steps run on the English-language web, even for Spanish, German or Turkish queries.
- What correlates with AI Overview visibility: branded web mentions (0.664) vs. backlinks (0.218) across 75k brands.

The takeaway: Let's call it GEO. Google says it's all still SEO, and that's true when SEO is done properly. We hurt SEO's reputation by doing small parts of the job and selling them as the whole thing. A new name is a clean slate, not a fix. The fix is doing the whole job: build for the person reading, get described accurately across the web, and sort out the plumbing before chasing scores.

Sources: Ahrefs, Peec AI, DEJAN, iPullRank, Vercel/MERJ, SE Ranking, Amplerank.
Speaker: https://pojda.dev | https://linkedin.com/in/pojda/

Transcript

1. Search Central Live Deep Dive 2026 Lightning Talk GEO is not "Just SEO" (or maybe it is?) Thiago Pojda SEO Director & Entrepreneur | SIXT SE
4. Nothing changed for two people No score to lose Never asked what the algorithm wanted. Everyone else Already playing the real game Server Logs, entities, brand description. Research changes the product.
5. Nothing changed for two people No score to lose Already playing Keyword research in public tools People Also Asked -> H2 A customer complaint now and then
6. A lot changed for you Don't let anyone tell you it didn't
7. We were told Be fast. Be useful.
8. We were told We reported Be fast. Be useful. A number.
9. We were told We reported Be fast. Be useful. A number. Doesn't get you budget
10. We were told We reported Be fast. Be useful. A number. Doesn't get you budget Gets you budget
11. Makes you findable AI crawlers don't run JS. 500M+ GPTBot fetches. Zero JS execution. GPTBot fetches JS 11.5% of the time. ClaudeBot 24%. Neither runs it. Either it's in the initial HTML, or it's invisible.
12. Makes you findable Makes you scoreable AI crawlers don't run JS. 500M+ GPTBot fetches. Zero JS execution. GPTBot fetches JS 11.5% of the time. ClaudeBot 24%. Neither runs it. llms.txt 300k domains, zero correlation. Either it's in the initial HTML, or it's invisible. Author bio boxes "E-E-A-T was never author bios" FAQPage schema Recommended in 2026, but deprecated in 2023.
13. Makes you findable Makes you scoreable AI crawlers don't run JS. 500M+ GPTBot fetches. Zero JS execution. GPTBot fetches JS 11.5% of the time. ClaudeBot 24%. Neither runs it. llms.txt 300k domains, zero correlation. Either it's in the initial HTML, or it's invisible. Author bio boxes "E-E-A-T was never author bios" FAQPage schema Recommended in 2026, but deprecated in 2023. AMP was the score version of speed. We didn't aim at faster, we aimed at ✅.
14. What we got wrong Slide 1 of 16 million Sorry, just 2 examples. Gary only gave me 7 mins.
15. What we got wrong Slide 1 of 16 million Sorry, just 2 examples. Gary only gave me 7 mins.
16. Schema 6M pages Cited pages had JSON-LD ~3x more often We sold it for a year. Ahrefs - 6M pages - 2026
17. Then, a re-test 1,885 pages added JSON-LD. 4k control pages. +2% ChatGPT +2% G AI Mode -4% G AI Overviews Ahrefs - Aug 2026 - Mar 2026
18. The markup was fine. The promise wasn't. Was only a sign of teams already doing everything else well. 0.04 Correlation between word count and citation probability - Ahrefs, 174k pages
19. Listicles +16.5 pp chance of being mentioned 1.17 positions earlier in the answer This is real. It works. 200k AI responses - 8 engines Peec AI, Sep 2025 - Mar 2026
20. It's not that engines trust them more
21. It's not that engines trust them more Grounding checks whether a claim exists. Not whether the source is honest. Dan Petrovic - Feb 2026 (Paraphrasing)
22. Google spent twenty years building defences against exactly this. LLM retrieval imported the corpus, not the defences.
23. Google spent twenty years building defences against exactly this. LLM retrieval imported the corpus, not the defences. The defences already exist, just not pointed at this yet.
24. We did this to links Still cleaning up that mess, huh?
25. Sorry, Matt 📸 not ai generated
26. The new shiny thing ✨
27. The engine writes its own questions. 10M+ prompts. 20M+ query fan-houts. Peec AI, Feb 2026
28. Even when your customer didn't ask in English 43% of research steps run on the English language web! Peec AI - 20M+ query fan-outs - Feb 2026 11% of 30M AI citations go to reddit.com, the single most-cited domain
30. 66% 77% 81% 82% 94% Spanish German French 88% 87% 88% Polish Portuguese Dutch Italian Turkish Nobody in this room was doing this in 2015. Including either of them.
31. 66% 77% 81% 82% 94% Spanish German French 88% 87% 88% Polish Portuguese Dutch Italian Turkish Nobody in this room was doing this in 2015 And very soon someone will be selling you a fan-out optimisation checklist
33. Our industry changed a lot overtime. Why would now be any different?
34. It's a probability distribution, not a ranking function. 92% of the time Gemini picks whichever page it read first 40-60% citation drift, month to month
35. You can't reverse-engineer a dice roll Only change the odds
36. "cosine drift" Not a metric. No such measurement exists. And cosine similarity (what it is imitating) can be measured, and still tell you nothing Steck, Ekanadham & Kallus - Netflix / Cornell - The Web Conference 2024
37. What correlates with showing up 75k brands Branded web mentions: 0.664 Backlinks: 0.218 Ahrefs - AI Overview brand visibility factors
38. Google says this is all still Just SEO
39. Google says this is all still Just SEO They are right
40. Google says this is all still Just SEO They are right Also, it doesn't matter, because Google is not who approves your headcount or budget
41. For both of them, GEO is just SEO For everyone else, what we were calling SEO was a subset
42. So say GEO
43. So say GEO When I say "SEO", my engineer hears hack
44. So say GEO When I say "SEO", my engineer hears When I say "SEO", my CFO hears hack the line going down
45. So say GEO When I say "SEO", my engineer hears When I say "SEO", my CFO hears When I say "SEO", my own brain hears hack the line going down what I've always done
46. A new word helps you create a clean slate
47. Somebody will invent the number. Nobody has it yet That's your window. Use it to become Her.
48. Somebody will invent the number. Nobody has it yet That's your window. Use it to become Her.
49. Thank you pojda.dev linkedin.com/in/pojda/ Ahrefs — We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved. (2026) · AI Overview Brand Visibility Factors (75,000 brands) · Short vs. Long Content in AI Overviews (174k pages, Dec 2025) Peec AI — The Listicle Rank Effect (200k responses, 8 engines) · ChatGPT searches in English, even when you don't (10M+ prompts, Feb 2026) Dan Petrovic / DEJAN — AI Spam (Feb 2026) · Rank Signal Recipe (92% position bias, Jul 2026) Mike King / iPullRank — The Case for Cloaking for LLMs (AI crawlers & JavaScript, May 2026) Vercel + MERJ (500M+ GPTBot fetches) SE Ranking (llms.txt, 300k domains) · Amplerank (30M citations) Steck, Ekanadham & Kallus — Is Cosine-Similarity of Embeddings Really About Similarity? (WWW '24)