I asked my AI coding partner to run 6 real-world SEO & AI Search questions through my live Open Knowledge Format (OKF) Brain across three Gemini models, shuffle columns A, B, and C, and hide the model names so I could vote completely blind. Try voting yourself below, then click See How Marie Voted + Token Costs!
Every run used the exact same OKF Brain index, Concise default prompt (~180-280 words), and live tool calls. Token counts come directly from real Gemini API usageMetadata (including thinking tokens).
| Model | Marie's Blind Score | Your Blind Picks | Avg Speed | Avg Length | Avg OKF Tools | Avg Tokens (In / Out / Think) | Est. Cost / Turn | Est. Cost / 1,000 Turns |
|---|
*Note on Neon speed: The 70.8s average was skewed by a single 3.5-minute API queue spike during testing. Typical Gemini 4 Neon replies ran 18-40s.
At ~$191 vs. ~$93 per 1,000 prompts (~$0.098 extra per prompt in Concise mode), here is what the monthly cost looks like across realistic full-month usage paces (extrapolated from early active users who have had their Brain for 1 to 2 weeks so far):
“Our informational gardening and recipe blog dropped 45% in Google Search Console clicks over the last 6 months, while impressions actually stayed flat. What is most likely happening, and what are the first 3 things we should check?”
When impressions stay flat while clicks drop 45%, your pages are still showing up in Google Search, but searchers are getting their answers right on the results page without needing to click through. For informational gardening and recipe blogs, this is almost always driven by summarizing the basic steps (like how to prune tomatoes or blanch asparagus) directly at the top of the page.
Here are the first 3 things I would check right away:
Do you want to pick one of your hardest-hit posts and brainstorm how to add a stronger reason to click?
When impressions stay flat while clicks plummet, your site is still appearing in search results, but searchers no longer need to click through to get their answer. In gardening and recipe niches, this almost always points to zero-click displacement from and Google directly answering informational queries at the top of the page.
Here is what is happening and the first three things to check:
What are two or three of the specific queries or pages that saw the biggest click drops? We can look at what Google is currently showing on those SERPs and brainstorm ways to add unique, non-commodity value.
A 45% drop in clicks while impressions stay flat is the classic signature of absorbing basic informational searches right on the results page. For gardening and recipe blogs especially, Google's AI can easily summarize general "world knowledge" (like a standard recipe or when to prune tomatoes), meaning your pages are still showing up and even grounding those answers, but searchers no longer feel the need to click through.
Here are the first 3 things I recommend checking right away:
If you'd like, share one or two of your hardest-hit post topics and we can brainstorm how to add unique value that gets people clicking again.
“Google AI Overviews are now answering our basic 'how to' and 'what is' questions directly in the SERP. Should we delete those informational articles or rewrite them? What does your research say actually works now?”
Do not rush to delete those articles just because AI Overviews are answering them, but do not leave them as basic, surface-level summaries either. My research shows that simply pruning pages rarely improves how Google views your overall site quality - as John Mueller noted, making a site smaller does not automatically make it better.
What actually works now is deciding which pages deserve a real upgrade from into something an AI summary cannot easily replace:
Concrete next step: Pull your top 5 declining informational posts and ask yourself what a real customer still gets stuck on after reading the basic AI answer. Add that lived experience right at the top of the page.
Want to pick one of your "how to" articles and brainstorm how to turn it into a non-commodity piece together?
Do not mass-delete those articles, and do not leave them as surface-level summaries. Rewriting the ones that matter to your business with real, first-hand depth is what works now, whereas pruning alone rarely fixes site-level quality.
Here is what my research shows is actually happening and what to do next:
Do you have a specific "how to" or "what is" article in mind? If you share the topic, we can brainstorm how to turn it into something an AI summary cannot replace.
Neither wholesale deleting nor generic rewriting will solve this on its own.
Googleโs AI answers excel at summarizing consensus facts, meaning basic definitional guides are treated as . Simply deleting these pages (the "pruning" playbook) will not magically boost your overall domain score; as Google's John Mueller pointed out, merely shrinking a site doesn't convince ranking systems that what remains is high quality.
Instead, the sites winning citations and clicks are pivoting from textbook definitions to authentic, lived experience:
A great practical exercise is the to spot where your content currently overlaps with commodity text.
Would you like to look at one of your specific "how-to" articles together and brainstorm how to inject unique information gain?
“We run a product comparison site with long 4,000-word 'Best X' buyer's guides. Our rankings have been sliding even though our content is thorough and well-researched. How should we rethink the top two scrolls of these pages?”
The traditional 4,000-word buyer's guide format is struggling because searchers do not want to scroll through an essay before finding a recommendation. If your top two mobile scrolls are filled with generic intros, "Why you should trust us" paragraphs, or table-of-contents widgets, visitors bounce back to the search results to find a direct answer.
Here is how I recommend restructuring the top of those comparison guides:
Would you like to take one of your declining comparison URLs and sketch out an above-the-fold layout test together?
Even the most thorough 4,000-word guide will slide in rankings if a visitor has to wade through generic intros, long disclosures, or "how to choose" filler just to see your actual recommendations. Google's updated helpful content guidance places heavy emphasis on - meaning you need to satisfy the shopper's core question immediately, before they get frustrated and bounce back to the search results.
Here is how I would rethink the top two scrolls of your buyer's guides:
As a concrete next step, pick two of your top-sliding guides, move the winning picks above the fold, and watch your engagement and rankings over the next few weeks. Want to share the top of one of your current guides so you and I can brainstorm what to trim?
When rankings slide on thorough 4,000-word buyer's guides, it is often because the actual answer (your top picks and comparison) is buried beneath long introductions, methodology blocks, or ad units. Google's systems heavily reward immediate user satisfaction, meaning visitors should get what they came for in the first two scrolls without having to hunt for it.
Here is how I would rethink the top of your guides:
Want to share the layout of one of your top guides so we can sketch out a faster, higher-converting top section together?
“We have 1,200 thin location pages in our sitemap that get almost zero clicks. Should we noindex them, delete and 410 them, or redirect them to our main service pages? Walk me through how to decide safely.”
Do not redirect 1,200 thin pages to a generic main service page (Google considers blanket redirects soft 404s), and avoid leaving them on noindex forever. For zero-click, boilerplate location pages that hold no backlinks or user demand, deleting and returning a 410 Gone (or 404) is the cleanest, safest route to strip away low-value drag.
A solid next step is running a quick URL export to bucket these into: (1) pages with backlink equity to redirect, (2) true hubs to enrich, and (3) zero-value URLs to 410.
Would you like me to help you sketch out an audit checklist from our to sort them quickly?
If those 1,200 location pages are thin, templated, and getting almost zero clicks, I would delete and 410 the vast majority of them rather than redirecting or noindexing them. Redirecting 1,200 dead-weight location pages to a handful of main service pages can look like a soft 404 to Google, and noindexing still wastes crawl resources without fixing the underlying issue.
Here is how to decide safely before you touch a single URL:
Do any of these 1,200 pages have real backlinks or local conversions, or are they all pure template pages?
For 1,200 thin location pages getting almost zero clicks, I would not mass-redirect them to your main service pages (Google often ignores mass redirects to generic pages and treats them like soft 404s). Instead, segment them by real business value first, then cleanly delete and 410 the true dead weight while removing those URLs from your sitemap.
Here is how I decide safely between the three options:
noindex if the page serves a real purpose for direct visitors or paid ad campaigns, even though it should not be in Google Search.Your safest next step: Export all 1,200 URLs into a spreadsheet with 12 months of Search Console clicks, conversions, and backlinks (like I outline in ). Start by 410-ing a test batch of 100 to 200 zero-value pages and monitor your core service pages for a few weeks before scaling up.
Are any of these location pages tied to physical offices or unique local case studies, or are they all templated?
“My boss just asked me if we should add an /llms.txt file and optimize for AI agents shopping on our site. Is /llms.txt actually worth doing right now, or is it hype?”
Here is the honest breakdown: /llms.txt is hype if your boss thinks it will boost Google Search rankings, because Google Search completely ignores it. However, it is worth considering for AI agents actively browsing and shopping on your site, where it acts like a simple instruction manual so agents do not have to guess how to navigate your pages.
Here is what matters right now and what I recommend doing:
llms.txt and UCP support to handle agentic checkout safely./llms.txt file (following the simple Markdown format at llmstxt.org), keep it short and focused on your core store navigation rather than dumping every URL on your site into it.What e-commerce platform are you running on? If you tell me your setup, I can help you decide whether a quick /llms.txt or a UCP check is the smarter first step.
It is not just hype, but it is also not an SEO ranking tactic. Google Search completely ignores /llms.txt for rankings and AI Overviews, so adding one will not boost your search traffic - however, it is becoming a practical instruction manual for AI shopping agents that actively browse your site.
Here is how I would frame this for your boss:
llms.txt as a standard way to point agents to product data and checkout flows. If you are on a custom site, Chrome Lighthouse now includes an audit for llms.txt under its "Agentic Navigation" checks.Are you running on Shopify or a custom e-commerce setup? If you'd like, we can sketch out a simple, 10-line /llms.txt file tailored to your store's main shopping flows.
The direct answer is that is not an SEO silver bullet, but it is a legitimate tool for autonomous browsing agents. If you expect it to boost your organic rankings or AI Overview visibility, that is pure hype; if you want to make life easier for AI agents navigating your catalog, it can be useful.
Here is the strategic reality and how to frame it for your boss:
llms.txt when crawling, indexing, and ranking content. AI engines fetch answers from their core web index and knowledge graphs, not from this text file.llms.txt is to help autonomous browser agents (like those tested in Chrome Lighthouse) quickly locate high-priority resources (such as product feeds or search paths) without sifting through messy HTML.llms.txt alongside frameworks like and . However, preparing for right now comes down to having impeccable Merchant Center feeds, accurate inventory data, clear pricing schema, and frictionless guest checkouts.If your technical team can spin up a clean llms.txt file in an afternoon following the , it does not hurt to publish one. Just make sure your team focuses its main energy on product feed fidelity and structured data.
Would you like me to look at what e-commerce platform you are running to see what structured feeds would make the biggest immediate impact?
“I'm a solo service-business owner with only 5 hours a week for SEO. Give me a realistic 30-day plan to make my site more visible in both Google Search and ChatGPT without wasting time on outdated tactics.”
With only 5 hours a week, the best move is to focus on Google Search first and ignore standalone "ChatGPT SEO" gimmicks. In my , Google Organic drove 96.5% of conversions compared to less than 1% of traffic from ChatGPT - and the exact same authority that wins in Google is what gets you cited by AI tools.
Here is a realistic 30-day plan (5 hours per week) that skips outdated tactics like cold link building and generic blog posts:
What type of service business do you run? If you share your niche, I can help you brainstorm that Week 3 content piece right now.
You do not need separate strategies for Google Search and ChatGPT. Both reward the exact same things: a clear business identity, real-world trust signals, and content rooted in your actual experience that AI cannot easily summarize away.
With 5 hours a week, here is a realistic 30-day plan based on my and :
What kind of service business do you run? If you share your niche, I can help you brainstorm your Week 3 content topic right now.
Here is the grounded reality: you do not need 20 hours a week, and you do not need separate playbooks for Google and ChatGPT. Both systems surface sources that demonstrate real-world authority, clear entity identity, and non-commodity insight.
With only 5 hours a week (20 hours total), the goal is to cut out low-return busywork (like churning out generic blog posts or chasing manufactured backlinks) and focus on high-impact signals.
Target: Help AI models understand exactly who you are, what service you offer, and where you operate.
Target: Move your pages from commodity summaries to experience-backed resources.
Target: Influence LLM knowledge bases and search algorithms where they actually look.
Target: Capture the specific questions clients ask before hiring you.
What type of service business do you run, and what do your primary service pages look like right now? We can look at one together and pinpoint the fastest quick-win adjustments.
Right now, 30 days of complimentary early access to Marie's Brain (pre-loaded with your site profile and full diagnostic report) is included with every $500 Diagnostic Report. Or join the public waitlist to be notified when standalone signups open.