ChatGPT's links are not its citations
Crawlmind Engineering··6 min read
A ChatGPT answer link is a clickable link to an external website placed in the answer text itself, and since early October it has become a separate list from the sources ChatGPT cites for that same answer.
That distinction used to be academic. Inline links were rare, so most teams treated "ChatGPT linked to us" and "ChatGPT cited us" as the same event. New data from Peec AI says they have split apart, and the split changes what a citation report can tell you about clicks.
#What Peec AI measured
Peec AI, a company that sells AI visibility monitoring, published daily figures on LinkedIn covering 27 August to 14 November 2026. PPC Land summarized the post on November 16 and read the numbers off the charts. Peec counted only "clickable links to external websites directly within ChatGPT's answers," and excluded citations and links that open the sources sidebar.
The headline figures, as PPC Land reports them:
- The share of conversations with at least one external link sat at about 10% until early October, jumped to 41% on October 9, peaked at 49.5% on October 14, and stood at 47.8% on November 14.
- The average number of links per linked conversation rose from roughly 2 to 4.
- Conversations with a single link fell from about 43% of linked chats to 18%, and almost 39% of linked chats now carry five or more.
- 63% of the external URLs did not overlap with the cited source URLs in the same conversation, and 47% of linked domains did not overlap with cited source domains.
The chart marks October 7 as the date GPT-6 became the default, while the steep rise starts two days later. PPC Land notes that a staggered rollout or a separate change could explain the gap, and that OpenAI has made no statement about the behavior.
#Why the two lists diverge
Citations are the pages ChatGPT retrieved and grounded the answer on. Inline links look like something else. Asked where the links point, Peec researcher David Konitzny replied that they "usually point to the homepage or a product page," according to the same PPC Land report. A commenter suggested they are mostly navigation links to the website of a brand named in the answer, and Konitzny said that was his assumption too.
Treat that as a working hypothesis, not a finding. Peec published no breakdown by link type. If it holds, the pattern is easy to picture: an answer cites a review site and a forum thread as its sources, then links the three brands it recommends to their homepages. The review site earns the citation. The brands earn the clicks.
The URL and domain figures also use different denominators, so PPC Land treats the 47% domain gap and the 63% URL gap as an indication of scale rather than a precise measure. Part of the URL gap may be a brand cited on a deep page and linked on its homepage, which is the same domain but a different URL.
#What this does to your reports
A citation tracker answers one question: which pages did the engine use as evidence. If ChatGPT now links to brands it did not cite, that tracker can show a brand as absent while the answer sends users straight to its homepage. The reverse also happens. A cited article may get no link in the answer text and depend on users opening the sources panel.
Three practical consequences follow.
Citation share stops predicting ChatGPT clicks on its own. Before October, a cited page was close to the only way to get a click from an answer. Now a brand mention with an inline link is a second path, and it may not touch your content pages at all.
Landing pages shift toward the homepage. If inline links favor homepages and product pages, the traffic they send will land there, not on the article that did the work of getting you into the answer. Reports that attribute AI traffic by landing page will credit the wrong content.
Attribution got noisier in the same window. OpenAI's publishers and developers FAQ says ChatGPT adds utm_source=chatgpt.com to referral URLs so publishers can track traffic from ChatGPT search results. On October 14, Search Engine Roundtable reported that the tag and the referrer had disappeared from ChatGPT links, so clicks were landing as direct traffic. The report was later updated after Adobe's Cedric Huesler said on October 15 that it was a bug and that tagging had returned. Anyone comparing ChatGPT referrals across early and mid October is comparing two different tracking regimes on top of two different linking behaviors.
OpenAI's own numbers will not settle this yet either. Search Engine Journal reports that Business Hub, which opened in beta on November 10, shows businesses how often ChatGPT mentions them and how many clicks it sends. It does not say whether those clicks came from a cited source or an inline link.
#The limits of the data
The Peec figures are useful because they are daily and cover the October change. They also have gaps worth stating plainly:
- The sample size, prompt set, market, language, plan, and login state are not disclosed.
- The method for extracting "cited sources" is not defined.
- There is no click or traffic data. The post itself says more links do not automatically mean more clicks.
- The share fell to the high 30s around October 20 and to about a third at the turn of the month before climbing again, per PPC Land's reading of the chart, and the post does not explain those swings.
- The study has not been replicated, and the publisher sells monitoring software.
That is enough to justify changing how you record answers. It is not enough to estimate what the change is worth in traffic.
#How to measure links and citations separately
- Store two lists per answer. For every tracked prompt, keep the cited source URLs and the inline answer links as separate fields. Never merge them into one "appeared in" count.
- Classify each inline link target. Tag it as homepage, product or pricing page, content page, or third-party page. This tests the homepage hypothesis on your own prompt set instead of relying on Peec's.
- Report three states for your brand. Cited and linked, cited only, linked only. The third state did not meaningfully exist before October and is the one a citation-only dashboard misses.
- Check the landing pages that inline links would hit. If your homepage and product pages now receive ChatGPT clicks, make sure they state what the product does, who it is for, and what it costs in plain text, because the user arrives there without the context the answer gave them.
- Annotate the mid-October window. Mark the reported GPT-6 default and the link jump on October 9, plus the tagging bug reported on October 14, in your analytics and visibility charts, so nobody reads the step change as the result of your own work.
- Compare tagged referrals with direct traffic on the same landing pages. If tagged ChatGPT visits fall while direct visits to the homepage rise in the same window, the gap is likely attribution, not demand.
The core habit is simple: a citation tells you the engine trusted your page, and a link tells you where it sent the user. Since October those are often different pages, and sometimes different companies. Record both, and report them side by side.
Related field notes
November 21, 2026 · 5 min
AI answers now hand out your coupon codes
ChatGPT and Google now show discount codes inside answers, and most codes chatbots return do not work. Make sure the codes they find are yours and current.
November 21, 2026 · 5 min
Shallow pages lose in AI search first
Google's head of Search says AI answers leave little room for pages that only restate facts. How to find your shallow pages and decide what to do with each.
November 20, 2026 · 5 min
The ten blue links are back as a tab
Google, Bing and now ChatGPT offer or test a view with only web links. That view ranks pages, not passages, and AI visibility reports often skip it.
Share or discuss
New posts, no spam. Roughly monthly. Unsubscribe with one click.