Skip to content
FRFree GEO audit

AI Visibility Observatory

AI Visibility Observatory: which sources AI engines cite when people are ready to buy

Rankings and comparison pages account for 29.9% of all sources cited by AI engines on buying questions, and 88.1% for Claude. The Archipel Observatory analyzes 406,626 answers from six engines collected between July 12 and October 11, 2026.

Get my free GEO auditBook a 30-min call

Updated

406,626

answers analysed

9

sectors

6

ai engines

63%

first-cited brand changes week to week

Share of rankings among cited sources, by engine
Claude88%Gemini62%Microsoft Copilot49%Perplexity33%ChatGPT29%Google AI Overviews25%
Searches where Google displays an AI Overview, by sector
Coaching and advisory92%Marketing agencies89%B2B software and tech88%Finance and wealth management83%Retail and fashion78%Events and corporate retreats77%Health, beauty and fitness76%Real estate67%Travel and leisure54%

The Archipel AI Visibility Observatory measures how AI engines answer purchase-intent questions: "which agency should I hire", "best software for", "who to call for". It covers 406,626 answers to over 1,700 buying questions tracked by Archipel across nine sectors, collected between July 12 and October 11, 2026. Most questions are in French and concern the French market; smaller sets cover the United States and the Netherlands. Informational questions ("how do I do my accounting") are excluded on purpose: we only look at the moment a buyer asks an AI who to choose.

Two data sources feed the Observatory. Profound supplies 368,556 answers from ChatGPT, Google AI Overviews, Perplexity and Microsoft Copilot. Peec AI supplies 38,070 answers from Claude and Gemini, plus the pages ChatGPT consults before answering. On top of that, five head-to-head comparisons put identical questions to two engines.

The headline finding is the weight of third-party rankings. Across all engines, rankings and comparison pages make up 29.9% of cited sources. The share varies widely by engine: 88.1% for Claude, 61.6% for Gemini, 48.5% for Copilot, 33.4% for Perplexity, 28.5% for ChatGPT and 25.1% for Google AI Overviews. Brand sites, media and blogs make up most of the rest; reviews and directories weigh 4.0%, YouTube 1.5%, forums 1.0%.

AI answers are unstable. For the same question on the same engine, the brand cited first changes from one week to the next in 63.3% of cases. A single screenshot of a ChatGPT answer proves nothing; visibility has to be measured over weeks. Google AI Overviews is also gaining ground: it appeared on 76.5% of tracked buying searches over the period, rising from 56.3% in July to 86.7% in October.

Engines do not read the same amount. Perplexity cites 9.7 sources per answer, Google AI Overviews 8.0, ChatGPT 4.6, Claude 4.2, Copilot 3.6 and Gemini 2.5. ChatGPT consults 14.5 pages per answer but cites only 3.8 of them: being read is not being cited.

Key figures

AI engines

Sectors

Source types

Head-to-head

Method

406,626 AI answers collected every day from July 12, 2026 to October 11, 2026 on over 1,700 purchase-intent questions tracked by Archipel (mostly in French, plus the United States and the Netherlands). ChatGPT, Google AI Overviews, Perplexity and Copilot: Profound data. Claude and Gemini: Peec AI data. Sources are classified automatically from their address and title. Figures are aggregated: no brand or client is published. Free to reuse with a link to this page.

FAQ

Frequently asked questions

What is the Archipel AI Visibility Observatory?
The Archipel AI Visibility Observatory is a dataset of 406,626 AI answers to over 1,700 buying questions tracked by Archipel, collected between July 12 and October 11, 2026, across six engines and nine sectors. It shows which sources ChatGPT, Claude, Gemini, Perplexity, Copilot and Google AI Overviews cite.
Which sources do AI engines cite most on buying questions?
On buying questions, AI engines cite brand sites, media and blogs most, then rankings and comparison pages, which account for 29.9% of cited sources. Reviews and directories follow with 4.0%, YouTube with 1.5% and forums with 1.0%.
Where does the Observatory data come from?
The Observatory data comes from Profound (368,556 answers from ChatGPT, Google AI Overviews, Perplexity and Copilot) and Peec AI (38,070 answers from Claude and Gemini). All questions are purchase-intent questions tracked by Archipel.
Which markets does the Observatory cover?
The Observatory covers mostly France, plus the United States and the Netherlands. Most of the over 1,700 tracked questions are in French, so results describe AI behavior on French-language buying questions first.
How often does the brand AI recommends first change?
The brand an AI engine cites first changes from one week to the next in 63.3% of cases, for the same question on the same engine. That is why AI visibility must be tracked weekly rather than checked once.

Shall we get started?

Free GEO audit: your rank on ChatGPT, Claude and Google AI Overviews, your competitors and your first actions. Reply within half a business day.

Get my free GEO auditBook a 30-min call