ChatGPT's source list averaged 38.8. Count the links on screen.

When a customer asks who to hire, two counts can both be called citations. One is the list of sources a tool returns with the answer. The other is a citation the answer shows. This week two vendor studies updated those counts. They do not measure the same thing.

What changed

On October 25, cloro published its AI Citation Tracker. The page counts 1,962,077 successful answers from May 1 through October 25, 2026. cloro’s customers track about 2,200 prompts in 38 countries. Those prompts lean commercial. In this study a source is an entry in the list the engine returns with the answer. That list can be longer than the links in the answer text.

On that count, company-reported, ChatGPT had the largest rise. In a ChatGPT answer that cited anything, the average list was 14.2 in the week of May 4. It was 38.8 in the week of October 19. That October week, every ChatGPT answer in the set had at least one source. The 38.8 does not come from dropping blank answers.

Google AI Mode had the largest fall on the same count. In an answer that cited anything, the average list was 13.9 in the week of May 4. It was 5.4 in the week of October 19. That October week, 87.1% of AI Mode answers had a source. The 5.4 leaves out the answers that had none.

On October 26, Foglift Research updated its GEO Statistics Report. This vendor study counts a citation the answer shows, and it counts a cited domain once inside each answer. Across 62,455 successful answers from March 20 through October 26, 2026 UTC, 77.7% included at least one shown citation. The answers averaged 6 shown citations. Both figures are company-reported.

By tool, still company-reported, over that same window:

• ChatGPT showed a citation on 94% of 11,071 answers and averaged 3.29.
• Perplexity showed a citation on 89.5% of 15,769 answers and averaged 9.54.
• Google AI Overviews showed a citation on 84.2% of 14,265 answers and averaged 7.52.
• Claude showed a citation on 60.7% of 9,820 answers and averaged 2.42.
• Gemini showed a citation on 52.4% of 11,530 answers and averaged 4.9.

youtube.com was the most-cited domain in that set, at 1.8% of 374,512 citations. That share is company-reported.

Keep the two counts apart

Foglift’s 3.29 and cloro’s 38.8 are not two readings of the same answers. The prompts differ. The dates differ. One counts shown citations and counts a domain once per answer. The other counts entries on the returned list, and only on answers that cited anything.

Subtracting 3.29 from 38.8 does not count hidden links. The studies do not support that math.

What they support is narrower. On one company-reported set, ChatGPT’s returned list got much longer by the week of October 19. On another, ChatGPT’s shown citations average 3.29 from March 20 through October 26. A person who hears that ChatGPT cites 38.8 sources, and then pictures 38.8 links on screen, is using the wrong count.

Google AI Mode moved the other way, and only on the returned-list count. On cloro’s set, the average list for an AI Mode answer that cited anything fell from 13.9 to 5.4. That is a shorter list on these prompts. It does not say which business a buyer sees.

Gemini shows why the rate needs a definition. On Foglift’s long window, 52.4% of Gemini answers showed a citation. On cloro’s customer prompts, for the week of October 19 only, 72.9% of Gemini answers had a returned source. That is not a before and after.

The list above is the spread across tools. Foglift’s Perplexity average is not its ChatGPT average or its Claude average. In that study, the tool changes the count.

The check to run this week

Pick one question a customer asks before they hire you. Use their words. For example: “Who can replace a water heater this week?”

Ask it once in ChatGPT and once in Google AI Mode. Save both answers with the date. Then write four lines for each tool.

1. How many links are printed in the answer.
2. In those printed links, mark you, your rival, or nobody.
3. If the screen also shows a longer source list, write how many entries it has. If you cannot see one, write “not shown.”
4. If your site appears only on that longer list, write “list only.” Do not count it as a printed link.

If ChatGPT prints three links and your rival is one of them, line 2 is rival. If you appear only on the longer list, line 2 stays nobody and line 4 says list only. The buyer can use the printed links.

Your lines are what you saw on that pass. They will not match either study. Ask the same question the same way next week before you change a page.

When a report quotes a citation number, ask three things first. Is it printed links, or the returned list? Is the average for every answer, or only for answers that cited something? Which tool, which dates, and whose prompts?

What to do this week

• Write one customer question in the customer’s words.
• Ask it in ChatGPT and in Google AI Mode. Save both answers with the date.
• Count the printed links. Mark each answer: you, your rival, or nobody.
• If a longer source list is on screen, write its length on its own line.
• If someone sends a citation number, ask which of the two counts it is.

Tell us who AI names instead of you. Book a war briefing with New Reward, and bring their name.