ChatGPTs UTM tags are back ⚡
As The Attribution Company... we got a bit nervous for a second. Heres everything were seeing along with some other fun updates from the OpenAI side:
What happened:
--> Thu Oct 8, about 2pm PT: ChatGPT stopped adding utm_source= to its links, and the referrer went with it.
--> Oct 9 to 14: attributable ChatGPT traffic ran about 50% below baseline across the sites our attribution models cover, bottoming at -67% on the morning of Oct 14.
--> Oct 14: OpenAI confirmed the bug. Oct 15, about 2am PT: the fix started rolling out.
What this looks like to an attribution model:
1// Our models picked up an abnormality with the usual flags we have in place. We saw global decreases in ChatGPT referred traffic, increases in direct, no changes to AI visibility, no changes to branded organic, and no changes to referral traffic from other models...
2// ...Direct, branded organic, AI visibility, citations, and referred traffic are all complex pieces of the same puzzle. The way that they move, or dont move, in tandem tells us so much about the impact of AI revenue on a given brand. Most of the time, we see changes in real world data that arent actually reflective of a brands performance in AI search. Being able to combine this with data on how the models behave yields much better insights.
3// 🐣 Easter egg 🐣 : For those who pay very close attention to slide 3, youll notice that branded organic clicks dropped right when the referrer tags started coming back. This is unrelated to the bug, and actually caused by a slight increase in OpenAIs propensity to cite owned media over earned and social, leading to more website visitors arriving directly from chat as opposed to needing to arrive via Google.
Where this goes from here:
OpenAI has spent the year building ad measurement (ads since February, a pixel and a conversions API by spring, a self-serve Ads Manager in May, about 20 measurement partners on Oct 5), and Gemini started tagging its own links in September.
A platform that sells ads needs attribution that works, and we expect that pressure to also reach the organic side. Were quite excited to start getting more data from the model providers to further increase the confidence intervals that were able to report on internally.
S/o Arjun Kumar Theo Nsereko Eren Karaaslan for the quick work here.
Show more