Home / Research / News, read closely

ChatGPT answers with direct website links jumped from about 10% to over 40% in October 2026

By Bainian16 November 20263 min read2 sources

10% → 41%ChatGPT answers with a direct website link

The short answer

Since GPT-6 rolled out on 8 October 2026, about 4 in 10 ChatGPT answers link a website directly, up from 1 in 10, and over half of those links go to home pages.

  • If your ChatGPT traffic 'fell' in mid-October, it probably didn't. A bug hid where those visitors came from, so they showed up as Direct.
  • The bug lasted from about 8 to 15 October. Compare that week with care.
  • ChatGPT links you when it names your brand. Being named in lists and articles now matters more.
How this was researched

Started from a screenshot of a LinkedIn post by David Konitzny (Peec AI); the full post needs a login and we couldn't read it. We checked it against the same-method data from his colleague Tomek Rudzki and public reports: Search Engine Journal, Search Engine Roundtable, the Web Traffic News briefing, Delante (12 client projects) and Global Gravity's write-up.

Short version: after GPT-6 rolled out on 7 and 8 October 2026, ChatGPT started putting website links right inside its answers far more often. About 1 in 10 answers had one before. About 4 in 10 have one now.

More than half of those links go to home pages.

At the same time, a bug made those visits hard to see. Many sites thought their ChatGPT traffic had dropped. It hadn’t.

Two terms first

  • UTM tag: a short label added to the end of a link, like ?utm_source=chatgpt.com. It tells your analytics where a visitor came from.
  • Referrer: the browser’s own note of which page sent a visitor.

When a link has neither, analytics tools such as GA4 (Google Analytics 4) file the visit under Direct, as if the person typed your address.

What happened, step by step

Date What happened
7 May 2026 ChatGPT starts adding clickable links to brand names. Similarweb: about 60% of that traffic lands on home pages.
7 to 8 Oct GPT-6 rolls out, first to paying users, then to everyone.
From 8 Oct Links inside answers lose their UTM tag and send no referrer. Visits show up as Direct.
9 Oct David Konitzny (Peec AI): answers with at least one website link jump from about 10% to 41%.
9 to 13 Oct Tomek Rudzki (Peec AI), across 1 million prompts: 42% of answers link a website directly, up from 8% to 10%. 54% of those links go to home pages.
14 Oct Konitzny: peak at 49.5%.
15 to 16 Oct Tags are back. Adobe’s Cedric Huesler says OpenAI confirmed it was a bug.

What these numbers count: links shown in answers, not clicks. Konitzny and Rudzki both work at Peec AI.

The “traffic drop” that wasn’t

  • According to Graphite (several hundred sites), ChatGPT visits in GA4 fell 22% to 57%. On phones, almost the only way to spot a ChatGPT visit is the UTM tag, so phones were hit hardest.
  • Adobe’s Cedric Huesler (about 500 pages): about 31,000 fewer ChatGPT visits a day, and about 26,000 more Direct visits a day. The new Direct visitors behaved like ChatGPT users.
  • Delante (12 client projects): 11 saw a drop, typically 40%. Five online shops saw 26% less revenue credited to ChatGPT.

One open question: Sébastien Billard saw ChatGPT’s own page-fetching bot visiting less, while its search bot visited more. So not every change may be a tracking issue.

Everyone selling these measurements (Graphite, Profound, Peec AI) also sells AI search monitoring. Keep that in mind.

Why it matters

1. AI answers now send real visitors

Since May, brand names get links. Since October, 4 in 10 answers include a direct link. Being named by ChatGPT is worth more than it was.

The links mostly go to home pages, attached to brand names. So ChatGPT has to know your brand and mention it first.

Other data points the same way:

  • “Best X” lists make up 43.8% of pages ChatGPT cites (Tim Soulo, summarising Ahrefs studies).
  • Being mentioned by name tracks with AI visibility more than links do: 0.664 vs 0.326 (Patrick Stox, Ahrefs).

3. Your mid-October numbers are wrong

If you read ChatGPT traffic in GA4, the week of 8 to 16 October can’t be compared with the weeks around it.

4. Your home page has to work for a stranger

More than half of these visitors land on your home page with no context. The first screen should say what this is and let them use it.

5. Cited is not the same as linked

“Cited in the sources panel” and “linked in the answer” are different things. Count them separately.

Limits of this evidence

  • I only saw the first half of Konitzny’s post. Rudzki’s 42% and 54%, from the same company, are the cross-check.
  • Every number comes from companies that sell AI monitoring. Most don’t publish sample sizes.
  • These count links shown, not clicks.
  • I found no November data on whether the link rate held.

✓ Do this next

Open your analytics, mark 8 to 16 October as unreliable, and look at ChatGPT plus Direct visits to your home page together for that week.

My takeThis looked like bad news and was the opposite. ChatGPT is sending more people to websites, not fewer. The sites that win are the ones it can name.

Sources

  1. Search Engine Journal: ChatGPT traffic drop appears to be a tracking problemsearchenginejournal.com
  2. Search Engine Roundtable: OpenAI drops UTM parameters from ChatGPT resultsseroundtable.com

Follow the research on X

The short version of every piece, plus daily notes on SEO and growth.