What I Actually Learned About SEO, AI, and Running a Business Last Year – 2025

The top 19 Things I learned about SEO, AI, and Running a 10-Year-Old Business Last Year

Every week, I do a check-in and a checkout.

It started as a way to make sure I wasn’t forgetting to balance work and life and to feel like I was making progress. Over time, it became something more useful: a running log of what I actually learned—not what I skimmed, not what I bookmarked, but what genuinely changed how I think or work.

Looking back at last year’s checkouts, a pattern emerges. Some of the lessons are deeply technical. Some have nothing to do with SEO and everything to do with staying sane while running a business.

Here’s the honest version.

The SEO + AI Things That Actually Mattered

I spent a lot of time last year separating signal from noise. AI makes that more complicated, not easier.

1. Ranking in AI Answers and LLMs

I learned what actually goes into ranking—or appearing—in LLMs and AI Mode. Not theories. Not vibes. The real inputs, the constraints, and the parts that are still opaque. 

The big takeaway: most AI answers rely on retrieval-augmented generation, which is SEO. The layers added are commentary on your brand and topic in authoritative online spaces for your niche, and an understanding of how to optimize for vector databases. Check out my post and checklist on the subject.

2. Decrease in Organic Clicks

I also conducted a deep dive and presented all the factors contributing to fewer organic clicks. It’s not just Google’s changes; there’s also a shift in searcher behavior, AND analytics tools/ data retention policies removing organic or relabeling clicks. This presentation will give you a good summary of the search apocalypse that happened in 2025 and will continue to worsen in 2026.

3. Drupal’s AI Features

I did a deep dive into which AI features Drupal currently supports and how to use them to optimize your website copy.

4. Structured Data is not used for LLMs

I confirmed something that keeps getting misunderstood: structured data is not being used for LLM search optimization. That doesn’t make structured data useless—it just means people are assigning it the wrong job. Schema still helps machines understand pages and how your brand is related to your other profiles and online mentions. It can still help with disambiguation. It’s just not a cheat code for AI visibility. Just like LLM.txt, it’s not a shortcut and isn’t even used by Google.

5. Site launches without SEO are always unpredictable.

I learned that developers can, in fact, roll out a website where only the homepage returns a 404. Yes, really. No one noticed. This is why technical QA still matters. On the upside, if this happens when you’re not yet on board as an SEO consultant, it means there’s nowhere to go but up in traffic. The takeaway? Migrate your sites with SEOs on board if at all possible.

6. I developed a tool to map redirect tests to the original mapping for enterprise sites.

I figured out how to check redirects and map the results to the redirect mapping sheet using Python in Google Colab to run redirect tests at scale, saving hours and catching things a crawler would miss. This is a solution I’ve been looking for for years- and I created a copy for you to use that you can play with. If you’ve never used Colab, no worries – LLMs can walk you through the steps. 

7. I built a few custom GPTs.

I learned how to build custom GPTs and built two. Not demos—tools I actually use. 

• SEO Tech Ticket Generator and 
• SEO Jargon to Plain English. 

8. So many new bots

Google’s bots

• Google-Extended: Used for training Gemini and grounding AI Overviews.
• Gemini-Deep-Research:
  • User-Agent String: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; Gemini-Deep-Research; +https://gemini[dot]google/overview/deep-research/) Chrome/135.0.0.0 Safari/537.36.
• Google-NotebookLM: A user-triggered agent that fetches content for individual users and typically ignores robots.txt by default.
• Google-Read-Aloud:
  • User-Agent Name: Google-Read-Aloud.
  • Function: Used by Chrome on Android, the Google App, and other text-to-speech services.
• Google Site Verifier:
  • Gemini iOS Agent
    • User-Agent Name: GeminiiOS 1.2025.417.
    • Function: Runs when a visitor clicks a link within the Gemini iOS app.
  • IP Addresses: Google maintains structured IP ranges for its special crawlers, which can be verified via their official documentation.
    • The PageSpeed Insights bot (User-agent: Google Page Speed Insights) 
    • The Rich Results Test uses the user agent: Google-InspectionTool. 

OpenAI & ChatGPT Agents

OpenAI has expanded its bot ecosystem to support both general training and its new agentic browser tools.

• ChatGPT-User:
  • Function: Used for direct user actions and Custom GPTs; it is not used for automatic web crawling or training.
  • User-Agent String: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot.
  • IP Endpoint: https://openai.com/chatgpt-user.json.
• ChatGPT Atlas (Agentic Browser):
  • User-Agent String: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0.0.0 Safari/537.36.
• GPTBot: Used specifically for general AI training data collection.
• OAI-SearchBot: Used specifically for AI search functionality.

Other Search and Social AI Agents

Various other platforms have introduced or updated their bots for search discovery and web scraping in 2025.

• DuckDuckGo Agents:
  • Names: DuckDuckBot and DuckAssistBot.
  • IP Endpoints: https://duckduckgo.com/duckduckbot.json and https://duckduckgo.com/duckassistbot.json.
• Common Crawl:
  • Name: CCBot.
  • IP Endpoint: https://index.commoncrawl.org/ccbot.json.
• Perplexity:
  • Name: PerplexityBot.
  • Note: Perplexity has been observed modifying its user agent or using residential IP addresses as proxies to circumvent blocks.

9. Google grounds its AI in its algorithms.

I learned that Google is training its own LLMs on its search algorithms. That alone should make everyone slow down before declaring SEO “dead” or “unchanged.”

10. Social media uses vector embeddings.

I learned that social media platforms also use vector embeddings, just like search engines. This means your social media posts should be measured for cosine similarity.

11. Substack does podcasts?

Yes, this was new to me, and now I’m currently pondering how to engage with my audience on Substack.

12. Find your customer’s journey.

This new tool, Listening Mind, is fascinating, and I am waiting to see if they will drop their monthly pricing.

13. ChatGPT’s Entity Understanding

I confirmed that ChatGPT understands entities far better than most people give it credit for. This has implications for how we structure content, not just how we write prompts.

14. VidIQ’s AI tools rock

Everyone should use VidIQ’s AI tools. It helps you craft your message in the context of your competition – doing the research in minutes.

15. Various Google Deep Research uses

• Assist in summarizing and drafting my monthly newsletter—not to replace thinking, but to accelerate synthesis. 
• Generate market opportunities for clients based on their personas, mission, and their competitors’ online activities.
• Promotional plans for clients based on goals and their market context.

The Business Lessons That Were Less Technical, but important for balance

16. When work spikes, I learned to use our babysitter to handle transportation to in-person homeschooling classes. It’s not an indulgence — it’s capacity planning.
17. I created Gmail email groups and immediately wondered why I hadn’t done it sooner. I am now no longer emailing the wrong “Brandon”. *sigh. Really. I should have discovered this sooner.
18. I learned that you can negotiate with lawyers more often than people assume. Rates are not always fixed. 
19. And I re-learned something I already knew, but apparently needed to keep learning: always send contracts to lawyers. Always. They find things you won’t, no matter how well you think you read the contract.

The Throughline

The common thread in all of this isn’t AI. It’s realism.

Systems are getting more complex, and AI tools are being launched constantly. 

My weekly checkouts force honesty. They surface what actually changed how you operate.

That’s what I want more of this year: time-saving tactics that keep me focused on client wins.  

SEO work isn’t getting simpler—but it is getting more interesting.