run an AI search visibility audit exactly like this before trying to rank in AI search for context, we got one of our customers to rank #1 in ChatGPT in only 2 weeks. here’s how (14 steps): 1. start by building a fixed set of 30–50 prompts your buyers would realistically ask before choosing a product. split them between category searches (“best [category] tools”), comparisons (“[product] vs [competitor]”), alternatives, specific use cases, pricing/cost questions and problem-based searches. don’t build the list around keywords you want to rank for. build it around actual buying decisions. 2. run the exact same prompt set through ChatGPT, Claude, Gemini and Google AI Mode. for every answer, record the model, date, companies mentioned, companies recommended and every cited URL. keep the prompts identical so you can actually compare the results. 3. separate a citation from a mention from a recommendation. if ChatGPT cites one of your blog posts while recommending three competitors, that is not the same as winning the query. track all three separately. 4. repeat your highest-value prompts more than once. AI answers are dynamic. one screenshot where you disappear means very little. you’re looking for gaps that keep showing up across multiple runs, related prompts and different engines. /5 group the prompts into topics instead of optimizing them one-by-one. “best auth API,” “SSO provider for SaaS,” “SAML authentication tools” and “SCIM providers” might all belong to the same buying topic. the goal is to understand which parts of the category you consistently own and which parts your competitors own. 6. now calculate the competitor gap for each topic. if a competitor appears in 18/25 relevant prompts and you appear in 4, start there. don’t waste time fixing a random query where both of you appeared once. 7. open every source the AI used when your competitors won. record which page was cited, which domain owns it, what type of page it is, which competitor it mentions and what claim or recommendation it appears to support. after 20–30 prompts you’ll usually start seeing the same pages and domains repeatedly. 8. before creating anything new, check whether you already have a page that should be winning that query. if the page exists, find out why it isn’t surfacing. does it actually answer the same intent? is the information outdated? is the useful information buried? does the competitor page contain proof, comparisons or data yours doesn’t? building another page can just create two weak pages instead of fixing the right one. 9. check the technical layer next. make sure the page is indexable, internally linked and returning normally to crawlers. for ChatGPT specifically, check that OAI-SearchBot isn’t blocked in robots.txt or by your CDN/WAF. for Google AI features, the page first needs to be indexed and eligible to appear in normal Google Search. 10. investigate the searches underneath the prompt. AI search doesn’t always retrieve pages using the exact sentence the user typed. Google AI Mode uses query fan-out to run related searches underneath a question. Bing Webmaster Tools now exposes the grounding queries its AI systems used to retrieve cited pages. use those related searches to understand what the engine is actually trying to find. 11. only NOW decide what needs to change. sometimes you need a new page. sometimes the right page already exists and needs better coverage. sometimes your information is outdated. sometimes the model keeps relying on third-party comparison pages where every competitor is mentioned except you. sometimes the problem is technical and writing anything would be a waste of time. 12. if you do create or update content, give the model information it cannot get from another generic article. use first-party data, real pricing, specific feature limits, screenshots, benchmarks, comparison criteria, customer results, use-case differences and actual limitations. another 2,000-word summary of information already on the internet gives the engine very little reason to retrieve you. 13. if third-party sources keep shaping the recommendation, work on those separately. find the comparison pages, industry sites, videos and discussions that repeatedly appear in the answers. figure out why competitors are represented there and you aren’t, then earn legitimate inclusion where it actually makes sense. don’t manufacture random mentions across the internet and call it GEO. 14. once the changes have been crawled, rerun the exact same audit. same prompts. same engines. same measurement. track whether your citation share increased, you started appearing in previously lost topics, competitor recommendations dropped and your brand moved from “mentioned” to actually recommended. then keep working on the clusters that are moving instead of starting another random content sprint. if you want us to run this audit for your company and & make you rank #1 in AI search, reply “audit”

We’ve spent the past year developing new primitives for GPUs. We believe this enables new algorithms to change how machines learn, open new markets, and dramatically reshape AI’s cost structure. We have raised a $1 billion Series A, co-led by Index and Lux.

Literal file tree 🌳