Why did ChatGPT's zero-source answers jump 8x in late 2026?

On 31 October 2026, the share of ChatGPT answers with zero cited sources jumped from 1.5% to 12.2% and stayed above 10% for the following nine days, according to Goodie's tracking of ChatGPT answers collected between 9 October and 9 November 2026. The jump lines up with OpenAI's rollout of GPT-6 and its new Intelligent UI, which renders some answers as widgets with no citation surface at all. It is best read as a second, sharper layer on top of a citation decline that has been building through 2026, not a one-off glitch.

What the 8x jump actually was

The numbers come from a specific before-and-after comparison. In the baseline period, 9 to 30 October, Goodie's study recorded 65,431 answers, of which 984 carried no sources at all. In the comparison period, 31 October to 9 November, there were 29,420 answers, with 3,601 carrying no sources. Citations per answer fell 17% across the same window, and from 31 October onward the zero-source rate never dropped back below 10%, a pattern based on that single tracked sample rather than a platform-wide audit.

That is the headline: a sudden, sustained step-change on a specific date, not a gradual drift.

What changed in ChatGPT right before the spike

The timing points to OpenAI's model rollout. GPT-6 started reaching ChatGPT users on 7 October 2026, bundled with a feature called Intelligent UI that assembles answers from text, charts, buttons, forms and small interactive tools rather than plain paragraphs, as described by an AI Engineering Simplified explainer on the GPT-6 rollout. GPT-6 Sol became the default for Plus, Pro, Business and Enterprise users on 7 October, with GPT-6 Luna following for Free and Go users from 8 October, according to Cellcog's coverage of the rollout schedule. The engine upgrade reportedly sped up web search responses by 44 percent and allowed real-time answer streaming, alongside the optional Intelligent UI for interactive charts and tools, per Android Headlines' reporting on the GPT-6 engine changes.

The model landed on 7-8 October. The zero-source spike came three weeks later, on 31 October. Goodie's own timeline log treats these as two separate events, calling October 31 "the second layer" rather than the direct trigger, based on Goodie's analysis of its own tracking data. For more on how the new model changes citation behaviour generally, see our piece on whether GPT-6 Astra changes how ChatGPT decides which sources to cite.

Is this a display bug or a real drop in sourcing?

There was a genuine display bug around the same period, but it was a different one. Before the zero-source spike, ChatGPT went through a UTM-tagging incident: by 14 October, about 67% of answers with cited links still included a tagged source link, before a sharper decline that evening, according to Profound's tracking of ChatGPT's link tagging. At 02:00 UTC on 16 October, only about 25% of answers with cited links included a tracking tag, but by 06:30 UTC that had recovered to roughly 99%, per the same Profound data. That incident affected whether links carried tracking tags, not whether sources were shown at all, and it resolved within hours. We covered that bug separately in our piece on whether the ChatGPT traffic drop in October 2026 was a UTM bug, now fixed, and on how dropped tracking tags made AI traffic numbers unreliable.

The 31 October zero-source spike is different. Goodie's own judgment is that there is "no evidence that OpenAI hid citations on October 31", which points to a genuine behavioural or retrieval shift rather than a UI masking bug, according to Goodie's study. Part of the mechanism is structural: some ChatGPT answers now render through a "genui_run" call that invokes named widgets using JSON arguments instead of standard cited text, leaving no citation surface at all, as described in Suganthan's research into ChatGPT's widget rendering. We have written more on this shift in our explainer on how ChatGPT's Intelligent UI shakes up what counts as a citation.

How this fits the bigger 2026 pattern of ChatGPT citing less

The October spike is not an isolated event. In the US, the share of ChatGPT answers with no citation at all doubled from 28% to 48% in March 2026, before rebounding in May, according to seoClarity's data reported by gogochimp. During that earlier February-to-April episode, citation volume across the US, UK, Canada, Germany and Italy dropped 86-94% from its February baseline by the end of April, arriving in two distinct waves roughly six weeks apart, based on seoClarity's tracking cited by Lyra. The US zero-citation rate roughly doubled during March alone, from 28% to 48%, while the UK moved from 24% to 40% in the same window, per the same Lyra report. We covered the Reddit-specific version of this earlier drop in our piece on Reddit's ChatGPT citations dropping 86% in four days.

Seen against that history, the October spike looks like another step down a staircase ChatGPT has been descending all year.

Why ChatGPT skips citations at all

The underlying mechanism is a pre-retrieval decision. GEO Lab's E042 experiment found a gate sits before retrieval even begins, and when that gate stays closed, ChatGPT answers from training data with no page to cite, regardless of how good any candidate page is, according to GEO Lab's E042 findings. On the default "auto" setting, the model can answer from training data and return an empty citation array, which reads as zero even on a site ChatGPT normally cites constantly, per Citability's audit of the citation trap.

Economics drive this. Retrieval costs money and adds latency, so when training data already carries an answer with high confidence, ChatGPT answers direct and skips the citation stack entirely, according to gogochimp's analysis of ChatGPT's citation behaviour. Separately, Ahrefs analysed 66.6 million free, logged-out GPT-5.6 Luna responses from August and September 2026 and found only 14.4% triggered a web search at all, meaning six in seven prompts were answered without looking anything up. We covered this in detail in our piece on ChatGPT skipping the web for six in seven prompts, per Ahrefs. When ChatGPT does search, it runs an average of about 1.6 fan-out queries per response, per Search Engine Journal's coverage of the same Ahrefs dataset.

What's still cited when ChatGPT does search

The mix of sources has shifted sharply even where citations survive. Ahrefs found Reddit citations in US answers fell from 19.1% in March 2026 to 2.3% in September, and Wikipedia fell from 14.1% to 2.9% over the same period, per Search Engine Journal's report on Ahrefs' findings. Meanwhile PubMed rose from 64th to 4th among the most-cited domains in US answers, while WebMD fell from 18th to 79th, a move toward primary and authoritative sources in the same data. A longitudinal personal-finance study found that of 215 pages ChatGPT cited most in October 2025, 107 (50%) get zero citations by September 2026, and the tracked group kept only 14% of its original citation volume, according to Green Flag Digital's citation study. In that study, bank, card issuer and government pages rose from 2% to 28% of top citations while Wikipedia and Reddit fell from 18% of top-500 citations to none.

Fan-out query analysis also shows ChatGPT increasingly hunting for official sources directly, with about 6.8% of background queries using the "site:" operator, three in four of those targeting brand websites, and use of the word "official" rising to 13.1% in September 2026, according to Startup Hub AI's analysis of ChatGPT's fan-out behaviour. A 4,000-prompt Ahrefs Brand Radar analysis found health answers cited government, medical and academic sites 82% of the time, definitions cited dictionaries 63% of the time (Cambridge alone at 42%), and brand questions cited company sites 59% of the time, per the same Startup Hub AI report.

What founders and marketers should do about it

Our view: the zero-source spike matters less as a single incident and more as proof that ChatGPT's citation behaviour is now unstable by design, shifting with every model and UI change OpenAI ships. Chasing any one snapshot of citation rates is a weaker strategy than building pages that win when ChatGPT does decide to search, whether that's primary data, official brand pages or authoritative institutional sources. The clearest limitation in all of this evidence is that it tracks what gets cited, not what gets recommended, and a page can shape an answer without ever appearing as a link. For a practical starting point, Sited's free AI visibility scorecard shows where your own site currently stands with ChatGPT and other AI answers. You may also find it useful to read our explainer on why topping Google means nothing to ChatGPT right now.

Frequently asked questions

Was the October 31 spike caused by a bug?

Goodie's own judgment is that there is no evidence OpenAI hid citations on that date, which points to a genuine retrieval or behavioural shift rather than a display bug, according to Goodie's study. A separate, unrelated UTM-tagging bug did hit ChatGPT in mid-October but resolved within hours, per Profound's tracking.

Did GPT-6 directly cause the zero-source jump?

The timing is suggestive but not proven as direct cause and effect. GPT-6 and Intelligent UI rolled out on 7-8 October, three weeks before the 31 October spike, which Goodie's timeline describes as a separate "second layer" of change, based on Goodie's analysis.

How often does ChatGPT even search the web?

Very rarely among free, logged-out users. Only 14.4% of 66.6 million responses triggered a web search in Ahrefs' August-September 2026 dataset, meaning most prompts were answered from training data alone, per Ahrefs' study covered by dev.to.

Which sites are gaining citations as others lose them?

Primary and official sources are gaining ground. PubMed rose from 64th to 4th among most-cited US domains between March and September 2026 while WebMD fell from 18th to 79th, and bank, card issuer and government pages rose from 2% to 28% of top citations in one finance-specific study, according to Search Engine Journal's Ahrefs coverage and Green Flag Digital's research.