Author: John Iwuozor
Profile: https://www.linkedin.com/in/john-iwuozor/
…aaannnnddd the full breakdown of my GPT-5.6 research is out!
Finished it earlier than planned and distilled the whole thing into 2,800+ words of pure goodness. If AI SEO is your jam, you probably need to read this
Beyond the findings I shared yesterday, a few more stood out:
- Fan-out isn’t just some enormous search tree. Once search triggered, GPT-5.6 generated an average of 4.29 queries. From the look of things, it’s getting focused on specific checks rather than burning through every possible branch.
- Brand-owned and retail sources dominated the citation mix. They made up 59.3% of sources, which makes sense when 76.7% of searches are using the site: operator to engage brand websites directly for product info.
- The average answer was 433 words and 3,161 characters. That is remarkably close to the 3,043-character average Profound found in its parrot problem study. ChatGPT rarely gives you only the answer; it keeps layering in context, caveats and related information.
- Citation volume was heavily concentrated. Out of 955 cited domains, the top 10 accounted for 30.5% of all citation appearances, while the top 50 accounted for 52.5%. It already echoes Ayomide's finding that AI recommendations tend to converge around a tight, repeating pool rather than continually discovering new options
Just go read the piece for more details. B2B and B2C brands trying to understand how to appear more in ChatGPT or figure out which prompts are worth tracking may find this useful.
Lots of inspo from the folks doing proper work in this space: Chris, Tomek, Suganthan, and Nick
Article in the comments 👇
Finished it earlier than planned and distilled the whole thing into 2,800+ words of pure goodness. If AI SEO is your jam, you probably need to read this
Beyond the findings I shared yesterday, a few more stood out:
- Fan-out isn’t just some enormous search tree. Once search triggered, GPT-5.6 generated an average of 4.29 queries. From the look of things, it’s getting focused on specific checks rather than burning through every possible branch.
- Brand-owned and retail sources dominated the citation mix. They made up 59.3% of sources, which makes sense when 76.7% of searches are using the site: operator to engage brand websites directly for product info.
- The average answer was 433 words and 3,161 characters. That is remarkably close to the 3,043-character average Profound found in its parrot problem study. ChatGPT rarely gives you only the answer; it keeps layering in context, caveats and related information.
- Citation volume was heavily concentrated. Out of 955 cited domains, the top 10 accounted for 30.5% of all citation appearances, while the top 50 accounted for 52.5%. It already echoes Ayomide's finding that AI recommendations tend to converge around a tight, repeating pool rather than continually discovering new options
Just go read the piece for more details. B2B and B2C brands trying to understand how to appear more in ChatGPT or figure out which prompts are worth tracking may find this useful.
Lots of inspo from the folks doing proper work in this space: Chris, Tomek, Suganthan, and Nick
Article in the comments 👇
Images:
