Does 50% prompt share contradict ChatGPT owning about 95% of AI referrals?

No. Prompt share measures where users ask questions, while referral share measures identifiable website visits generated by AI platforms. An engine can lose share of overall prompt activity while still producing a much larger share of measurable outbound referral traffic if users on other platforms click external links less frequently.

Why does Claude's prompt share look high versus conversations?

Comscore reported that Claude users generated about 9.1 prompts per conversation in June 2026 compared with approximately 5.4 for ChatGPT. Longer or more iterative sessions increase an engine's share of total prompts relative to its share of conversations.

ChatGPT Prompt Share Fell to 50%: Comscore Q2 for B2B SaaS

Q: What did Comscore claim about ChatGPT?

Comscore reported that ChatGPT's share of U.S. AI prompt volume fell from 70% in January 2026 to 50% in June 2026, while Gemini rose to 30% and Claude to 11%. Comscore also reported that ChatGPT remained the category leader on conversation volume and multiplatform visitors.

Q: Should we drop ChatGPT-focused GEO?

No. ChatGPT remains a major source of AI usage and measurable referral traffic, so it should remain part of a multi-engine visibility program. Teams should add dedicated monitoring for Gemini, Claude, and other relevant engines rather than shifting from one single-engine strategy to another.

Q: How should reporting change this week?

Separate AI reporting into distinct metrics: prompt presence by engine, AI referral sessions, citation and recommendation share, and Google Search Console generative AI impressions. Avoid combining them into one blended AI visibility score because each metric represents a different stage of discovery and buyer behavior.

Written byAyush SharmaVP, SEO & AI Search, DerivateX

Reviewed byPawan BhargavSr. Content Writer, DerivateX

Published Sep 29, 2026Updated Sep 28, 2026

6 min read

TL;DR

On September 22, 2026, Comscore released its Q2 2026 AI Intelligence Report: US ChatGPT prompt-volume share fell from 70% (January 2026) to 50% (June 2026).
In the same window, Gemini rose 17% → 30% and Claude 2% → 11% (more than fivefold). ChatGPT still logged 168 million desktop conversations in June and 99 million multiplatform visitors (up from 59 million a year earlier).
Google desktop searches with an AI Overview rose from 25.8% (July 2025) to 39.4% (June 2026). Claude averaged 9.1 prompts per conversation vs ChatGPT 5.4, which helps explain why Claude’s prompt share outruns its conversation count (ppc.land deep dive of the same report).
Pair this with DerivateX’s Sep 25 BrightEdge angle: ChatGPT still owns ~95.1% of measurable AI referral traffic in August 2026. Both can be true. Prompt share ≠ referral share ≠ citation or recommendation share.
Comscore sells AI Intelligence measurement products, so treat the series as a vendor panel with a product stake, then replicate on your own engines. Start with a free AI visibility audit if you need prompt, referral, and shortlist lines separated before you reallocate.

ChatGPT can still send most of the AI clicks you can see in GA4 and still lose twenty points of the prompts your buyers type.

That is the useful tension between Comscore’s Q2 prompt pie and BrightEdge’s August referral concentration. One measures what people ask. The other measures what hits your site with a clean AI referrer. B2B SaaS pipeline care about both, plus whether your brand is named inside the answer.

If your board slide still says “we optimized for ChatGPT,” this report is the check that Gemini and Claude are now material prompt destinations, especially for longer enterprise-style sessions.

What Comscore measured

From the press release and report download page, the public panel findings that matter for GEO:

Prompt-volume share (Jan → Jun 2026): ChatGPT 70% → 50%, Gemini 17% → 30%, Claude 2% → 11%. Copilot held near 5%. Grok and Perplexity lost ground on the rounded pie (ppc.land reconstructs the seven-assistant split).
Category leadership without monopoly: ChatGPT reached 168 million desktop conversations in June 2026 and grew multiplatform visitation to 99 million from 59 million year over year, while its share of prompts fell as usage spread.
Claude intensity: Desktop conversations climbed from ~1.1 million (Oct 2025) to a spring high near 40 million, then 22.3 million in June (still ~20× October). Average prompts per conversation in June: Claude 9.1, Copilot 6.9, Gemini 6.2, ChatGPT 5.4.
Reach still expanding: In June 2026, 35% of desktop users and 29% of mobile users visited an AI assistant (vs 25% and 12% in January 2025).
AI Overviews as default search furniture: Google desktop searches with an AI Overview 25.8% → 39.4% (Jul 2025 → Jun 2026). Bing Copilot Search 11.8% → 17.3% over the same span.

Smriti Sharma, Comscore’s Chief Analytics and Technology Officer, put the commercial read this way: “As competition among AI assistants intensifies and prompt volume shifts across platforms, brands and publishers can no longer assume that a single AI experience defines the market.”

What the report measures vs what it misses

Useful for operators

A US panel view of where prompt volume sits, not just unique visitors.
Conversation and prompts-per-conversation context that explains Claude’s share jump.
Parallel context on AI Overviews coverage and early sponsored placements inside ChatGPT shopping/travel chats.

Caveats (do not skip)

Comscore sells AI Intelligence products. Vendor panels have a stake in making AI measurement feel urgent. Replicate on your own properties before a board claim.
Prompt pies appear built heavily on the US desktop panel. Mobile reach is rising faster, but the headline share series is not a clean global multi-device census.
ppc.land documents caption and release slips (growth rates vs starting levels, “plateau” language vs a ~45% June dip from Claude’s April peak, mismatched source footnotes). The direction of fragmentation holds; treat single-sentence PR framing as secondary to the charts.
Prompt share is not citation share, recommendation share, or referral share. A Claude shortlist with no click never appears in GA4.

Prompt share vs referral share (why both headlines can be true)

Metric	What the Sep 22–25 window says	What it means for pipeline
Prompt-volume share (Comscore Q2)	ChatGPT ~50% of prompts by June; Gemini 30%; Claude 11%	Where buyers type. Multi-engine prompt frames are no longer optional.
AI referral share (BrightEdge Aug, DerivateX Sep 25 piece)	ChatGPT ~95.1% of measurable AI → website sessions	Where analytics sees exits. ChatGPT still owns the clean referrer trail.
Citation / recommendation share	Not this Comscore headline; engines diverge hard (~9% citation overlap, rec vs cite)	Whether you are named or linked inside the answer.

Weak read: “ChatGPT is dying, move all GEO to Gemini.”

Better read: ChatGPT remains the volume and referral floor. Gemini and Claude now take enough prompts that ChatGPT-only content and corroboration plans under-serve B2B shortlists, especially on Claude’s longer sessions.

Operator playbook / worked example

Layer 1: Split the scoreboard

Report three lines weekly, never one blended “AI visibility” number:

Prompt presence: named / recommended / cited on a fixed buyer-prompt set in ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews / AI Mode.
Referral sessions: GA4 (and CRM) by AI source, ChatGPT-heavy for arrivals.
Google Gen AI impressions: GSC generative surfaces, separate from chat referrals.

Layer 2: Weight Claude for B2B enterprise prompts

Claude’s 9.1 prompts per conversation and fivefold share rise matter for long evaluation chats (security questionnaires, integration fit, “best X for Series B”). Keep a fixed Claude frame even when Claude referral share is tiny.

Layer 3: Do not ChatGPT-only your GEO stack

If Gemini now holds ~30% of prompt volume in Comscore’s US panel, category pages, comparison truth, and third-party corroboration that only “work” in ChatGPT leave a third of the typed demand under-covered.

Layer 4: Worked example (mid-market RevOps SaaS)

Priority prompts: “best RevOps platform for Series B,” “HubSpot RevOps alternatives with Salesforce sync,” “pipeline hygiene software for B2B.”

Confirm ChatGPT still drives most AI referral sessions (many stacks will).
Run the same three prompts weekly in ChatGPT, Gemini, and Claude. Log vendor slots and owned / influenced citations separately.
Put incremental corroboration where Claude or Gemini name competitors and skip you, even if those engines send few sessions.
Put conversion hardening on the ChatGPT exit path when arrival is the constraint.
Use AI Overview coverage as a Google surface metric, not as a proxy for chat prompt share.

That turns a 47-page vendor report into a measurement queue instead of a panic reorg.

How DerivateX thinks about it

Treat Comscore as a prompt-fragmentation signal with a measurement-product COI, then verify engine by engine on your category.
Keep prompt share, referral share, citation share, and recommendation share as separate lines. BrightEdge’s referral concentration and Comscore’s prompt pie are complementary, not contradictory.
Do not let ChatGPT’s referral monopoly erase Gemini/Claude shortlist work. Do not let Claude’s prompt-share rise erase ChatGPT conversion work.
Budget follows the constraint: missing names on Claude/Gemini → corroboration and entity work; weak ChatGPT exits → first-screen answer completion and demo path.

If you want a baseline that separates those lines on your category, request a free AI visibility audit. Engagement options stay on pricing. Adjacent reading: AI citation overlap, recommendation vs citation share, and case studies where multi-engine visibility showed up in CRM.

FAQ

What did Comscore claim about ChatGPT?

That ChatGPT’s share of US AI prompt volume fell from 70% in January 2026 to 50% in June 2026, while Gemini rose to 30% and Claude to 11%, even as ChatGPT remained the category leader on conversations and multiplatform visitors.

Is this independent research?

No. Comscore sells AI Intelligence and related measurement products. Use the series as a directional panel signal and replicate with your own prompt frames and analytics.

Does 50% prompt share contradict ChatGPT owning ~95% of AI referrals?

No. Prompt share is where questions go. Referral share is where measurable clicks to websites come from. ChatGPT can lose prompt share and still dominate clean referrers if other engines convert less often to outbound sessions.

Why does Claude’s prompt share look high vs conversations?

Because Claude averaged about 9.1 prompts per conversation in June vs ChatGPT’s 5.4 in Comscore’s data. Longer sessions lift prompt share relative to conversation count.

Should we drop ChatGPT-focused GEO?

No. ChatGPT remains the volume and referral floor. Add Gemini and Claude frames; do not replace ChatGPT work with a single-engine pivot.

How should reporting change this week?

Split prompt presence, AI referral sessions, citation/recommendation share, and GSC Gen AI impressions. Stop one blended AI score from running budget.