Hi everyone,
I’ve seen Codex being used for both SEO research and GEO research, but I’m curious about the tracking side.
SEO has clearer metrics through tools like Google Search Console: rankings, impressions, clicks, traffic, and conversions.
GEO seems harder to measure because AI answers may change based on the platform, prompt, location, user context, and time.
For anyone using Codex for this:
- What data sources do you connect it to?
- How do you track brand mentions and citations in AI answers?
- Do you test a fixed list of prompts regularly?
- How do you compare visibility across ChatGPT, Gemini, Perplexity, and other AI platforms?
- Do you track only mentions, or also referral traffic and conversions?
- How do you know whether an improvement came from GEO work rather than normal SEO?
The main thing I want to understand is how you make GEO tracking consistent and reliable enough to support business decisions.
If you have built a working process, I’d appreciate it if you could share your setup, the metrics you track, and any real cases or results.
I treat GEO as a repeatable sampling exercise, not an exact ranking report.
-
Keep a fixed set of 30 to 50 prompts split by discovery, comparison, and purchase intent. Run them weekly on ChatGPT, Gemini, Perplexity, and Google AI Overviews using the same location and logged-out setup where possible. Save the full response, date, model, brand position, citation URL, and competitors mentioned.
-
Connect Codex to Search Console, GA4, Shopify orders, and the saved prompt results. Track mention rate, citation rate, share of mentions versus competitors, AI referral sessions, add-to-cart rate, and orders. Use UTMs on any URLs you control, but also group referrers such as chatgpt.com and perplexity.ai in GA4.
-
For attribution, annotate every content or schema change and compare the affected prompt group against an unchanged group for 4 to 6 weeks. SEO and GEO overlap, so I would not claim causation from mentions alone.
I’m also working on ClickFrom.ai, which automates the structured product data and AI citation tracking part, but the spreadsheet process above is enough to start today.
@Nekocha ,
I like the idea of treating GEO as a sampling exercise rather than a fixed ranking report. One thing add is to track answer accuracy not just whether a brand is mentioned.
For example an AI might mention your store but give the wrong price, product details, availability or shipping information. I’d record those errors alongside mentions and citations.
Over time that gives you a better picture of whether your GEO work is actually improving visibility and how accurately AI systems represent the brand.
You can also use the “seo-audit” skill - which is primarily designed for Claude - with Codex. I’ve customized it so that it can crawl the site at any time using the crawling data from the backend. It provides good insights, but it’s no substitute for a tried-and-true tool.
My advice is always to aim for diversity. If you rely solely on one tool - or, in this case, just one skill or prompt - that’s the wrong approach. Tools like SEMRush, Seobility, and others have years of experience that can’t be captured in a single prompt. They approach topics differently, have connections with Google and others (which are essential for insights into their tools), and tend to be more up-to-date.
You can supplement the technical SEO in the theme with Claude, Codex, Kimi, and other AI agents. As a supplement, it’s great
Hey Nekocha, @Nekocha
There’s no equivalent of Search Console for GEO yet, so most of this has to be built manually. A fixed panel of 30-50 prompts run weekly across ChatGPT, Gemini, and Perplexity works well, log whether you’re mentioned, in what position, and whether the info is accurate (wrong pricing or outdated details can hurt more than no mention). Don’t try to cleanly separate GEO from SEO credit, they overlap too much, just watch directional traffic change over 4-6 week windows.
The part worth emphasizing: mentions alone don’t tell you if it’s driving revenue. Check Shopify’s Analytics > Sessions by referrer first to see if AI platforms show up at all. If you want to tie that traffic to actual orders and revenue, Mipler reports let you filter by referring domain or UTM source to see conversion volume over time, more useful than citation counts alone.
Seeing any AI referral traffic yet, or still setting up tracking?
Thanks for sharing this approach. Could you share a few real examples from the 30–50 prompts you created, perhaps 2–3 prompts each for discovery, comparison, and purchase intent? I’d be interested to see how you structure them for different search intents.
Thank you for sharing this perspective. Tracking answer accuracy alongside mentions and citations makes a lot of sense, especially when incorrect product, pricing, or shipping information could be more harmful than no mention at all.
Have you tested this approach in practice and seen successful results? If so, do you have any tips, tricks, or workflows for tracking these errors and improving answer accuracy over time?
I test many of these approaches every day. I’m happy to share the following initial insights, since this isn’t a secret.
I can recommend the “Claude-seo” skill. It’s publicly available and already provides good coverage of what a Shopify store needs. However, the search speed must be throttled during the scanning process; otherwise, a 429 error will occur. I’ve adjusted this locally on my end to avoid any conflicts.
The skill covers a very broad range of areas. Technology. SEO. AI. If you’re familiar with it, it also offers direct fixes in the code. That’s more extensive, though, which is why I’ve automated it on my end. It’s a solid foundation, but the skill has one major limitation: it doesn’t have access to keywords, visibility, or other data. DataForSEO can be integrated, but it doesn’t have historical data. Automatic visibility comparisons? None. Tracking? It’s feasible with AI, but it’s time-consuming. Alerts that trigger automatically? Not feasible with AI either.
AI is great for one-off audits and fixes. The skills aren’t bad either. It only becomes problematic when we move on to automation, historical data, and tracking. For that, I always recommend external tools like Semrush, etc., which have years of experience.
In any case, these days you have to make the most of whatever comes your way. Those who work exclusively with AI are no better off than those who work exclusively with tools. The hybrid approach combining AI and tried-and-true tools is what makes the difference. Today, I manage SEOBility in the German market using Claude, Codex, GLM, and others - depending on the use case. They can analyze the data faster and more effectively than I ever could. With GSC integrated, it’s even more powerful.
The question shouldn’t be how to track SEO and GEO, but how to integrate all these technologies with one another.