OpenAI says ChatGPT query volume is 20% of Google's; the referral numbers say 0.36%

Both figures can be true at once because they measure different stages. The 20% is query demand: how often people ask ChatGPT something they used to ask a search box. The 0.36% is delivered traffic: how often a site actually gets a session. An assistant that answers the question and cites a couple of links will always send a fraction of the referrals that ten blue links send. Rising query share alongside a near-flat referral share is the normal shape of this market, not a contradiction to resolve. The figures behind the second number are public: Ahrefs' own AI vs Search benchmark, updated 3 November 2026 across 118,671 sites, shows Google at 24.74% of tracked traffic, ChatGPT at 0.36% and all AI assistants together at 0.46% (see Ahrefs, AI vs Search Traffic Analysis). For a budget decision, only the second number moves sessions and revenue.

What should change your plan is distribution, and Google's own framing. Google used its Q3 2026 call to describe Search and AI merging into one product, with Pichai pointing to recursive self-improvement (his words on the call were that he expects "continued acceleration because of RSI" over the next twelve months) and Search becoming an agent manager, and it is testing removal of the AI Mode button while moving brand-query AI Overviews into the right-hand column (see Search Engine Journal's write-up of the call, and Alphabet's own transcript of the 28 October 2026 call). A citation economy has a flat floor and a steep ceiling: most sites get nothing, a few authority properties get named again and again. That is why the writing-and-praying version of generative engine optimisation fails, and why a small set of citable, fact-dense pages aimed at your highest-intent queries beats 500 generic FAQ articles.

Fix measurement before content. Most analytics stacks still fold assistant traffic into organic search, which credits the wrong channel and hides whether your content work is landing. Classify AI referrers explicitly in the GA4 BigQuery export instead of trusting the default channel grouping, using the event-level and session-level attribution fields Google documents in BigQuery traffic attribution data:

-- GA4 BigQuery: separate AI-assistant referrals from classic search
WITH s AS (
  SELECT traffic_source.source AS src, traffic_source.medium AS med,
         CONCAT(user_pseudo_id, CAST(event_timestamp AS STRING)) AS sid
  FROM `myproj.analytics_123456789.events_*`
  WHERE event_name = 'session_start'
    AND _TABLE_SUFFIX BETWEEN '20260901' AND '20261031'
)
SELECT CASE
    WHEN src IN ('chatgpt.com','perplexity.ai','gemini.google.com',
                 'copilot.microsoft.com','claude.ai','search.brave.com')
      THEN 'ai_assistant'
    WHEN src = 'google.com' AND med = 'referral'
      THEN 'google_referral_check'  -- referrer-stripped assistant clicks land here
    WHEN src = 'google.com' AND med = 'organic' THEN 'google_organic'
    ELSE 'other' END AS channel,
  COUNT(DISTINCT sid) AS sessions
FROM s GROUP BY 1 ORDER BY 2 DESC;

One GCC-specific reality. Arabic-language material is thin in the corpora these assistants draw on, so they default to English sources, which leaves an Arabic-first brand invisible in the answers its own customers now receive. That is a content gap you can close on your own domain, not a platform feature you can wait for. The original numbers are worth reading in full: Big Technology and The Verge.

Our approach at Azrty is instrumentation before content. Run a fixed prompt panel through one gateway so the same money terms hit ChatGPT, Gemini, Perplexity and Copilot under one budget and one audit trail: FastLLM Proxy is an OpenAI-compatible gateway in front of your own model servers and 80 hosted providers, with routing, budgets and access control in one place. Version the prompt set in PromptForge, our shared prompt library, and write the monthly results back into Kryton, our self-hosted knowledge base with wiki links and a knowledge graph, so this month's run is directly comparable with last month's. Our team in the UAE does the deployment and the support.