The 20-second version
- The fact: on June 30, 2026, Semrush and Kevin Indig published a study on ChatGPT's reasoning modes and the sources it cites.
- The key number: only 25.6% of cited domains overlap between minimal reasoning (Instant) and high reasoning (Thinking). Nearly 3 sources in 4 change.
- Why it matters: depending on which mode the user triggers, your brand may be cited — or entirely absent — for the exact same question.
- What to do: optimize for both modes, not one, and cover the sub-questions of a buyer journey in depth.
On June 30, 2026, Semrush published a study with analyst Kevin Indig showing that ChatGPT's reasoning mode almost entirely changes the sources it cites, according to the analysis published on the Semrush blog. The experiment tested 100 prompts, each run twice (200 responses), across 20 buyer journeys in four categories. The core finding is blunt: only 25.6% of cited domains overlap between minimal reasoning, which produces fast Instant-style answers, and high reasoning — the "Thinking" mode.
Put another way: for the same question asked of ChatGPT, nearly three cited sources in four change depending on whether the model "thinks" or not. This isn't a marginal variation. It's two different visibility surfaces behind a single interface.
What the numbers show
Thinking mode doesn't just answer more slowly — it casts a much wider net before writing. Across the 100 prompts tested, high reasoning fired 1,130 internal web searches versus 245 in minimal mode, roughly 4.6x more sub-queries. As a result, it cites more often and cites more sources.
The second lesson is about the nature of the sources chosen. Switching to Thinking mode flips the mix: user-generated content recedes, institutional sources advance.
| Source type | Instant mode | Thinking mode |
|---|---|---|
| 15% | 7% | |
| User-generated content / reviews | 14.3% | 6% |
| Government & academic sources | 1.9% | 8.8% |
| Official documentation | 12.4% | 17.5% |
In short, the more ChatGPT thinks, the further it moves away from forums and reviews toward authoritative documents. That's consistent with what we already saw in how AI Overviews select their citations: generative engines reward structured, verifiable sources.
Why this matters for small businesses
The practical problem fits in one line from the study, relayed by Danny Goodwin in Search Engine Land on July 1: your content "may appear in fast ChatGPT answers but disappear when users ask more complex questions." And it's often the complex, late-stage questions that precede a purchase.
The gap varies by sector. The citation-rate gain between Instant and Thinking reaches +28 points in finance, +24 in health/lifestyle, +16 in B2B SaaS and just +4 in consumer tech. If you're a service SMB or a B2B publisher, most of your visibility plays out in the mode you don't control: the one the user triggers when they ask their real, decisive question.
The GEO angle. Thinking mode also follows the thread of a conversation: in the study, full-funnel brand continuity appeared only 4 times out of 20 under high reasoning, and never in minimal mode. Being cited early, on discovery questions, therefore raises your odds of staying present through to the decision. That's exactly the logic of being citable by an AI rather than simply well ranked.
What to do now
- Optimize for both modes, not one. A page that wins in Instant doesn't automatically win in Thinking. Test your key queries in both settings and note who gets cited.
- Cover the depth of the journey. Thinking mode breaks a question into dozens of sub-queries. Address the peripheral questions (comparisons, criteria, edge cases), not just the main query.
- Strengthen trustworthy third-party references. Documentation, official sources, mentions in serious media: these are what gain ground when the model reasons.
- Measure citability, mode by mode, inside a structured GEO strategy, rather than tracking one single traffic number.
Our take
This study confirms a shift we could feel coming: generative visibility isn't a single ranking, it's a mosaic of surfaces that depend on the setting, the sector, and the moment in the conversation. Optimizing for "ChatGPT" no longer makes sense: you have to optimize for a ChatGPT that, depending on whether it thinks or not, doesn't talk about the same sources. For a small business, the right read isn't to chase every mode, but to produce content solid and complete enough to be picked up regardless of the mode. Depth becomes a competitive advantage again.
Sources
- → Semrush & Kevin Indig: study "Only 25% of cited sources overlap between ChatGPT's reasoning modes," June 30, 2026.
- → Search Engine Land (Danny Goodwin): "ChatGPT Thinking mode changes which brands get cited," July 1, 2026.
- → Growth Memo, Kevin Indig: detailed methodology (100 prompts × 2, 20 buyer journeys, 4 categories).
Frequently asked questions
What did the Semrush + Kevin Indig study find about ChatGPT reasoning modes?
Why does Thinking mode cite different sources?
What should brands do about this shift?
What this article does not cover
We analyze here the SEO/GEO implications of the Semrush study for publishers and small businesses. This article is not an independent replication of the experiment, nor an audit of its methodology: the figures cited come from the analysis published by Semrush and Kevin Indig, on a sample of 100 prompts and 20 buyer journeys, whose results may vary by sector, language, and the evolution of OpenAI's models. Always check your own queries in both modes before drawing conclusions.
Growth and SEO content strategist, I founded Cicéro to help businesses build lasting organic visibility — on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.
LinkedIn