All posts

5 min read

Read is not cited: ChatGPT consulted 14.5 pages per answer and cited 3.8

Archipel’s October AI Visibility Observatory covers 406,626 AI answers. Three of its findings argue for checking weekly and for fixing pages that get read but never quoted.

Vishal Verma, author of Read is not cited: ChatGPT consulted 14.5 pages per answer and cited 3.8Vishal Verma
See the numbers
MeasurePages per answer
Consulted14.5
Cited3.8
Source: Archipel Marketing, AI Visibility Observatory, October 2026. Archipel is a GEO agency.

Archipel is a GEO agency in France. It publishes an AI Visibility Observatory, and its October 2026 edition came out on Oct 11. Archipel sells GEO audits, so this is an agency study. The questions are mostly about buying in France, in French.

What the Observatory covers

The edition analyses 406,626 answers from six AI engines to more than 1,700 buying questions, collected from Jul 12 to Oct 11, 2026. ChatGPT, Google AI Overviews, Perplexity and Copilot come from Profound data (368,556 answers). Claude and Gemini come from Peec AI data (38,070 answers).

The questions span nine sectors. The biggest by answers are B2B software and tech (145,731) and travel and leisure (84,359). Retail and fashion has 33,429. Sources are classified automatically from their address and title. Archipel publishes no brand or client names.

Being read is not being cited

Archipel reports that ChatGPT consults 14.5 pages per answer and cites 3.8. So a page can be read by ChatGPT and still not appear as a source. The page does not publish how it tells a page consulted from a page cited, so treat the exact numbers as Archipel’s own.

The first brand named keeps changing

The brand cited first changed from one week to the next in 63.3% of cases. A spot at the top of one answer is not a spot you keep.

AI Overviews is on most buying searches

Google AI Overviews appeared on 76.5% of tracked searches across the period. The rate rose from 56.3% in July to 86.7% in October.

Rankings are a big source

Rankings and comparison pages are 29.9% of all cited sources, and 88.1% for Claude, 61.6% for Gemini and 48.5% for Copilot. Archipel adds that ChatGPT cut forums from 6.1% of its sources in July to 0.1% in October.

Read the numbers with care

  • Most questions are in French and about France, with some for the United States and the Netherlands.
  • Most answers are from B2B software and travel, not consumer goods.
  • It is aggregated data, with sources classified by machine.
  • It is the first edition, so there is little history to lean on.

What to do this week

Tick these off as you go.

How Shopthru.ai helps

Shopthru.ai is an AI agents team for consumer brands. It asks ChatGPT, Gemini, Google AI, Perplexity and Copilot what your shoppers ask, finds where AI gets your products, prices or policies wrong, and drafts each fix into an approval inbox. It ships only on your yes, then checks again. Instagram and Amazon are next.

Sources

See what AI would buy from you.