Although promising, SearchGPT’s responses for e-commerce can still be improved, particularly because of the absence of flows.
If SearchGPT is far from being comparable to Google on the e-commerce side, the generative search engine still offers advances, compared to ChatGPT in particular. Under the hood is the GPT-4o language model. A RAG, or “Retrieval Augmented Generation” approach, for more accurate and precise answers. A more important index. Diversified sources. Updated answers. It has been available to all users for a few days.
Sobriety, structuring and duplicates
Regarding the different styles of e-commerce queries, “Search GPT uses the same sources for their processing as for other types of queries,” says Alexis Rylko, SEO technical director at iProspect France. “That is, the 12 best results from Bing as well as content provider partners, i.e. the major media sites. From these sources, SearchGPT selects those it wishes to use as citations. It is then on the latter that the tool generates a response and provides the corresponding links to the cited sources Given the typology of e-commerce queries, in the cited sources, we mainly find the pages of e-commerce sites, product comparison articles, items offers and promotions on media sites.”
This approach results in a diversity of sources on certain types of e-commerce queries studied. Note that on these, the results appear in a refined manner. They appear in the main column, with the sources cited at the end. Information is often synthesized, structured and updated. Let’s also note that 4 images of a product can appear, often at the top of these links.
Informational e-commerce queries: the most convincing “Quotes”
Informational e-commerce queries concern the search for information sometimes linked to a possible purchase. They deal with a need for opinions, or even a desire to compare, as for the queries “Lenovo PC review”, “iPhone 15 and 16 comparison”…
On these, we note, according to our tests, a paucity of sources displayed. For example, on the query “Lenovo PC review”, the site https://meilleurtest.fr/ appears in 3 of the 5 responses generated in the right column (red arrows on the image). There are also duplicates, with identical URLs. To discover new links, you must click on “Sources” at the bottom of this list of results. This allows you to arrive in the “Citations” column on the right (black arrows). The sources are then more diversified.
Navigational e-commerce queries rather stingy with sources
Navigational e-commerce queries refer to searches to access a site or a web brand, without going through the browser address bar. In the e-commerce context, this may be linked to a purchase, but not necessarily, as with “voyage sncf” or “airbnb” queries.
On the query “SNCF travel” for example, SearchGPT structures and summarizes its response. But it simply presents the SNCF platform and links to the Wikipedia, SNCF Connect, SNCF voyageurs sites. There are duplicates. The “Citation” section is hardly more exhaustive in terms of the links mentioned. It does not display, for example, links to Ouigo, or Trainsline, as Google does for this query. You have to go to the bottom of this list to find more varied links.
Transactional e-commerce queries diversify links
Transactional e-commerce queries are more purchase-oriented. They may include terms like “order”, “buy” or even “price”. SearchGPT regularly displays, according to our tests, four images of the product from a site, on this style of query. Then, always, a synthetic and updated text. The sources are relatively diversified, and direct almost exclusively to sites linked more or less directly to sales.
An interesting opportunity
E-commerce on SearchGPT therefore still seems far from matching that of Google. This contains announcements, a “Product sites” section, links to purchasing sites but also comparison sites with rich snippets, or even a “People also ask” module. The answers are more direct. “For its part, SearchGPT continues to respond to e-commerce queries as a purchasing guide,” notes Alexis Rylko. “It can point to websites and addresses where to buy a product, compare options and provide advice, but it cannot redirect directly to specific products and perform this transactional function. This is partly because Internet users do not yet consider ChatGPT as a gateway to e-commerce. OpenAI still has to carry out important work to encourage these new reflexes among users.
For him, it is above all the absence of flow that is failing with SearchGPT “The heart of e-commerce is based on the product flow, standardized, structured and up to date. Without having access to it, it is impossible to become a real engine of e-commerce search We observe this with the examples of AI Overview deployed around the world, because it is an AI added on top of the Google Shopping product feed. The real turning point will arrive when SearchGPT integrates this. flow of products, and Good foundations already exist in Microsoft Merchant Center.”
Another explanation for the general poverty of sources for e-commerce queries: the purpose of SearchGPT, more in conversation than in direct response, according to Roman Fougerolle, senior SEO consultant at Noiise. “The objective of an AI search engine is to offer a relevant response to the Internet user according to their query, without directly highlighting the sources. An AI engine wants to keep the Internet user in its interface. By creating a discussion with the latter, this allows it to improve. The future will tell us how AI search engines position themselves, particularly with advertising.”
If SearchGPT thus displays certain limits on e-commerce, note that, according to BrightEdge, the growth of OpenAI’s generative search engine would be 150% from one month to the next. Striking figures, compared to those of Perplexity.ai and Claude, which only grew by 22% over this same period.
It can therefore be interesting for the SEO to position themselves on SearchGPT. For those surveyed, the first thing to do is to be present in the Bing index. “We recommend using the Bing Webmasters Tools tool, the equivalent of Google Search Console”, breathes Roman Fougerolle. “Once the page is present in the Bing index, it is necessary to determine the factors which will influence the positioning pages in SearchGPT If the page corpus is identical between Bing and it, the relevance criteria differ. These are not yet precisely known, but several semantic optimizations can be carried out to influence an AI. According to the SEO, citations, or statistics, are “very effective”. Citing sources or producing original content deemed “effective”.
For Alexis Rylko, “among Bing search results, SearchGPT will select citations, and will particularly seek to include those coming from various sources. This is to avoid excessive dependence on a single site or content provider. Thus, if we take the query “adidas gazelle”, SearchGPT will choose one or two sources to cite among the sellers, one or two to cite among the informative articles presenting the collection, its history, particularities, or celebrities behind it. Thus, on queries. multi-intentional, offering content covering each of the needs increases the chances of being cited.”
