Three AI search signals landed between 6 and 13 July 2026. GPT-5.6 shipped, Search Engine Land published a 6.77M-session study saying ChatGPT owns 92.4% of AI referral traffic, and Kevin Indig's Growth Memo #21 showed AI Overviews and AI Mode decoupling. Two of the three, read carelessly, will push Indian brand teams to make the same allocation mistake this quarter: pour Q3 AEO/GEO budget into ChatGPT alone. Read against the June 2026 Cited Index (226 Indian brands, 5 AI engines, 760 citations) they say something different. This piece walks through each signal, what it means literally, and where the Cited data changes the read.
Signal 1: The 92.4% ChatGPT referral number and what it actually measures
On 6 July, David Bell at Search Engine Land published a clickstream study covering 6.77 million AI-driven sessions across 166 GA4 properties, running November 2024 through May 2026. The headline number: ChatGPT now accounts for 92.4% of trackable AI referral traffic, up from 84% in December 2025. The other four engines together (Perplexity, Gemini, Copilot, Claude) sit at roughly 8%. Perplexity is down 61% from its March 2025 peak. Copilot is down 96% from its August 2025 peak. Claude grew 64x from November 2024 to May 2026, but off a base of 133 sessions.
Take the study at face value on what it does measure. It measures traffic. Traffic is what happens after a user reads an AI answer, sees a source cited, decides to click, and lands on a page instrumented with GA4. That last chain has a lot of loss inside it. Most AI answers name brands without linking them, and users who get a satisfying answer often do not click through. Traffic is the tiny visible slice that survives the answer-generation surface.
Citation share is a different measurement. It is the answer to a different question: when an AI engine constructs an answer to a category prompt, which brands and URLs does it cite in the response itself? A citation without a click still counts. That is the layer where the buying journey forms, because the brand names the user reads in the answer are the shortlist they act on next, whether they click or open a new tab or go straight to Amazon.
On the June 2026 Cited Index the picture is different. Aggregate across 760 brand-level citations, ChatGPT (openai) accounts for 27.2%. Google AI Mode is the single largest engine at 35.9%. Google AI Overviews adds another 20.0%. Perplexity and Gemini share the rest at 7.8% and 9.1%.
| Engine | Share of citations (June 2026 Cited Index) | Bell / SEL trackable referral share |
|---|---|---|
| ChatGPT | 27.2% | 92.4% |
| Google AI Mode | 35.9% | (rolled into Google) |
| Google AI Overviews | 20.0% | (rolled into Google) |
| Perplexity | 7.8% | ~6.5% (of the residual) |
| Gemini | 9.1% | ~2% (of the residual) |
| Grok / Copilot / Claude | Not tested this edition | ~2% combined |
Two different measurements of the same market. Both are correct on their own terms. A brand that budgets from the 92.4% number and spends Q3 exclusively on ChatGPT-shaped citations is optimising for the surface that survives to GA4. It is under-investing in the four engines where roughly 73% of the actual brand-naming in AI answers is happening. Indian D2C is more exposed to the mismatch, not less: ChatGPT ships in India at a $4.50/month tier and referral traffic has surged, but AI Mode is rolling out into the same categories where the majority of high-intent product prompts get run.
The Yellow.ai case study we published Tuesday shows the shape of this at brand level. Yellow.ai gets cited in about 45% of ChatGPT answers in its category. On Perplexity its share is 13.8%. If the team read only their GA4 referral report, they would conclude ChatGPT is the only engine that matters and skip a Perplexity gap that is actually costing them enterprise-buyer visibility.
Signal 2: GPT-5.6, and what to watch as it rolls out
OpenAI released the GPT-5.6 family on 9 July per multiple secondary sources: Sol, Terra, and Luna, with token pricing at $5/$30, $2.50/$15, and $1/$6 per million tokens respectively. The rebrand replaces the older monotonic naming with generation-plus-tier. Profound shipped same-day tracking support. OpenAI has not published a retrieval-pipeline change note, so the citation mechanics that decide which brands ChatGPT names in a category answer are unchanged as of publish date.
That leaves two things to watch as the rollout lands.
First, default-tier substitution. If the pricing shift moves a meaningful share of ChatGPT free-tier users onto Luna, brand citation on those responses will look different. Historically the smallest-tier models have smaller context windows and lean harder on cached, higher-authority retrieval. Which means small-brand citations get squeezed harder than large-brand citations, because the low-context path favours whatever domain the model already knows.
Second, reasoning-tier asymmetry. Sol as flagship and Luna as budget will likely produce the same source-mix asymmetry that ChatGPT Thinking mode already shows versus Instant mode. Our Reddit AI visibility gap analysis documented that reasoning mode cuts Reddit's citation share by roughly half. If Sol pulls a Thinking-style source mix and Luna pulls an Instant-style mix, the category exposure changes with which tier the user is on. Categories built on forum and social-source citation (Audio & Wearables at 30% forum-plus-social, Skincare & Beauty at 28%) will read the difference the fastest.
Cited will re-run the June brand set on the GPT-5.6 family in the July Index and publish the delta. Whether the retrieval mechanics changed at all is empirically checkable in three weeks. Until then, the honest answer is: unchanged as far as OpenAI has said, wait for the data.
Signal 3: Google AIO and AI Mode are diverging, and Indian brands were never on the same curve
Want to know how your brand scores on these same metrics?
We'll run 20 prompts across 3 AI platforms and send your report within 24 hours.
On 10 July, Kevin Indig published Growth Intelligence Brief #21 with a fresh 4-week window covering 2,600 companies across 26 verticals. Three numbers frame it. Google AI Overviews mentions fell 3.2% (5.47M to 5.29M). AI Mode mentions grew 22.5% (547,193 to 670,163). ChatGPT mentions fell 28.4% (1.20M to 860,743). SEO visibility rose across all five major social platforms in the same window: Reddit up 18.1%, Facebook 20.1%, Instagram 21.4%, LinkedIn 43.3%, X 38.6%. Indig's read is that the coupling between traditional SEO signal and AI Overviews signal is coming apart.
For Indian D2C brands the split reads differently. On the June 2026 Cited Index, AI Mode is already the larger of the two Google surfaces: 35.9% of all citations vs 20.0% for AI Overviews. The two surfaces are pulling different source-type mixes. Per category:
| Category | AI Mode top source types | AI Overviews top source types |
|---|---|---|
| Skincare & Beauty | YouTube (10 of 12 YouTube citations run via AI Mode) | Brand sites (Beminimalist and Earth Rhythm at 2 AIO citations each, Plum and Foxtale at 1 each) |
| Travel & Luggage | Reddit (8 of 15 Reddit citations), YouTube (5 of 6), brand sites | Brand sites (Samsonite, Flipkart, Suitcase.repair, Assembly) |
| Audio & Wearables | Reddit (12 of 22), Amazon (6 of 25), YouTube (6 of 6) | Amazon (7 of 25), Flipkart (2), Facebook (2) |
Same category, same underlying prompt set, materially different source lists per Google surface. This is not a UI-tier variation. It is two different retrieval pipelines producing two different citation ladders.
Two implications for Indian brand teams. First, category-level Google strategy needs to split AI Mode and AI Overviews as separate targets, not lumped under "Google AI." The evidence for AI Mode citation lives in YouTube presence, Reddit brand-mention density, and, for e-commerce categories, product-page visibility on Amazon. The evidence for AI Overviews citation lives more heavily on brand-owned domains, particularly for Skincare & Beauty. Second, Indig's global signal that AIO is contracting while AI Mode expands is running ahead of India, where AI Mode has been the larger surface all quarter. Optimising only for AI Overviews is competing for the smaller and shrinking half.
Where the three signals land together
Three signals, one takeaway. GPT-5.6 is a wait-and-measure event, not a decision-forcing one until we see the July retrieval data. The 92.4% ChatGPT traffic-share number is real, but it measures a different thing than citation share, and Indian brand teams that budget from it will systematically underinvest in the Google AI surfaces that are now producing 55.9% of category-level brand citations in India. The Indig divergence is validated at the global level and is running further along in India, where AI Mode has been the larger Google surface for the whole June cycle.
The on-site content playbook we shipped Monday still holds: fix the 4% of citation surface that lands on your own domain, get author attribution, schema, and freshness right, and the citations you should have won stop leaking. But the harder Q3 decision is off-site. If your category is 55% Google-surface-cited and 27% ChatGPT-cited, budget accordingly. That is the number to plan against, not the traffic slice.
Where to go from here
If you want to see per-engine citation numbers for your own brand and category rather than the D2C aggregate above, dashboard.getcited.in/signup gets you the same tracking layer we used to build these numbers, applied to your prompt library, on the same weekly cadence.