Key Takeaways
- The common mechanism: a language model (LLM) that, to generate a response, combines what it has learned during training with a real-time search of a web index, and then writes a summary citing sources.
- Indexes: Bing for ChatGPT and Copilot, Google for Gemini and AI previews, and a comprehensive, customized index for Perplexity.
- The differences: number and visibility of sources, importance of freshness, brand positioning, share of use.
- For a website: Rank well on Google and Bing, be indexable, up-to-date, and consistent across third-party platforms.
AI assistants have become entry points for searching, comparing, and choosing. To be visible on them, you need to understand what they do when you ask them a question: they don’t “know” the web—they consult it. This article explains the common mechanism, compares the five most widely used search engines in France in 2026, and draws practical conclusions for a website. The resulting strategy is described in GEO Marketing.
What Happens When You Ask a Question
- Comprehension: The language model (LLM) analyzes the question and its context (previous conversation, location if permitted).
- Decision to search: For factual, recent, or local questions, the assistant runs a search; for general questions, he or she can answer off the top of their head.
- Requests: It rephrases the question into one or more standard search queries, which it sends to a search engine (Bing, Google, or its own).
- Reading: It retrieves the first relevant pages (5 to 20), reads them, and extracts the useful passages.
- Summary: He writes a response based on these passages and his own knowledge, attributing the statements to their sources.
- Display: The answer, including links or source references, and sometimes follow-up questions.
This has two consequences: a page that does not appear among the top results for the reformulated query is not read; a page that is read but whose content cannot be properly extracted is not cited. Details on citations can be found in Enriched Citation in AI.
A Comparison of the Five Engines
| Engine | Publisher, template | Search Index | Sources Displayed | Special Features | Usage in France, 2026 |
|---|---|---|---|---|---|
| ChatGPT (with search) | OpenAI, GPT-5, and Beyond | Bing, supplemented by OAI-SearchBot | Numbers in the text, list of sources at the end of the response | Most frequently used; readily recommends brands; user history; Shopping | The first assistant, far ahead |
| Perplexity | Perplexity: Several models to choose from | In-house index + Bing | Source lists at the top, numbers in the text, 5 to 10 sources | Most like a "search engine"; very up-to-date; cites many sources | Advanced users, professionals |
| Gemini | Links below the answer; sometimes upon request | Integrated with Android, Workspace, and Chrome; cited less frequently | Experiencing Strong Growth Through the Google Ecosystem | ||
| Google AI Insights and AI Mode | Google, Gemini | Source citations in the margins, links by paragraph | On the results page itself; the largest actual audience | As seen by the majority of Internet users | |
| Co-pilot | Microsoft, OpenAI models | Bing | Numbered links | Integrated with Windows, Edge, and Office; B2B | Businesses, Microsoft users |
Other assistants include: Claude (Anthropic, web search, professional audience), Le Chat de Mistral (French, growing), Meta AI (in WhatsApp and Instagram), and Grok (X). Their mechanisms are similar. The question of choosing between AI and Google is addressed in Should we target AI companies rather than Google?.
What these engines have in common
- They are based on existing indexes: No consumer search engine has a comprehensive index that is independent of Google and Bing. SEO remains the gateway.
- They prefer recent, accurate, and well-organized sources: a visible date, numbers, tables, and a direct answer under a question-form heading.
- They cross several sources Before recommending a brand: websites, reviews, comparisons, media, online communities (Reddit, forums), Wikipedia; see Sources of Generative AI.
- They vary: The same question yields different answers from one run to the next; the measurement requires repeated readings.
- They read HTML without executing the JavaScript, using identifiable bots; see AI-powered indexing.
What Sets Them Apart, and What That Means for You
| Difference | Practical Implications |
|---|---|
| ChatGPT and Copilot use Bing | Check indexing and rankings in Bing (Bing Webmaster Tools)—a task that is often overlooked |
| Perplexity cites a lot and quickly | The first search engine where a new, well-structured page appears; a good testing ground |
| Gemini and AI previews read Google | Traditional SEO efforts have a direct impact |
| ChatGPT recommends brands based on multiple sources | Focus on building a presence in comparison sites, reviews, and the media—not just on the website |
| AI previews are on the Google page | The greatest impact on clicks; track this using a ranking tool that detects the preview |
What this means for a website
- Be indexed and rank highly on Google and Bing for queries that voice assistants rephrase (your customers' questions).
- Write pages in which each section answers a question, providing a direct answer, facts, and tables; see Structuring Content for LLMs.
- Make the entity clear and consistent: who you are, what you do, for whom, where, with what evidence—presented the same way everywhere.
- Be present on platforms that AI systems scan: reviews, directories, LinkedIn, YouTube, and comparison sites.
- Measure by engine, using a fixed list of questions; see GEO's KPIs.
How GreenRed Can Help You
Instead of juggling multiple tools, GreenRed's GEO and AI module brings these metrics together in a single dashboard, compares them over time, and shows you which actions to prioritize. You can try it for free—no credit card required—on the page Rates.
Frequently Asked Questions
Does ChatGPT have its own search engine?
Partially. ChatGPT relies primarily on Bing’s index, supplemented by its own OAI-SearchBot, which crawls certain pages. It does not have a complete, independent index. Therefore, being indexed and well-ranked in Bing is a prerequisite for being cited by ChatGPT.
Why do AI responses vary from one time to the next?
Language models are probabilistic: the same question yields different phrasings and, at times, different sources. The indices also change continuously. To measure visibility, you need to repeat each question several times and calculate averages, as the GEO monitoring tools do.
Which AI engine is the most widely used in France?
ChatGPT is by far the most widely used assistant; however, Google’s AI snippets reach the largest audience since they appear in the search results seen by the majority of internet users. Gemini is gaining ground on Android and Chrome, while Perplexity remains the tool of choice for advanced users, and Copilot is the go-to tool for Microsoft businesses.
Should we optimize differently for each search engine?
The foundation is the same: structured content, facts, a coherent entity, presence on third-party platforms, and indexing in Google and Bing. The differences are minor: check Bing for ChatGPT and Copilot, test first on Perplexity, which responds quickly, and track AI-generated snippets using a ranking tool. A single strategy, measured by search engine.