Why Graphite dropped its AI versus search comparison
Graphite’s revised November report explains how missing mobile app data affected its earlier comparison and why AI sessions and referrals measure different things.
Graphite has dropped its earlier comparison between AI usage and search engine usage. In a November 9 update, the research team said its earlier analysis omitted sessions in search engine apps. Google’s Android app is especially important to the comparison because people use it for traditional search, translation, image recognition, and Discover. Graphite says it cannot reliably separate those activities in its data. Graphite’s revised report and methodology
What Graphite’s new estimate measures
Using Similarweb data, Graphite estimated 76 billion worldwide web and app sessions in September 2026 across ChatGPT, Gemini, Claude, Grok, Perplexity, and DeepSeek. A session can include several prompts or no web search at all. The estimate describes usage across those six products. It can’t show how many searches it handled or how much traffic it sent to publishers.
Which metric answers your question?
| Metric | Question it answers | What to record |
|---|---|---|
| AI sessions | How often was an AI product used? | Products, web and app coverage, geography, and period |
| Web search events | How often did a product run a search? | The provider’s definition of a search and the surfaces it covers |
| Referral sessions | How many visits did our site attribute to an AI product? | Recognized referrers, landing pages, and the analytics date range |
Build a report around your own landing pages
Start with the same set of landing pages and two comparable periods. Count sessions from the AI referrers your analytics can identify and keep a separate row for organic search. Record the referral rules you used so a later change in classification does not look like growth. Then compare engagement or conversions for each group if your measurement setup supports that comparison. Guide to tracking AI referrals in GA4
If you also track appearances in AI answers, use a separate column and keep the prompt sample fixed. Visits and appearances can move in different directions. An answer can show a link without sending a visit, and a visit can arrive without a recognizable referrer. That distinction is especially important when a report combines visibility data with on-site results. Guide to connecting AI visibility and referral data
Keep learning
Continue this topic
Next in this topic
How often ChatGPT cites OpenAI’s publisher partners
Earlier in this topic
Popular brands earned more citations in an AI search audit
Research
Ask a question or join the discussion