Trends

A year of change in AI search

These charts follow the same set of long-running studies month by month, so changes reflect the engines, not a change in who was measured.

Engines
14 → 4
sources per Google AI Mode answer, before and after late August 2026

Google AI Mode cut its sources by two-thirds

In the last week of August 2026, Google AI Mode went from about 14 distinct domains per answer to about 4, and it has stayed there into November. It shows up in the same week across studies tracked on two independent platforms, so it reflects a change in Google’s product rather than in measurement. In October AI Mode averaged 3.7 sources per answer, against 7.7 for Perplexity, 7.5 for ChatGPT and 6.4 for Google AI Overviews; only Gemini (3.2) cites fewer. Each remaining AI Mode citation now carries far more weight.

Engines
22%
of OpenAI API answers cite Wikipedia

Each engine has its favourite platforms

YouTube leads on Google AI Overviews, and LinkedIn shows up in 13% of Perplexity answers and 12% of AI Mode answers. The OpenAI API cites Wikipedia in 22% of answers, against 7% for the ChatGPT app. Claude almost never cites Reddit or YouTube.

Engines
42% → 30%
earned-media share of Claude citations, Sonnet 4.6 → 5

Model upgrades move citations overnight

Within the same studies, the switch from Claude Sonnet 4.6 to Sonnet 5 in August 2026 cut earned media’s share of citations from 42% to 30%, and sources per answer dipped from 6.6 to 6.2. Visibility has to be tracked by model version, not just by engine.

Shopping and product cards

Share of answers that show shopping or product cards

Ads inside AI answers

Share of answers carrying an ad placement

Community and reference platforms over time

Share of answers retrieving each platform, same set of studies every month

Model updates change what gets cited

Profile of answers from each model version. Versions ran in different sets of studies, so read differences as indicative; the engine insights above quote within-study comparisons.

EngineModel versionIn useAnswersSources / answerBrand sitesEarned mediaSocial & UGCLength
ChatGPTChatGPT (app)Aug 25 – Nov 26157,7688.431%31%5%3,523
ClaudeClaude Sonnet 4Nov 25 – Nov 2624,0558.038%23%6%2,910
ClaudeClaude Sonnet 4.6Apr 26 – Aug 2629,2707.437%28%7%4,081
ClaudeClaude Sonnet 5Aug 26 – Sep 266,1246.532%29%5%3,063
GeminiGemini (app)Nov 25 – Nov 26106,3053.744%20%6%3,047
Microsoft CopilotMicrosoft Copilot (app)Apr 26 – Nov 262,5583.249%17%4%3,225
OpenAI APIGPT-4o SearchNov 25 – Nov 265,2402.258%14%1%1,038
OpenAI APIGPT-5 SearchDec 25 – Jul 2635,6956.537%28%3%6,243
OpenAI APIGPT-5.6 LunaJul 26 – Sep 268,8294.146%8%2%4,302
PerplexityPerplexity Sonar APIAug 25 – Sep 262,90414.331%33%6%1,552
PerplexityPerplexity (app)Oct 25 – Nov 26144,6148.133%30%7%1,417
Notes.
  • Web search, shopping and ad rates come from daily tracking of the same question sets across engines, on a fixed panel of studies present in at least 75% of months.
  • Platform lines follow a fixed panel of long-running studies so that month-to-month changes are not caused by studies starting or ending.
  • Model versions are shown where at least 2,000 answers were collected from that version.
Your brand in AI search

How is your brand showing up in AI answers?

We benchmark how ChatGPT, Gemini, Google AI Overviews, Perplexity, Claude and Copilot describe, cite and recommend brands, then build the content and earned-media plan to change it.

Want to understand your brand opportunities and risks on AI search? Contact We AI & Data Advisors for a sample audit: AIAdvisors@wecommunications.com

Request a sample audit