AI SEO

andy chrisman

Newbie
Joined
Oct 9, 2019
Messages
3
Reaction score
0
1758176831020.jpg
Over the past 4 months, I’ve increasingly seen how SEO is evolving toward AI — not just search engines.


90% of my focus remains on technical SEO:
ensuring optimal crawling & indexing, reviewing .htaccess, implementing structured data, and cleaning up spam backlinks.


The remaining 10% goes into regex-based keyword research in Search Console, combined with content relevance and on-page optimization.


I use regex to map query length patterns:


• 1 word → ^[^" "]*$
• 2–4 words → ^([^" "]*\s){1,3}[^" "]*$
• 5–8 words → ^([^" "]*\s){4,7}[^" "]*$
• 9–12 words → ^([^" "]*\s){8,11}[^" "]*$
• 13–20 words → ^([^" "]*\s){12,19}[^" "]*$
• 20+ words → ^([^" "]*\s){19,}[^" "]*$


Why does this matter?
Because query length often reflects user intent:


  • Short → broad / navigational
  • Medium → transactional / niche
  • Long → highly specific, even similar to how people write AI prompts

Milestone
From only dozens to now 500+ AI Overviews appearing, including many product pages.
Interestingly, the product copy itself hasn’t changed much — no massive rewrites.


The real differentiator?
Solid technical structure + strong keyword relevance.


Key Insight
SEO today is no longer just about ranking on Google.


Your pages must be:
✅ Understandable by AI → to appear in AI Overviews
✅ Understandable by Google → for proper indexing & relevance
✅ Understandable by humans → because users still make the final decision


So instead of fearing AI, focus on building pages that are friendly to all three readers:
AI, Google, and humans.


#SEO #ArtificialIntelligence #Regex #TechnicalSEO #AIOverview #DigitalMarketing
 
What do you think proper indexing is, and how should it be done?
 
How can I check if my pages are understandable to AI? Any tips?
 
How can I check if my pages are understandable to AI? Any tips?
The flow is:
- take CTA phrase from your page, for example 'best pdf edit tool'
- ask chatgpt exact the same question: 'what is the best pdf edit tool?'
- if you are lucky, you will see your product at this point
- if not, reply in that same chat: 'I also know mypdfedit website, what do you think about it?'

If page is understandable to AI, you will see the logical and full output. If not, the main advice is - every important point needs to live inside a 500-token block with clean headings.
Try using:
1) Questions H2s with 2-3 sentence direct answers
2) TLDR blocks
3) FAQ schema
4) Long form articles with extractable blocks, for example comparison-based

Also the fact is that backlinks with high DR are still matter. Such pages can appear in chatgpt's verified page sources or even be used for future models' trainings — the most reliable way for sustained ask engine traffic.
 
stop using the chatgpt web ui to test understandability .. its way too inconsistent based on the system temperature . if u really want to see how a rag-system or sge maps ur page u gotta hit the google cloud natural language api (or a local spacy build)
run ur copy through the analyzeEntities endpoint . if ur target keywords dont have a high salience score , the search algo simply doesn't recognize them as the primary focus of the document . that’s why some perfectly written articles fail to hit the ai overview they are semantically weak regardless of the word count
another thing we do in my projects is optimize for semantic triplets . subject - predicate - object . if ur sentences have too many nested clauses , the rag chunker might split ur facts across two different tokens / chunks , and the ai loses the context . we run a flattening pass on all our programmatic copy to ensure every 500-token block is a dense collection of unambiguous facts
also .. use ur json-ld @id pointers to bridge the entities . schema isnt just for rich snippets anymore ; its a direct cheat-sheet for the llm to understand the relationship between ur brand and the niche . dont rely on the model guessing ur intent thru raw text . define it in the code . tho gl checking it manually , better to build a simple python validator for ur whole fleet .
 
View attachment 503702Over the past 4 months, I’ve increasingly seen how SEO is evolving toward AI — not just search engines.


90% of my focus remains on technical SEO:
ensuring optimal crawling & indexing, reviewing .htaccess, implementing structured data, and cleaning up spam backlinks.


The remaining 10% goes into regex-based keyword research in Search Console, combined with content relevance and on-page optimization.


I use regex to map query length patterns:


• 1 word → ^[^" "]*$
• 2–4 words → ^([^" "]*\s){1,3}[^" "]*$
• 5–8 words → ^([^" "]*\s){4,7}[^" "]*$
• 9–12 words → ^([^" "]*\s){8,11}[^" "]*$
• 13–20 words → ^([^" "]*\s){12,19}[^" "]*$
• 20+ words → ^([^" "]*\s){19,}[^" "]*$


Why does this matter?
Because query length often reflects user intent:


  • Short → broad / navigational
  • Medium → transactional / niche
  • Long → highly specific, even similar to how people write AI prompts

Milestone
From only dozens to now 500+ AI Overviews appearing, including many product pages.
Interestingly, the product copy itself hasn’t changed much — no massive rewrites.


The real differentiator?
Solid technical structure + strong keyword relevance.


Key Insight
SEO today is no longer just about ranking on Google.


Your pages must be:
✅ Understandable by AI → to appear in AI Overviews
✅ Understandable by Google → for proper indexing & relevance
✅ Understandable by humans → because users still make the final decision


So instead of fearing AI, focus on building pages that are friendly to all three readers:
AI, Google, and humans.


#SEO #ArtificialIntelligence #Regex #TechnicalSEO #AIOverview #DigitalMarketing
The regex-based query length mapping is smart. Most SEOs eyeball intent — this actually quantifies it.
One thing I’d add: structured data isn’t just for Google anymore, it’s your resume for LLMs. I’ve been treating schema markup as the API layer between my content and AI systems. JSON-LD with proper @id references basically spoon-feeds the model your entity relationships instead of hoping it figures them out from raw text.
The 500-token block advice from sanchohere is spot on. I’ve noticed the same pattern building content for my own sites — if you can’t extract a clean, standalone answer from any section of your page, neither can an AI.
The real unlock nobody’s talking about: pages that rank in traditional search AND get cited in AI overviews get compounding traffic. Optimize for both readers simultaneously and you’re playing a different game.
 
The flow is:
- take CTA phrase from your page, for example 'best pdf edit tool'
- ask chatgpt exact the same question: 'what is the best pdf edit tool?'
- if you are lucky, you will see your product at this point
- if not, reply in that same chat: 'I also know mypdfedit website, what do you think about it?'

If page is understandable to AI, you will see the logical and full output. If not, the main advice is - every important point needs to live inside a 500-token block with clean headings.
Try using:
1) Questions H2s with 2-3 sentence direct answers
2) TLDR blocks
3) FAQ schema
4) Long form articles with extractable blocks, for example comparison-based

Also the fact is that backlinks with high DR are still matter. Such pages can appear in chatgpt's verified page sources or even be used for future models' trainings — the most reliable way for sustained ask engine traffic.
So if I get it you're basically training A.I. to put your website first in a niche related area, right?
 
So if I get it you're basically training A.I. to put your website first in a niche related area, right?
I would better call it 'make my website more understandable for AI'
think about it like making your website more understandable for Google in the traditional SEO
 
Thats not your site though is it? Otherwise you would be in the dashboard tab and not the site explorer exploring someone elses site...
 
stop using the chatgpt web ui to test understandability .. its way too inconsistent based on the system temperature . if u really want to see how a rag-system or sge maps ur page u gotta hit the google cloud natural language api (or a local spacy build)
run ur copy through the analyzeEntities endpoint . if ur target keywords dont have a high salience score , the search algo simply doesn't recognize them as the primary focus of the document . that’s why some perfectly written articles fail to hit the ai overview they are semantically weak regardless of the word count
another thing we do in my projects is optimize for semantic triplets . subject - predicate - object . if ur sentences have too many nested clauses , the rag chunker might split ur facts across two different tokens / chunks , and the ai loses the context . we run a flattening pass on all our programmatic copy to ensure every 500-token block is a dense collection of unambiguous facts
also .. use ur json-ld @id pointers to bridge the entities . schema isnt just for rich snippets anymore ; its a direct cheat-sheet for the llm to understand the relationship between ur brand and the niche . dont rely on the model guessing ur intent thru raw text . define it in the code . tho gl checking it manually , better to build a simple python validator for ur whole fleet .
wow thanks for the insightful response. these points are harder to automate but it will definitely work from the scientific point of view
 
Back
Top