ChatGPT and Perplexity: Their Local Search Strategies

ChatGPT and Perplexity: A Distinct Approach to Local Search

How do artificial intelligences like ChatGPT and Perplexity choose sources to answer questions such as "what is the best insurance in Paris"? An exclusive study presented at SMX Paris 2026 by Idriss Khouader, co-founder of Meteoria, and Thibault Renouf, co-CEO of Partoo, analyzed approximately 15,000 geolocated prompts covering about twenty sectors and 200 French cities to provide quantified answers to this question.

Two AIs, Two Methods for Querying the Web

When users pose questions to ChatGPT or Perplexity, these language models do not simply draw from their pre-existing databases. They generate queries, called query fan-outs, which they send to search engines to obtain real-time sources. According to Idriss Khouader, this intermediate step is crucial for determining which sources will be analyzed and which brands will appear in the final response. The study presented at SMX Paris reveals significant differences between the two AIs at each stage of this process.

ChatGPT: Long and Bilingual Queries, Perplexity: Short and Stable Queries

ChatGPT systematically generates queries in the language of the prompt as well as in English, regardless of the country of origin of the question.

• "If I submit a prompt in Spanish, it will do the same: a query in Spanish and a query in English. This means that to influence ChatGPT's responses, it is necessary to have content in English, as half of the retrieved sources will be in English," explained Idriss Khouader.

These queries are on average long, with 11.5 words, and very precise.

Perplexity, in its free version, takes a different approach: a single search per prompt, in the original language, with short queries averaging about 5 words, very close to the initial prompt.

ChatGPT Explores, Perplexity Repeats

To measure the persistence of AI behaviors, the study submitted the same prompt 100 times to each AI. The results show that ChatGPT explores extensively. For a prompt like "what is the best insurance in Paris in the 10th arrondissement," 63 out of 67 query fan-outs retrieved were unique, exploring variations around insurance, comparison sites, reviews, and specific options. Perplexity, on the other hand, produces nearly the same query with each submission ("best insurance Paris 10th 2026").

This difference in behavior directly translates to the sources used. ChatGPT uses an average of 22.7 sources per response, and over 100 submissions of the same prompt, the total corpus reaches 132 unique URLs.

• "That's huge. This means that to maximize the influence on ChatGPT's response to a prompt, you need to influence these 132 sources," emphasized Idriss Khouader.

Perplexity mobilizes fewer sources per response (18 on average) and presents a much more stable corpus: only 40 unique URLs across 100 submissions.

Implications for Local Visibility

These operational differences have direct implications for local marketing and SEO professionals. With ChatGPT, the variability of sources means that a brand may appear in one response and disappear in the next. The challenge is to be present on as many pages likely to be scraped. With Perplexity, given its more restricted and stable corpus, the challenge is to be positioned on the few recurring sources.

The study also identified three categories of sources favored by the AIs for local queries:

• Store locators from brands (pages like "agency + city"), which constitute the primary source of data used by AIs for geolocated queries.

• Directories (Mappy, 118 000, Petit Futé, etc.), which Thibault Renouf describes as "the comeback of small directories: we were told for years 'do people really go to 118 000 to find a point of sale?' But now, AIs are going there."

• Comparators, favored by LLMs because they concentrate many brands on a single page, reducing the processing cost compared to scraping multiple individual sites.

To measure the effectiveness of their visibility in AI responses, three indicators were presented:

• Visibility rate: frequency of the brand's appearance across 30 submissions of the same prompt.

• Average position in the response: the earlier a brand is mentioned, the more it is remembered.

• Source rate: percentage of times a page from the site has been used as a data source.

These metrics, still new, complement traditional SEO indicators in a rapidly evolving search landscape.