# AI search optimization starts with an honest audit Source: https://www.smartecomseo.com/geo-agency/ai-visibility-audit/ Updated: 2026-09-29 Before anyone promises you ChatGPT visibility, you should know where you stand. We run the questions your buyers actually ask across every major assistant, record every answer with a date, and show you who gets named, what is said about you and why. ## 50 to 100 buyer prompts, across six AI surfaces We test ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode and Microsoft Copilot. Prompts are written from your keyword data, customer questions and sales team, grouped into six types. ### Category "best" prompts "Best outdoor pizza oven for a small patio." The prompts where shortlists are made. ### Problem-led prompts "How do I keep a greenhouse warm in winter without gas?" Questions that lead to a product. ### Comparison prompts "Brand A vs brand B" and "slate vs MDF pool table". Where buyers narrow down. ### Price-bound prompts "Best e-bike under $2,000 for hills." Budget filters that decide who is even considered. ### Brand prompts "Is [brand] legit?", "Where is [brand] made?", "[brand] warranty". How assistants describe you. ### Alternative prompts "Alternatives to [competitor]." Whether you appear when buyers look past the leader. ## Every answer, logged the same way | | What we record | Why it matters | |---|---|---| | Mentioned | Is your brand named at all? | The basic measure of visibility | | Cited | Is a page on your domain linked as a source? | Citations can send traffic; mentions alone rarely do | | Position | First recommendation, one of several, or a footnote | Buyers read the first names most | | Description | What the assistant says about you, word for word | Wrong facts about price, origin or warranty cost sales | | Competitors | Which brands are named instead of or beside you | The basis for share of voice | | Sources | Which domains the assistant cited | Shows where mentions would help most | | Evidence | Dated screenshot and full answer text | So the baseline can be checked later | ## Two weeks from prompts to fix list 1. **Build the prompt set** (Days 1 to 3): From your Search Console queries, keyword research, reviews, support tickets and sales questions. You approve the list before we run it. 2. **Run and record** (Days 3 to 8): Each prompt run on each assistant in clean sessions for your target market, more than once where answers vary, with screenshots and full text saved. 3. **Score share of voice** (Days 8 to 10): Mentions and citations for you and each competitor, by assistant and by prompt type. 4. **Check the inputs** (Days 5 to 10): Crawler access, indexing, schema, Shopify agentic storefront settings, product data completeness and brand fact consistency. 5. **Write the fix list** (Days 10 to 14): Every finding tied to a fix, ranked by likely impact and effort, with a 30, 60 and 90-day plan. ## Audit deliverables ### Prompt log Every prompt, assistant, date and answer in a shared Google Sheet. ### Screenshot archive Dated screenshots for every answer, so the baseline can be checked later. ### Share of voice report You against your top competitors, by assistant and prompt type. ### Brand description review What assistants say about you, with every inaccurate fact flagged. ### Citation source map The domains assistants rely on in your category, and whether you appear there. ### Prioritized fix list Technical, catalog, content, entity and mention fixes, in order. ## Why the audit is a sample, and why that is fine AI assistants do not give everyone the same answer. Results change with location, language, account history, memory and the day you ask. So an audit is a sample, not a census, and we say so in the report. What makes the sample useful is consistency. We use the same prompts, the same markets and the same recording method every time. That lets us compare your brand against competitors on equal terms today, and compare you against yourself when we re-run the set in three or six months. ### The fixes we usually find - **Access.** A bot-protection app or robots rule blocking an AI crawler, more often than you would expect. - **Wrong facts.** An assistant repeating an old price, a discontinued model or the wrong country of origin, usually traced to an outdated page or listing. - **Missing subtopics.** Competitors cited for questions you never answered on your site. - **Thin product data.** Products in Shopify Catalog with too little detail to match buyer questions. - **No outside voice.** Competitors appearing in review sites and forum threads where you are absent. ### For Chinese brands Assistants often describe Chinese brands vaguely or confuse them with the manufacturer's company name. Part of the audit checks exactly how you are described, and whether the facts buyers care about, such as warranty, support and where orders ship from, are stated correctly. See [brand trust for Chinese brands](/chinese-brands/brand-trust/). The audit can stand alone, or become the baseline for ongoing work on [ChatGPT SEO](/geo-agency/chatgpt-seo/), [ChatGPT Shopping](/geo-agency/chatgpt-shopping/) and [AI Overviews](/geo-agency/google-ai-overviews/). How we report progress afterwards is on the [results page](/geo-agency/results/). ## AI visibility audit questions ## Get your AI baseline before your competitors do Tell us your store, your top competitors and the markets you sell to. We will scope the prompt set with you. [Request an audit](/contact/) ## Frequently asked questions ### What does an AI visibility audit include? A set of 50 to 100 buyer prompts run across ChatGPT, Perplexity, Gemini, Google AI Overviews, AI Mode and Copilot, with every answer logged and screenshotted. You get share of voice against competitors, a review of how assistants describe your brand, a map of the sources they cite, technical and catalog checks, and a prioritized fix list. ### Which AI assistants do you test? ChatGPT, Perplexity, Google Gemini, Google AI Overviews, Google AI Mode and Microsoft Copilot. Those cover most AI-assisted product research in Western markets today. If your buyers use another assistant heavily, such as Claude or a regional tool, we can add it to the prompt set. ### How do you choose the prompts? From your data first: Search Console queries, keyword research, product reviews, support tickets and the questions your sales team hears. We group them into discovery, comparison, price, brand and alternative prompts, and you approve the list before we run it, so the audit reflects how your buyers actually ask. ### Why do AI answers change when I ask the same question? Assistants personalize by location, language, account history and memory, and they retrieve fresh sources each time. The same prompt can name different brands on different days. That is why we run prompts in clean sessions for your target market, repeat prompts where answers vary, and report trends across many prompts instead of single answers. ### Can the audit be repeated to measure progress? Yes. The prompt set, markets and recording method stay fixed, so a re-run in three or six months compares like with like. Clients on ongoing GEO work get a smaller tracked prompt set re-run monthly, with dated screenshots, alongside AI referral sessions and orders from Shopify.