Search traffic is only part of the story now. If an AI assistant answers for your category before sending a visitor to your site, you need to know whether it cites you, what it says, and how often that happens.
Why this belongs in SEO
Traditional SEO asks: can Google find my page, and does it rank? AEO asks: can an assistant answer the user's question? GEO asks: when the answer is generated, does it choose my site as a source? These questions overlap, but they are not identical. A page can rank in Google and still be absent from ChatGPT, Perplexity, Gemini, or Claude. Another page can be cited by one assistant and ignored by the others because each system can use different retrieval, ranking, source selection, and response generation steps.
For a founder or marketing lead seeing search traffic fall, the practical issue is not whether AI is replacing SEO overnight. It is whether a new decision point has appeared between intent and click. If the assistant gives a direct answer and names competitors, your brand may disappear before the user reaches the results page.
The five-step manual test
Start with the query a real customer would type. Do not test a broad brand query first. Use a category and location query such as 'best accounting software for freelancers in Toronto' or 'emergency plumber in Austin'. The more specific the query, the more useful the result.
Test ChatGPT first.
Open ChatGPT and ask the exact query. Read the full answer, not only the final recommendation. Search inside the response for your company name, product name, domain, and category terms. A citation buried in a sentence is different from a link at the bottom, but both matter.Repeat the same query on Perplexity.
Perplexity often shows sources and search context. Check whether your page is cited in the answer, in the sources list, or not at all. If a competitor appears, note which claim helped it: price, locality, reviews, documentation, or a specific use case.Repeat on Claude and Gemini.
Use the same wording. Do not summarize the first answer and ask a new question, because small wording changes can change retrieval. Record each result separately. Being cited by ChatGPT says little about Claude; being absent from Gemini does not prove you are invisible everywhere.Check Google and AI Overviews.
Run the same query in Google. If an AI Overview appears, read it as you would read a featured snippet. Look for your brand, your domain, and the competitor that replaced you. Then check Google Search Console to confirm that the page you expect to be cited is indexed. A live AI search cannot cite a page that the search index does not know.Write down the result.
Create a simple table with date, engine, query, cited, absent, competitor cited, and short note about what the assistant said. Include the URL of the cited page. Without this trace, the next check is not comparable.
One check is a photo; a series is a film
A single check answers only one narrow question: were you cited on this date, with this query, on this engine? It does not show a trend. Generative systems can change because the web changed, a competitor published a stronger page, a source became more accessible, or the model selected a different set of references.
Run the same test for several weeks. The goal is to measure frequency: how often you appear, how often a competitor appears instead, and whether a page update improves citation over time. For example, if you add a clear service page for 'same-day appliance repair in Denver' and include useful details such as service area, hours, pricing range, FAQs, and structured data, you might see no change in week one. In week three, ChatGPT may start citing the page for that query while Perplexity still prefers an established local directory. That pattern tells you where to focus.
Track four states: cited, absent, competitor cited, and mixed. Mixed means the assistant mentions your brand but links to a different page, or cites you for one claim and a competitor for another. This is useful because it shows whether the problem is discovery, authority, relevance, or content coverage.
What to do with the data
Do not treat AI citations as a vanity metric. Treat them as a map of how assistants make your category understandable. If an assistant cites your site but misstates your offer, your issue may be page clarity. If it cites a competitor for price and not for quality, your page may need stronger proof. If it cites no one, the query may be too broad or your page may not answer the exact intent.
Use the same query set you already use for SEO: product, category, local, comparison, and support queries. Review the results weekly. Compare before and after changes to titles, headings, FAQs, internal links, schema, and pages that summarize your offering. Keep the raw notes, because the useful signal is often the repeated pattern, not one surprising answer.
For e-commerce teams, the business context matters. Contentsquare reported in 2026 that average e-commerce conversion in France is between 1.5% and 3%, with mobile conversion around 2% versus 3.4% on desktop. That means a missing citation is not only an image problem. It can mean fewer visits from a channel where each conversion is already valuable, especially on mobile.
A tool such as ZenRadar can automate this repeated check across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews, then store the results over time. The manual version is enough to learn the method; the repeated version is what turns observations into a decision.
Source: https://zenradar.organizen.ai/articles/comment-savoir-si-une-ia-vous-recommande
Top comments (0)