How many sources an AI engine puts in an answer decides how many places there are to win. That number is not fixed. This page tracks it week by week for six engines, with the share of answers that cite a source at all and the model that serves ChatGPT.
Published October 15, 2026 · 1,853,557 answers · May 1, 2026 to October 15, 2026 · updated each day
ChatGPT: 14.2 → 37.1
ChatGPT had the largest rise in sources. The mean number of sources in an answer that cited one was 14.2 in the week of May 4, 2026 and 37.1 in the week of October 12, 2026.
Google AI Mode: 13.9 → 4.0
Google AI Mode had the largest fall. The mean number of sources in an answer that cited one was 13.9 in the week of May 4, 2026 and 4.0 in the week of October 12, 2026.
Gemini: 67.1%
Gemini cites a source least often. Its share of answers with a source was 65.0% in the week of May 4, 2026 and 67.1% in the week of October 12, 2026.
gpt-6: 99.9%
In the week of October 12, 2026, gpt-6 served 99.9% of the ChatGPT answers that named a model.
An answer with no source has no citation to win. Each chart shows, for one engine, the share of its answers that carried one source or more. For Google AI Overview it is the share of Google searches that returned an AI Overview with sources.
ChatGPT
Perplexity
Copilot
Gemini
Google AI Mode
Google AI Overview
Each chart shows the mean number of sources in an answer that has one or more. The six charts share one scale. A source here is an entry in the list the engine returns with the answer, which can be longer than the links shown in the answer text.
ChatGPT
Perplexity
Copilot
Gemini
Google AI Mode
Google AI Overview
| Engine | Answers with a source, first week | Last week | Sources in a cited answer, first week | Last week |
|---|---|---|---|---|
| ChatGPT | 96.8% | 100.0% | 14.2 | 37.1 |
| Perplexity | 99.9% | 100.0% | 10.1 | 10.7 |
| Copilot | 95.9% | 93.8% | 4.6 | 5.5 |
| Gemini | 65.0% | 67.1% | 6.1 | 10.6 |
| Google AI Mode | 95.5% | 88.5% | 13.9 | 4.0 |
| Google AI Overview | 81.1% | 74.9% | 9.8 | 5.5 |
The first and last week are the first and last with four or more measured days for that engine, so they can differ between engines. Each chart above names its weeks.
ChatGPT picks the model for each answer, and the answer names it. The chart shows each model's share of the answers that named a model. 71.3% of the ChatGPT answers in the period named one. A week with fewer than 500 such answers has no point.
Figures are for the week of Oct 12.
Each row counts the answers one model served. The models served in different periods, so a difference between two rows can come from the period and not from the model.
| Model | Served | Answers | With a source | Sources in a cited answer |
|---|---|---|---|---|
| gpt-6 | October 8, 2026 to October 15, 2026 | 13,944 | 100.0% | 37.2 |
| gpt-5-6 | August 7, 2026 to October 14, 2026 | 88,400 | 97.8% | 38.0 |
| gpt-5-5 | May 23, 2026 to September 11, 2026 | 89,411 | 99.5% | 14.5 |
| gpt-5-5-mini | July 6, 2026 to August 28, 2026 | 4,205 | 99.8% | 15.7 |
| gpt-5-3-mini | May 23, 2026 to July 6, 2026 | 32,803 | 70.1% | 12.1 |
The share of ChatGPT answers that ran one web search or more, the share that carried one ad or more, and the share that carried a shopping card. An ad is counted when OpenAI served it with the answer.
Answers with a web search
Answers with an ad
Answers with a shopping card
The share of Google searches that returned an AI Overview. The searches are full questions, not keywords. It was 80.3% in the week of October 12, 2026. The highest week was the week of July 27, 2026, at 86.0%.
Searches with an AI Overview
The figures come from cloro's monitoring: about 2,200 prompts in 38 countries, each sent to each engine once a day through the cloro API. The page counts 1,853,557 successful answers from May 1, 2026 to October 15, 2026. The prompts are the ones cloro's customers track, so they lean commercial. Read a level as a figure for this prompt set and a change over time as the signal.
cloro. (October 15, 2026). AI Citation Tracker: How Many Sources Each AI Engine Cites. cloro Research. https://cloro.dev/research/ai-citation-tracker/
More studies are in the research index. For the same measures on a fixed, topic-balanced query set, see the AI Search Index.
In each engine's last full week measured, an answer that cited a source had on average: ChatGPT 37.1 (the week of October 12, 2026), Perplexity 10.7 (the week of September 28, 2026), Copilot 5.5 (the week of October 12, 2026), Gemini 10.6 (the week of October 12, 2026), Google AI Mode 4.0 (the week of October 12, 2026) and Google AI Overview 5.5 (the week of October 12, 2026). The figures are from cloro's daily monitoring prompts.
In each engine's last full week measured: ChatGPT 100.0% (the week of October 12, 2026), Perplexity 100.0% (the week of September 28, 2026), Copilot 93.8% (the week of October 12, 2026), Gemini 67.1% (the week of October 12, 2026), Google AI Mode 88.5% (the week of October 12, 2026) and Google AI Overview 74.9% (the week of October 12, 2026). For Google AI Overview the figure is the share of Google searches that returned an AI Overview with sources.
Yes. The mean number of sources in a ChatGPT answer that cited one was 14.2 in the week of May 4, 2026 and 37.1 in the week of October 12, 2026.
Yes. The mean number of sources in a Google AI Mode answer that cited one was 13.9 in the week of May 4, 2026 and 4.0 in the week of October 12, 2026.
In the week of October 12, 2026, 96.7% of ChatGPT answers ran one web search or more. The highest week was the week of May 4, 2026, at 100.0%. The figures are from cloro's daily monitoring prompts.
In the week of October 12, 2026, 1.7% of ChatGPT answers had an ad. The highest week was the week of August 17, 2026, at 27.4%. The figures are from cloro's daily monitoring prompts.
In the week of October 12, 2026, 0.2% of ChatGPT answers had a shopping card. The highest week was the week of August 10, 2026, at 13.2%. The figures are from cloro's daily monitoring prompts.
In the week of October 12, 2026, 80.3% of Google searches returned an AI Overview. The highest week was the week of July 27, 2026, at 86.0%. The figures are from cloro's daily monitoring prompts.
In the week of October 12, 2026, gpt-6 served 99.9% of the ChatGPT answers that named a model. 71.3% of the ChatGPT answers in the full period named one.
cloro returns the answer and its source list from ChatGPT, Perplexity, Gemini, Copilot and Google's AI surfaces as structured data. Start with 500 free credits a month.
