Sacred-AI
← Back to News
ChatGPT Traffic Didn't Drop – Our Analytics Did

ChatGPT Traffic Didn't Drop – Our Analytics Did

October 20, 2026·By Admin User
#weekly-news#research

ChatGPT Traffic Didn't Drop – Our Analytics Did

Oct 19, 2026 · @Mads Barclay Paaskesen

Last week I published my Part 2 brand test: six of nine AI engines now know my app Flixlar by name. Then, a few days later, half the SEO world started panicking that ChatGPT traffic had fallen off a cliff.

So which is it? Are AI engines sending us more visitors, or fewer? The honest answer: for about a week, a lot of us simply couldn't see them.

📉 What Happened

Starting around October 9, sites saw ChatGPT-attributed sessions in Google Analytics 4 fall sharply. Graphite CEO Ethan Smith reported drops of 22% to 57% across the sites his company tracks, and LLM Pulse measured ChatGPT sessions about 46% lower on Oct 9–11 than on the same days in the previous two weeks. Total traffic, meanwhile, stayed roughly flat.

That last detail is the giveaway. Visitors didn't vanish. The labels did.

The explanation Smith and others landed on: ChatGPT stopped adding the utm_source=chatgpt.com tag (and a referrer) to links in its answers. Without either, GA4 has nothing to go on, so the visit lands in "Direct."

⚠️ According to Adobe's Cedric Huesler, OpenAI confirmed the bug on October 14, and a fix started rolling out early on October 15, with tagged clicks back to roughly 95–100% of the week before on the domains he tracks. As of Search Engine Journal's report, OpenAI hadn't published anything official, and its release notes for Oct 1–15 didn't mention link tracking.

🕵️ Where the Clicks Actually Went

The most useful data point came from Huesler. Comparing Oct 9–13 with the week before, on about 500 pages where ChatGPT drove at least 5% of visits, ChatGPT visits fell by about 31,000 a day, while Direct visits rose by about 26,000 a day. Not a perfect match, and it's only four days of sampled data, but close enough to tell the story.

Two other things changed at the same time, and they matter more long term:

  • Answers are changing shape. Profound found the share of cited answers with at least one tagged source link fell from about 98% on Oct 8 to 66% by Oct 14, while answers using OpenAI's new "Intelligent UI" grew from 2% to about 35%.
  • More bare URLs, mostly homepages. Peec AI's Tomek Rudzki found direct URLs in 42% of responses, up from about 8–10%, and 54% of those pointed to homepages. (Those are displayed links, not confirmed clicks.)

A fair caveat: most of these numbers come from companies that sell AI-search tooling. I still think they're worth reading, just with a pinch of salt.

[Mads: if you have it, add one line here on what Flixlar / sacred-ai.com saw in GA4 around Oct 8–15. Even "too little ChatGPT traffic to tell" is an honest data point.]

🧰 The Tools That Arrived the Same Month

Here's the irony: while ChatGPT's tracking wobbled, the other big players quietly gave us better instruments.

  • Bing Webmaster Tools – Topics Portfolio (beta, Oct 13). Inside the AI Performance report, it groups the grounding queries where your site gets cited into topic clusters, with rough topic size, how many sites compete, and a "Top Opportunity Queries" list where others are cited but you aren't. Microsoft warns that small sites may see a short list or nothing at all.
  • Microsoft Clarity – Brand Terms in AI Visibility (Sep 29). You can now define your brand's name variants, abbreviations and typos, so Clarity can split branded from non-branded AI queries more accurately.
  • Gemini – UTM parameters on outgoing links (reported Oct 1). Google added them without an announcement; Google's John Mueller confirmed seeing them on Reddit. The exact parameter values aren't documented, so check your own reports rather than trusting a filter someone posted.

For a small brand like Flixlar, that Brand Terms setting is exactly the problem I've been testing by hand for ten weeks: "Flixlar", "flixlar.com", and whatever creative misspellings people type.

🧪 What I'm Changing in My Own Setup

My brand tests so far have been manual: same prompts, nine engines, screenshots, dates. That tells me whether AI knows Flixlar. It doesn't tell me whether anyone clicks. So for Part 3, I'm adding a measurement layer:

  1. Annotate October 8 in GA4. Huesler now uses Oct 8–15 as the affected window. Anything in that week gets a footnote, not a conclusion.
  2. Watch ChatGPT and Direct side by side. Per landing page, and for the homepage. If one drops while the other rises, it's a labelling problem, not a traffic problem.
  3. Count the UTM, not the referrer. Huesler notes most ChatGPT clicks still arrive without a referrer, so the utm_source=chatgpt.com tag is the signal to trust.
  4. Connect Bing Webmaster Tools and Clarity. Add Flixlar's brand terms and see whether Topics Portfolio has enough data for a site my size. [Mads: add result once you've checked.]
  5. Keep the manual tests. Analytics shows clicks. Prompt tests show whether you're mentioned at all. You need both.

💡 Key Takeaways

  • The early-October ChatGPT "traffic drop" looks mostly like a tracking bug: missing UTM tags pushed visits into Direct.
  • A fix reportedly began rolling out Oct 15, but OpenAI hasn't confirmed it publicly. Footnote Oct 8–15 in your reports.
  • AI answers are changing shape (Intelligent UI, more bare homepage URLs), so tracking will keep getting harder, not easier.
  • Bing and Clarity now offer free AI-visibility reporting. Use them, but treat the numbers as approximate.
  • Being known by AI and getting clicks from AI are two different metrics. Measure both.

💬 Final Thoughts

I almost wrote a gloomy post about AI search traffic drying up. Glad I waited a week.

Did you see the drop in your own analytics, and did it come back after Oct 15? I'd love to compare notes. Reply on LinkedIn or through the contact page, and I may include your numbers in Part 3.

Sources (checked 19 Oct 2026)