Chad Hetherington

ChatGPT’s nearly 1 billion weekly active users are a pretty big tell that search has diversified beyond Google and traditional search engine results pages (SERPs).

Back when SEO specialists were dialed into single-target-keyword strategies, they were optimizing websites and content for one phrase and therefore one SERP — where almost everyone saw the same results, depending on region. But because LLMs and AI-powered search platforms are more dynamic, that’s changing.

Cited sources between ChatGPT’s different reasoning models hardly overlap, meaning users see different results and citations for the exact same query depending on which reasoning mode they’re using. Cited domains across ChatGPT’s reasoning modes only overlap ~25% of the time, according to new data from Semrush.

Let’s look at the numbers and what they mean for marketers and SEOs navigating AI visibility in their search strategies.

ChatGPT’s Citations Vary Significantly Between Reasoning Modes

Semrush analyzed hundreds of prompts across ChatGPT’s Instant and Thinking modes to understand how reasoning affects the sources AI cites. The results suggest that Thinking mode often follows a completely different research path compared to Instant mode for the same queries.

The domains cited by Instant and Thinking mode overlap just 25.6% of the time. Or, to put it differently, roughly three-quarters of the websites ChatGPT references during a search change depending on how much reasoning it applies.

That has important implications for brands measuring AI visibility, since visibility in one ChatGPT experience doesn’t necessarily translate to another. That’s on top of the fact that most tracking tools don’t distinguish between reasoning modes, but rather show general metrics for ChatGPT as a whole.

This reinforces an important point, though: AI visibility isn’t a single ranking metric to optimize for. It’s becoming an ecosystem of experiences influenced by user intent, reasoning depth and the sources AI chooses to trust.

Thinking Mode Conducts More Research

The citation differences make more sense when you look at how Thinking mode arrives at its answers.

Compared to Instant mode, Thinking mode performed roughly 4.6 times more web searches during the study. Rather than issuing one broad search, it frequently broke prompts into multiple smaller research tasks before generating a response.

For example, a prompt asking for the best CRM for a mid-sized B2B business might trigger searches related to:

  • Pricing.
  • Features.
  • Integrations.
  • Security.
  • Customer reviews.
  • Documentation.
  • Customer support.

Each of those research steps creates additional opportunities for brands to be discovered.

The takeaway here is that ChatGPT increasingly behaves less like a traditional search engine retrieving a single result and more like a researcher gathering evidence before presenting a recommendation.

Authoritative Sources Become More Important

The study also found that Thinking mode changes which sources it prefers.

Compared to Instant mode, citations from Reddit and other user-generated content sites declined noticeably, while government websites, academic resources and official documentation became more common.

That doesn’t necessarily mean community discussions no longer matter. Reddit and review sites still influence AI systems, particularly during product research. But when ChatGPT spends more time reasoning through a problem, it appears more likely to validate information against authoritative primary sources.

For marketers, this reinforces several best practices that extend beyond traditional SEO:

  • Publish comprehensive, accurate product/service documentation.
  • Demonstrate expertise with original research and educational content.
  • Earn mentions from credible industry publications and organizations.

These types of assets provide the depth and authority higher-reasoning modes appear to value.

Complex Buying Decisions Generate More Citations

The differences become even more pronounced for comparison and purchase-related queries.

Thinking mode generated substantially more searches and citations when evaluating products or vendors than Instant mode, Semrush found. Rather than relying on a handful of sources, it assembled information from a much broader range of websites before making a recommendation.

Today’s buying journeys already involve prospects researching products across Google, ChatGPT, Gemini, Perplexity and countless other sources. If AI systems are now conducting dozens of their own research steps before presenting recommendations, brands need visibility throughout that information ecosystem rather than on just a handful of webpages.

AI Visibility Is Becoming Less Predictable

Historically, if your page ranked first for an important keyword, most users searching that keyword would encounter roughly the same results.

While Google personalized search results based on factors like location and search history, the core SERP remained relatively stable. One older study from 2017 found that only about 12% of results changed due to personalization, with the top-ranked organic results being the least likely to vary.

Large language models don’t work that way. AI search is fundamentally generative, so even the underlying sources can change dramatically from one response to the next based on model, reasoning mode, conversation history and the evidence gathered during generation — or some combination thereof.

So, instead of asking, “Do we rank for this query?” marketers increasingly need to ask:

  • Are we cited across multiple AI platforms?
  • Are we earning mentions from authoritative third-party sources?
  • Does our content comprehensively answer the questions AI systems are researching?
  • Are we visible throughout the entire buyer journey, not just at the final decision stage or early awareness stage?

The Bigger Picture

These findings don’t suggest SEO is becoming less important. If anything, they reinforce many of the same principles that have driven long-term search success: publish helpful content, build topical authority, earn trustworthy backlinks and create resources people genuinely reference and find helpful.

How AI systems consume information, especially between different models and modes, will likely continue to change. And who knows what the future truly holds. But for now, we know that success depends less on winning a single ranking and more on consistently establishing your brand as a credible source wherever AI and your target audience go looking for answers.

Note: This article was originally published on contentmarketing.ai.